How to use _info_filter method in autotest

Best Python code snippet using autotest_python

catimages.py

Source:catimages.py Github

copy

Full Screen

1#!/usr/bin/python2# -*- coding: utf-8 -*-3"""4Image by content categorization derived from 'checkimages.py'.5Script to check uncategorized files. This script checks if a file6has some content that allows to assign it to a category.7This script runs on commons only. It needs also external libraries8(see imports and comments there) and additional configuration/data9files in order to run properly. Most of them can be checked-out at:10 http://svn.toolserver.org/svnroot/drtrigon/11(some code might get compiled on-the-fly, so a GNU compiler along12with library header files is needed too)13This script understands the following command-line arguments:14-cat[:#] Use a category as recursive generator15 (if no given 'Category:Media_needing_categories' is used)16-start[:#] Start after File:[:#] or if no file given start from top17 (instead of resuming last run).18-limit The number of images to check (default: 80)19-noguesses If given, this option will disable all guesses (which are20 less reliable than true searches).21-single:# Run for one (any) single page only.22-train Train classifiers on good (homegenous) categories.23X-sendemail Send an email after tagging.24X-untagged[:#] Use daniel's tool as generator:25X http://toolserver.org/~daniel/WikiSense/UntaggedImages.php26"""27#28# (C) Kyle/Orgullomoore, 2006-2007 (newimage.py)29# (C) Pywikipedia team, 2007-2011 (checkimages.py)30# (C) DrTrigon, 201231#32# Distributed under the terms of the MIT license.33#34__version__ = '$Id$'35#36# python default packages37import re, urllib2, os, locale, sys, datetime, math, shutil, mimetypes, shelve38import StringIO, json # fallback: simplejson39from subprocess import Popen, PIPE40import Image41#import ImageFilter42scriptdir = os.path.dirname(sys.argv[0])43if not os.path.isabs(scriptdir):44 scriptdir = os.path.abspath(os.path.join(os.curdir, scriptdir))45# additional python packages (non-default but common)46try:47 import numpy as np48 from scipy import ndimage, fftpack#, signal49 import cv50 # TS: nonofficial cv2.so backport of the testing-version of51 # python-opencv because of missing build-host, done by DaB52 sys.path.append('/usr/local/lib/python2.6/')53 import cv254 sys.path.remove('/usr/local/lib/python2.6/')55 import pyexiv256 import warnings57 with warnings.catch_warnings():58 warnings.simplefilter("ignore")59 import gtk # ignore warning: "GtkWarning: could not open display"60 import rsvg # gnome-python2-rsvg (binding to librsvg)61 import cairo62 import magic # python-magic (binding to libmagic)63except:64 # either raise the ImportError later or skip it65 pass66# pywikipedia framework python packages67import wikipedia as pywikibot68import pagegenerators, catlib69import checkimages70import externals # allow import from externals71# additional python packages (more exotic and problematic ones)72# modules needing compilation are imported later on request:73# (see https://jira.toolserver.org/browse/TS-1452)74# e.g. opencv, jseg, slic, pydmtx, zbar, (pyml or equivalent)75# binaries: exiftool, pdftotext/pdfimages (poppler), ffprobe (ffmpeg),76# convert/identify (ImageMagick), (ocropus)77# TODO:78# (pdfminer not used anymore/at the moment...)79# python-djvulibre or python-djvu for djvu support80externals.check_setup('colormath') # check for and install needed81externals.check_setup('jseg') # 'externals' modules82externals.check_setup('jseg/jpeg-6b') #83#externals.check_setup('_mlpy') #84externals.check_setup('_music21') #85externals.check_setup('opencv/haarcascades') #86externals.check_setup('pydmtx') # <<< !!! test OS package management here !!!87externals.check_setup('py_w3c') #88externals.check_setup('_zbar') #89import pycolorname90#import _mlpy as mlpy91from colormath.color_objects import RGBColor92from py_w3c.validators.html.validator import HTMLValidator, ValidationFault93#from pdfminer import pdfparser, pdfinterp, pdfdevice, converter, cmapdb, layout94#externals.check_setup('_ocropus')95locale.setlocale(locale.LC_ALL, '')96###############################################################################97# <--------------------------- Change only below! --------------------------->#98###############################################################################99# NOTE: in the messages used by the Bot if you put __botnick__ in the text, it100# will automatically replaced with the bot's nickname.101# Add your project (in alphabetical order) if you want that the bot start102project_inserted = [u'commons',]103# Ok, that's all. What is below, is the rest of code, now the code is fixed and it will run correctly in your project.104################################################################################105# <--------------------------- Change only above! ---------------------------> #106################################################################################107tmpl_FileContentsByBot = u"""}}108{{FileContentsByBot109| botName = ~~~110|"""111# this list is auto-generated during bot run (may be add notifcation about NEW templates)112#tmpl_available_spec = [ u'Properties', u'ColorRegions', u'Faces', u'ColorAverage' ]113tmpl_available_spec = [] # auto-generated114# global115useGuesses = True # Use guesses which are less reliable than true searches116# all detection and recognition methods - bindings to other classes, modules and libs117class _UnknownFile(object):118 def __init__(self, file_name, file_mime, *args, **kwargs):119 self.file_name = file_name120 self.file_mime = file_mime121 self.image_size = (None, None)122 # available file properties and metadata123 self._properties = { 'Properties': [{'Format': u'-', 'Pages': 0}],124 'Metadata': [], }125 # available feature to extract126 self._features = { 'ColorAverage': [],127 'ColorRegions': [],128 'Faces': [],129 'People': [],130 'OpticalCodes': [],131 'Chessboard': [],132 'History': [],133 'Text': [],134 'Streams': [],135 'Audio': [],136 'Legs': [],137 'Hands': [],138 'Torsos': [],139 'Ears': [],140 'Eyes': [],141 'Automobiles': [],142 'Classify': [], }143 def __enter__(self):144 return self145 def __exit__(self, type, value, traceback):146 pass147 def getProperties(self):148 self._detect_HeaderAndMetadata() # Metadata149 self._detect_Properties() # Properties150 return self._properties151 def getFeatures(self):152 pywikibot.warning(u"File format '%s/%s' not supported (yet)!" % tuple(self.file_mime[:2]))153 return self._features154 def _detect_HeaderAndMetadata(self):155 # check/look into the file by midnight commander (mc)156 # use exif as first hint - in fact gives also image-size, streams, ...157 exif = self._util_get_DataTags_EXIF()158 #print exif159 result = { 'Software': exif['Software'] if 'Software' in exif else u'-',160 'Output_Extension': exif['Output_extension'] if 'Output_extension' in exif else u'-',161 'Desc': exif['Desc'] if 'Desc' in exif else u'-',162 'DescProducer': exif['DescProducer'] if 'DescProducer' in exif else u'-',163 'DescCreator': exif['DescCreator'] if 'DescCreator' in exif else u'-',164 'Comment': exif['Comment'] if 'Comment' in exif else u'-',165 'Producer': exif['Producer'] if 'Producer' in exif else u'-',}166 #'Comments': exif['Comments'] if 'Comments' in exif else u'-',167 #'WorkDesc': exif['WorkDescription'] if 'WorkDescription' in exif else u'-',168 ##'Dimensions': tuple(map(int, exif['ImageSize'].split(u'x'))),}169 #'Dimensions': tuple(exif['ImageSize'].split(u'x')) if 'ImageSize' in exif else (None, None),}170 #'Mode': exif['ColorType'], }171# TODO: vvv172#* metadata template in commons has to be worked out and code adopted173#* like in 'Streams' a nice content listing of MIDI (exif or music21 - if needed at all?)174#* docu all this stuff in commons175#* docu and do all open things on "commons TODO list"176#177#178#179#(* initial audio midi support (music21))180#[TODO: docu on Commons ... / template ...]181# TODO: if '_detect_History' is not needed here, moveit back into _JpegFile !!!182 #print "self._detect_History()"183 #print self._detect_History()184 # https://pypi.python.org/pypi/hachoir-metadata (needs 'core' and 'parser')185 #186 #from hachoir_core.error import HachoirError187 #from hachoir_core.stream import InputStreamError188 #from hachoir_parser import createParser189 #import hachoir_core.config as hachoir_config190 #191 #from hachoir_metadata import extractMetadata192 #193 #hachoir_config.debug = True194 #hachoir_config.verbose = True195 #hachoir_config.quiet = True196 #197 ## Create parser198 #try:199 # parser = createParser(self.file_name.decode('utf-8'),200 # real_filename=self.file_name.encode('utf-8'),201 # tags=None)202 # #print [val for val in enumerate(parser.createFields())]203 # desc = parser.description204 # ptags = parser.getParserTags()205 #except (InputStreamError, AttributeError):206 # desc = u'-'207 # ptags = {}208 #209 ## Extract metadata210 #try:211 # # quality: 0.0 fastest, 1.0 best, and default is 0.5212 # metadata = extractMetadata(parser, 0.5)213 # #mtags = dict([(key, metadata.getValues(key))214 # mtags = dict([(key, metadata.getValues(key)) # get, getItem, getItems, getText215 # for key in metadata._Metadata__data.keys()#])216 # if metadata.getValues(key)])217 #except (HachoirError, AttributeError):218 # mtags = {}219 #220 ##result = {'parser_desc': desc, 'parserdata': ptags, 'metadata': mtags}221 ##print result222 #print {'parser_desc': desc, 'parserdata': ptags, 'metadata': mtags}223 #224 ### Display metadatas on stdout225 ##text = metadata.exportPlaintext(priority=None, human=False)226 ##if not text:227 ## text = [u"(no metadata, priority may be too small, try priority=999)"]228 ##print u'\n'.join(text)229 self._properties['Metadata'] = [result]230 #print self._properties['Metadata']231 return232 def _detect_Properties(self):233 # get mime-type file-size, ...234 pass235 def _util_get_DataTags_EXIF(self):236 # http://tilloy.net/dev/pyexiv2/tutorial.html237 # (is UNFORTUNATELY NOT ABLE to handle all tags, e.g. 'FacesDetected', ...)238 239 if hasattr(self, '_buffer_EXIF'):240 return self._buffer_EXIF241 res = {}242 enable_recovery() # enable recovery from hard crash243 try:244 if hasattr(pyexiv2, 'ImageMetadata'):245 metadata = pyexiv2.ImageMetadata(self.file_name)246 metadata.read()247 248 for key in metadata.exif_keys:249 res[key] = metadata[key]250 251 for key in metadata.iptc_keys:252 res[key] = metadata[key]253 254 for key in metadata.xmp_keys:255 res[key] = metadata[key]256 else:257 image = pyexiv2.Image(self.file_name)258 image.readMetadata()259 260 for key in image.exifKeys():261 res[key] = image[key]262 263 for key in image.iptcKeys():264 res[key] = image[key]265 266 #for key in image.xmpKeys():267 # res[key] = image[key]268 except IOError:269 pass270 except RuntimeError:271 pass272 disable_recovery() # disable since everything worked out fine273 274 275 # http://www.sno.phy.queensu.ca/~phil/exiftool/276 # MIGHT BE BETTER TO USE AS PYTHON MODULE; either by wrapper or perlmodule:277 # http://search.cpan.org/~gaas/pyperl-1.0/perlmodule.pod278 # (or use C++ with embbedded perl to write a python module)279 data = Popen("exiftool -j %s" % self.file_name, 280 shell=True, stdout=PIPE).stdout.read()281 if not data:282 raise ImportError("exiftool not found!")283 try: # work-a-round for badly encoded exif data (from pywikibot/comms/http.py)284 data = unicode(data, 'utf-8', errors = 'strict')285 except UnicodeDecodeError:286 data = unicode(data, 'utf-8', errors = 'replace')287 #res = {}288 data = re.sub("(?<!\")\(Binary data (?P<size>\d*) bytes\)", "\"(Binary data \g<size> bytes)\"", data) # work-a-round some issue289 for item in json.loads(data):290 res.update( item )291 #print res292 self._buffer_EXIF = res293 294 return self._buffer_EXIF295 def _detect_History(self):296 res = self._util_get_DataTags_EXIF()297 #a = []298 #for k in res.keys():299 # if 'history' in k.lower():300 # a.append( k )301 #for item in sorted(a):302 # print item303 # http://tilloy.net/dev/pyexiv2/api.html#pyexiv2.xmp.XmpTag304 #print [getattr(res['Xmp.xmpMM.History'], item) for item in ['key', 'type', 'name', 'title', 'description', 'raw_value', 'value', ]]305 result = []306 i = 1307 while (('Xmp.xmpMM.History[%i]' % i) in res):308 data = { 'ID': i,309 'Software': u'-',310 'Timestamp': u'-',311 'Action': u'-',312 'Info': u'-', }313 if ('Xmp.xmpMM.History[%i]/stEvt:softwareAgent'%i) in res:314 data['Software'] = res['Xmp.xmpMM.History[%i]/stEvt:softwareAgent'%i].value315 data['Timestamp'] = res['Xmp.xmpMM.History[%i]/stEvt:when'%i].value316 data['Action'] = res['Xmp.xmpMM.History[%i]/stEvt:action'%i].value317 if ('Xmp.xmpMM.History[%i]/stEvt:changed'%i) in res:318 data['Info'] = res['Xmp.xmpMM.History[%i]/stEvt:changed'%i].value319 #print res['Xmp.xmpMM.History[%i]/stEvt:instanceID'%i].value320 result.append( data )321 elif ('Xmp.xmpMM.History[%i]/stEvt:parameters'%i) in res:322 data['Action'] = res['Xmp.xmpMM.History[%i]/stEvt:action'%i].value323 data['Info'] = res['Xmp.xmpMM.History[%i]/stEvt:parameters'%i].value324 #data['Action'] = data['Info'].split(' ')[0]325 result.append( data )326 else:327 pass328 i += 1329 330 self._features['History'] = result331 return332class _JpegFile(_UnknownFile):333 # for '_detect_Trained'334 cascade_files = [(u'Legs', 'haarcascade_lowerbody.xml'),335 (u'Torsos', 'haarcascade_upperbody.xml'),336 (u'Ears', 'haarcascade_mcs_leftear.xml'),337 (u'Ears', 'haarcascade_mcs_rightear.xml'),338 (u'Eyes', 'haarcascade_lefteye_2splits.xml'), # (http://yushiqi.cn/research/eyedetection)339 (u'Eyes', 'haarcascade_righteye_2splits.xml'), # (http://yushiqi.cn/research/eyedetection)340 #externals/opencv/haarcascades/haarcascade_mcs_lefteye.xml341 #externals/opencv/haarcascades/haarcascade_mcs_righteye.xml342 # (others include indifferent (left and/or right) and pair)343 (u'Automobiles', 'cars3.xml'), # http://www.youtube.com/watch?v=c4LobbqeKZc344 (u'Hands', '1256617233-2-haarcascade-hand.xml', 300.),] # http://www.andol.info/345 # ('Hands' does not behave very well, in fact it detects any kind of skin and other things...)346 #(u'Aeroplanes', 'haarcascade_aeroplane.xml'),] # e.g. for 'Category:Unidentified aircraft'347 def __init__(self, file_name, file_mime, *args, **kwargs):348 _UnknownFile.__init__(self, file_name, file_mime)349 self.image_filename = os.path.split(self.file_name)[-1]350 self.image_path = self.file_name351 self.image_path_JPEG = self.image_path + '.jpg'352 self._convert()353 def __exit__(self, type, value, traceback):354 #if os.path.exists(self.image_path):355 # os.remove( self.image_path )356 if os.path.exists(self.image_path_JPEG):357 os.remove( self.image_path_JPEG )358 #image_path_new = self.image_path_JPEG.replace(u"cache/", u"cache/0_DETECTED_")359 #if os.path.exists(image_path_new):360 # os.remove( image_path_new )361 def getFeatures(self):362 # Faces (extract EXIF data)363 self._detect_Faces_EXIF()364 # Faces and eyes (opencv pre-trained haar)365 self._detect_Faces()366# TODO: test and use or switch off367 # Face via Landmark(s)368# self._detect_FaceLandmark_xBOB()369 # exclude duplicates (CV and EXIF)370 faces = [item['Position'] for item in self._features['Faces']]371 for i in self._util_merge_Regions(faces)[1]:372 del self._features['Faces'][i]373 # Segments and colors374 self._detect_SegmentColors()375 # Average color376 self._detect_AverageColor()377 # People/Pedestrian (opencv pre-trained hog and haarcascade)378 self._detect_People()379 # Geometric object (opencv hough line, circle, edges, corner, ...)380 self._detect_Geometry()381 # general (opencv pre-trained, third-party and self-trained haar382 # and cascade) classification383 # http://www.computer-vision-software.com/blog/2009/11/faq-opencv-haartraining/384 for cf in self.cascade_files:385 self._detect_Trained(*cf)386 # barcode and Data Matrix recognition (libdmtx/pydmtx, zbar, gocr?)387 self._recognize_OpticalCodes()388 # Chessboard (opencv reference detector)389 self._detect_Chessboard()390 # general (self-trained) detection WITH classification391 # BoW: uses feature detection (SIFT, SURF, ...) AND classification (SVM, ...)392# self._detectclassify_ObjectAll()393 # Wavelet: uses wavelet transformation AND classification (machine learning)394# self._detectclassify_ObjectAll_PYWT()395 # general file EXIF history information396 self._detect_History()397 398 return self._features399 # supports a lot of different file types thanks to PIL400 def _convert(self):401 try:402 im = Image.open(self.image_path) # might be png, gif etc, for instance403 #im.thumbnail(size, Image.ANTIALIAS) # size is 640x480404 im.convert('RGB').save(self.image_path_JPEG, "JPEG")405 self.image_size = im.size406 except IOError, e:407 if 'image file is truncated' in str(e):408 # im object has changed due to exception raised409 im.convert('RGB').save(self.image_path_JPEG, "JPEG")410 self.image_size = im.size411 else:412 try:413 # since opencv might still work, try this as fall-back414 img = cv2.imread( self.image_path, cv.CV_LOAD_IMAGE_COLOR )415 cv2.imwrite(self.image_path_JPEG, img)416 self.image_size = (img.shape[1], img.shape[0])417 except:418 if os.path.exists(self.image_path_JPEG):419 os.remove(self.image_path_JPEG)420 self.image_path_JPEG = self.image_path421 except:422 self.image_path_JPEG = self.image_path423 # FULL TIFF support (e.g. group4)424 # http://code.google.com/p/pylibtiff/425 # MIME: 'image/jpeg; charset=binary', ...426 def _detect_Properties(self):427 """Retrieve as much file property info possible, especially the same428 as commons does in order to compare if those libraries (ImageMagick,429 ...) are buggy (thus explicitely use other software for independence)"""430 result = {'Format': u'-', 'Pages': 0}431 try:432 i = Image.open(self.image_path)433 except IOError:434 pywikibot.warning(u'unknown file type [_JpegFile]')435 return436 # http://mail.python.org/pipermail/image-sig/1999-May/000740.html437 pc=0 # count number of pages438 while True:439 try:440 i.seek(pc)441 except EOFError:442 break443 pc+=1444 i.seek(0) # restore default445 # http://grokbase.com/t/python/image-sig/082psaxt6k/embedded-icc-profiles446 # python-lcms (littlecms) may be freeimage library447 #icc = i.app['APP2'] # jpeg448 #icc = i.tag[34675] # tiff449 #icc = re.sub('[^%s]'%string.printable, ' ', icc)450 ## more image formats and more post-processing needed...451 #self.image_size = i.size452 result.update({ #'bands': i.getbands(),453 #'bbox': i.getbbox(),454 'Format': i.format,455 'Mode': i.mode,456 #'info': i.info,457 #'stat': os.stat(self.image_path),458 'Palette': str(len(i.palette.palette)) if i.palette else u'-',459 'Pages': pc,460 'Dimensions': self.image_size,461 'Filesize': os.path.getsize(self.file_name),462 'MIME': u'%s/%s' % tuple(self.file_mime[:2]), })463 #self._properties['Properties'] = [result]464 self._properties['Properties'][0].update(result)465 return466 # .../opencv/samples/c/facedetect.cpp467 # http://opencv.willowgarage.com/documentation/python/genindex.html468 def _detect_Faces(self):469 """Converts an image to grayscale and prints the locations of any470 faces found"""471 # http://python.pastebin.com/m76db1d6b472 # http://creatingwithcode.com/howto/face-detection-in-static-images-with-python/473 # http://opencv.willowgarage.com/documentation/python/objdetect_cascade_classification.html474 # http://opencv.willowgarage.com/wiki/FaceDetection475 # http://blog.jozilla.net/2008/06/27/fun-with-python-opencv-and-face-detection/476 # http://www.cognotics.com/opencv/servo_2007_series/part_4/index.html477 # https://code.ros.org/trac/opencv/browser/trunk/opencv_extra/testdata/gpu/haarcascade?rev=HEAD478 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_eye_tree_eyeglasses.xml')479 #xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_eye.xml')480 if not os.path.exists(xml):481 raise IOError(u"No such file: '%s'" % xml)482 #nestedCascade = cv.Load(483 nestedCascade = cv2.CascadeClassifier(xml)484 # http://tutorial-haartraining.googlecode.com/svn/trunk/data/haarcascades/485 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_frontalface_alt.xml')486 # MAY BE USE 'haarcascade_frontalface_alt_tree.xml' ALSO / INSTEAD...?!!487 if not os.path.exists(xml):488 raise IOError(u"No such file: '%s'" % xml)489 #cascade = cv.Load(490 cascade = cv2.CascadeClassifier(xml)491 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_profileface.xml')492 if not os.path.exists(xml):493 raise IOError(u"No such file: '%s'" % xml)494 cascadeprofil = cv2.CascadeClassifier(xml)495 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_mcs_mouth.xml')496 if not os.path.exists(xml):497 raise IOError(u"No such file: '%s'" % xml)498 cascademouth = cv2.CascadeClassifier(xml)499 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_mcs_nose.xml')500 if not os.path.exists(xml):501 raise IOError(u"No such file: '%s'" % xml)502 cascadenose = cv2.CascadeClassifier(xml)503 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_lefteye_2splits.xml')504 if not os.path.exists(xml):505 raise IOError(u"No such file: '%s'" % xml)506 cascadelefteye = cv2.CascadeClassifier(xml) # (http://yushiqi.cn/research/eyedetection)507 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_righteye_2splits.xml')508 if not os.path.exists(xml):509 raise IOError(u"No such file: '%s'" % xml)510 cascaderighteye = cv2.CascadeClassifier(xml) # (http://yushiqi.cn/research/eyedetection)511 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_mcs_leftear.xml')512 if not os.path.exists(xml):513 raise IOError(u"No such file: '%s'" % xml)514 cascadeleftear = cv2.CascadeClassifier(xml)515 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_mcs_rightear.xml')516 if not os.path.exists(xml):517 raise IOError(u"No such file: '%s'" % xml)518 cascaderightear = cv2.CascadeClassifier(xml)519 scale = 1.520 # So, to find an object of an unknown size in the image the scan521 # procedure should be done several times at different scales.522 # http://opencv.itseez.com/modules/objdetect/doc/cascade_classification.html523 try:524 #image = cv.LoadImage(self.image_path)525 #img = cv2.imread( self.image_path, cv.CV_LOAD_IMAGE_COLOR )526 img = cv2.imread( self.image_path_JPEG, cv.CV_LOAD_IMAGE_COLOR )527 #image = cv.fromarray(img)528 if img == None:529 raise IOError530 531 # !!! the 'scale' here IS RELEVANT FOR THE DETECTION RATE;532 # how small and how many features are detected as faces (or eyes)533 scale = max([1., np.average(np.array(img.shape)[0:2]/500.)])534 except IOError:535 pywikibot.warning(u'unknown file type [_detect_Faces]')536 return537 except AttributeError:538 pywikibot.warning(u'unknown file type [_detect_Faces]')539 return540 #detectAndDraw( image, cascade, nestedCascade, scale );541 # http://nullege.com/codes/search/cv.CvtColor542 #smallImg = cv.CreateImage( (cv.Round(img.shape[0]/scale), cv.Round(img.shape[1]/scale)), cv.CV_8UC1 )543 #smallImg = cv.fromarray(np.empty( (cv.Round(img.shape[0]/scale), cv.Round(img.shape[1]/scale)), dtype=np.uint8 ))544 smallImg = np.empty( (cv.Round(img.shape[1]/scale), cv.Round(img.shape[0]/scale)), dtype=np.uint8 )545 #cv.CvtColor( image, gray, cv.CV_BGR2GRAY )546 gray = cv2.cvtColor( img, cv.CV_BGR2GRAY )547 #cv.Resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR ) 548 smallImg = cv2.resize( gray, smallImg.shape, interpolation=cv2.INTER_LINEAR )549 #cv.EqualizeHist( smallImg, smallImg )550 smallImg = cv2.equalizeHist( smallImg )551 t = cv.GetTickCount()552 faces = list(cascade.detectMultiScale( smallImg,553 1.1, 2, 0554 #|cv.CV_HAAR_FIND_BIGGEST_OBJECT555 #|cv.CV_HAAR_DO_ROUGH_SEARCH556 |cv.CV_HAAR_SCALE_IMAGE,557 (30, 30) ))558 #faces = cv.HaarDetectObjects(grayscale, cascade, storage, 1.2, 2,559 # cv.CV_HAAR_DO_CANNY_PRUNING, (50,50))560 facesprofil = list(cascadeprofil.detectMultiScale( smallImg,561 1.1, 2, 0562 #|cv.CV_HAAR_FIND_BIGGEST_OBJECT563 #|cv.CV_HAAR_DO_ROUGH_SEARCH564 |cv.CV_HAAR_SCALE_IMAGE,565 (30, 30) ))566 #faces = self._util_merge_Regions(faces + facesprofil)[0]567 faces = self._util_merge_Regions(faces + facesprofil, overlap=True)[0]568 faces = np.array(faces)569 #if faces:570 # self._drawRect(faces) #call to a python pil571 t = cv.GetTickCount() - t572 #print( "detection time = %g ms\n" % (t/(cv.GetTickFrequency()*1000.)) )573 #colors = [ (0,0,255),574 # (0,128,255),575 # (0,255,255),576 # (0,255,0),577 # (255,128,0),578 # (255,255,0),579 # (255,0,0),580 # (255,0,255) ]581 result = []582 for i, r in enumerate(faces):583 #color = colors[i%8]584 (rx, ry, rwidth, rheight) = r585 #cx = cv.Round((rx + rwidth*0.5)*scale)586 #cy = cv.Round((ry + rheight*0.5)*scale)587 #radius = cv.Round((rwidth + rheight)*0.25*scale)588 #cv2.circle( img, (cx, cy), radius, color, 3, 8, 0 )589 #if nestedCascade.empty():590 # continue591 # Wilson, Fernandez: FACIAL FEATURE DETECTION USING HAAR CLASSIFIERS592 # http://nichol.as/papers/Wilson/Facial%20feature%20detection%20using%20Haar.pdf593 #dx, dy = cv.Round(rwidth*0.5), cv.Round(rheight*0.5)594 dx, dy = cv.Round(rwidth/8.), cv.Round(rheight/8.)595 (rx, ry, rwidth, rheight) = (max([rx-dx,0]), max([ry-dy,0]), min([rwidth+2*dx,img.shape[1]]), min([rheight+2*dy,img.shape[0]]))596 #smallImgROI = smallImg597 #print r, (rx, ry, rwidth, rheight)598 #smallImgROI = smallImg[ry:(ry+rheight),rx:(rx+rwidth)]599 smallImgROI = smallImg[ry:(ry+6*dy),rx:(rx+rwidth)] # speed up by setting instead of extracting ROI600 nestedObjects = nestedCascade.detectMultiScale( smallImgROI,601 1.1, 2, 0602 #|CV_HAAR_FIND_BIGGEST_OBJECT603 #|CV_HAAR_DO_ROUGH_SEARCH604 #|CV_HAAR_DO_CANNY_PRUNING605 |cv.CV_HAAR_SCALE_IMAGE,606 (30, 30) )607 nestedObjects = self._util_merge_Regions(list(nestedObjects), overlap=True)[0]608 if len(nestedObjects) < 2:609 nestedLeftEye = cascadelefteye.detectMultiScale( smallImgROI,610 1.1, 2, 0611 #|CV_HAAR_FIND_BIGGEST_OBJECT612 #|CV_HAAR_DO_ROUGH_SEARCH613 #|CV_HAAR_DO_CANNY_PRUNING614 |cv.CV_HAAR_SCALE_IMAGE,615 (30, 30) )616 nestedRightEye = cascaderighteye.detectMultiScale( smallImgROI,617 1.1, 2, 0618 #|CV_HAAR_FIND_BIGGEST_OBJECT619 #|CV_HAAR_DO_ROUGH_SEARCH620 #|CV_HAAR_DO_CANNY_PRUNING621 |cv.CV_HAAR_SCALE_IMAGE,622 (30, 30) )623 nestedObjects = self._util_merge_Regions(list(nestedObjects) +624 list(nestedLeftEye) + 625 list(nestedRightEye), overlap=True)[0]626 #if len(nestedObjects) > 2:627 # nestedObjects = self._util_merge_Regions(list(nestedObjects), close=True)[0]628 smallImgROI = smallImg[(ry+4*dy):(ry+rheight),rx:(rx+rwidth)]629 nestedMouth = cascademouth.detectMultiScale( smallImgROI,630 1.1, 2, 0631 |cv.CV_HAAR_FIND_BIGGEST_OBJECT632 |cv.CV_HAAR_DO_ROUGH_SEARCH633 #|CV_HAAR_DO_CANNY_PRUNING634 |cv.CV_HAAR_SCALE_IMAGE,635 (30, 30) )636 smallImgROI = smallImg[(ry+(5*dy)/2):(ry+5*dy+(5*dy)/2),(rx+(5*dx)/2):(rx+5*dx+(5*dx)/2)]637 nestedNose = cascadenose.detectMultiScale( smallImgROI,638 1.1, 2, 0639 |cv.CV_HAAR_FIND_BIGGEST_OBJECT640 |cv.CV_HAAR_DO_ROUGH_SEARCH641 #|CV_HAAR_DO_CANNY_PRUNING642 |cv.CV_HAAR_SCALE_IMAGE,643 (30, 30) )644 smallImgROI = smallImg[(ry+2*dy):(ry+6*dy),rx:(rx+rwidth)]645 nestedEars = list(cascadeleftear.detectMultiScale( smallImgROI,646 1.1, 2, 0647 |cv.CV_HAAR_FIND_BIGGEST_OBJECT648 |cv.CV_HAAR_DO_ROUGH_SEARCH649 #|CV_HAAR_DO_CANNY_PRUNING650 |cv.CV_HAAR_SCALE_IMAGE,651 (30, 30) ))652 nestedEars += list(cascaderightear.detectMultiScale( smallImgROI,653 1.1, 2, 0654 |cv.CV_HAAR_FIND_BIGGEST_OBJECT655 |cv.CV_HAAR_DO_ROUGH_SEARCH656 #|CV_HAAR_DO_CANNY_PRUNING657 |cv.CV_HAAR_SCALE_IMAGE,658 (30, 30) ))659 data = { 'ID': (i+1),660 'Position': tuple(np.int_(r*scale)), 661 'Type': u'-',662 'Eyes': [],663 'Mouth': (),664 'Nose': (),665 'Ears': [],666 'Pose': (), }667 data['Coverage'] = float(data['Position'][2]*data['Position'][3])/(self.image_size[0]*self.image_size[1])668 #if (c >= confidence):669 # eyes = nestedObjects670 # if not (type(eyes) == type(tuple())):671 # eyes = tuple((eyes*scale).tolist())672 # result.append( {'Position': r*scale, 'eyes': eyes, 'confidence': c} )673 #print {'Position': r, 'eyes': nestedObjects, 'confidence': c}674 for nr in nestedObjects:675 (nrx, nry, nrwidth, nrheight) = nr676 cx = cv.Round((rx + nrx + nrwidth*0.5)*scale)677 cy = cv.Round((ry + nry + nrheight*0.5)*scale)678 radius = cv.Round((nrwidth + nrheight)*0.25*scale)679 #cv2.circle( img, (cx, cy), radius, color, 3, 8, 0 )680 data['Eyes'].append( (cx-radius, cy-radius, 2*radius, 2*radius) )681 if len(nestedMouth):682 (nrx, nry, nrwidth, nrheight) = nestedMouth[0]683 cx = cv.Round((rx + nrx + nrwidth*0.5)*scale)684 cy = cv.Round(((ry+4*dy) + nry + nrheight*0.5)*scale)685 radius = cv.Round((nrwidth + nrheight)*0.25*scale)686 #cv2.circle( img, (cx, cy), radius, color, 3, 8, 0 )687 data['Mouth'] = (cx-radius, cy-radius, 2*radius, 2*radius)688 if len(nestedNose):689 (nrx, nry, nrwidth, nrheight) = nestedNose[0]690 cx = cv.Round(((rx+(5*dx)/2) + nrx + nrwidth*0.5)*scale)691 cy = cv.Round(((ry+(5*dy)/2) + nry + nrheight*0.5)*scale)692 radius = cv.Round((nrwidth + nrheight)*0.25*scale)693 #cv2.circle( img, (cx, cy), radius, color, 3, 8, 0 )694 data['Nose'] = (cx-radius, cy-radius, 2*radius, 2*radius)695 for nr in nestedEars:696 (nrx, nry, nrwidth, nrheight) = nr697 cx = cv.Round((rx + nrx + nrwidth*0.5)*scale)698 cy = cv.Round((ry + nry + nrheight*0.5)*scale)699 radius = cv.Round((nrwidth + nrheight)*0.25*scale)700 #cv2.circle( img, (cx, cy), radius, color, 3, 8, 0 )701 data['Ears'].append( (cx-radius, cy-radius, 2*radius, 2*radius) )702 if data['Mouth'] and data['Nose'] and data['Eyes'] and (len(data['Eyes']) == 2):703 # head model "little girl" for use in "MeshLab":704 # http://www.turbosquid.com/FullPreview/Index.cfm/ID/302581705 # http://meshlab.sourceforge.net/706 D3points = [[ 70.0602, 109.898, 20.8234], # left eye707 [ 2.37427, 110.322, 21.7776], # right eye708 [ 36.8301, 78.3185, 52.0345], # nose709 [ 36.6391, 51.1675, 38.5903],] # mouth710 #[ 119.268, 91.3111, -69.6397], # left ear711 #[-49.1328, 91.3111, -67.2481],] # right ear712 D2points = [np.array(data['Eyes'][0]), np.array(data['Eyes'][1]),713 np.array(data['Nose']), np.array(data['Mouth']),]714 D2points = [ item[:2] + item[2:]/2. for item in D2points ]715 neutral = np.array([[np.pi],[0.],[0.]])716 # calculate pose717 rvec, tvec, cm, err = self._util_get_Pose_solvePnP(D3points, D2points, self.image_size)718 #data['Pose'] = tuple(rvec[:,0])719 check = not (err[:,0,:].max() > 0.5)720 if not check:721 rvec = neutral # reset to neutral pose722 tvec = np.array([[0.],[0.],[100.]]) # reset to neutral position (same order as max of D3points)723 pywikibot.warning(u'Could not calculate pose of face, too big errors. '724 u'(looks like neutral pose/position is somehow singular)')725 ## debug: draw pose726 ##rvec *= 0727 #mat, perp = self._util_getD2coords_calc(np.eye(3), cm, rvec, tvec, hacky=False)728 ## from '_util_drawAxes(...)'729 #for i, item in enumerate(mat.transpose()):730 # p = tuple((50+10*item).astype(int))[:2]731 # cv2.line(img, (50, 50), p, (0., 0., 255.), 1)732 # cv2.putText(img, str(i), p, cv2.FONT_HERSHEY_PLAIN, 1., (0., 0., 255.))733 #cv2.imshow("win", img)734 #cv2.waitKey()735 # calculate delta to neutral pose736 drv = -cv2.composeRT(-rvec, np.zeros((3,1)),737 neutral, np.zeros((3,1)))[0]738 rvec = cv2.Rodrigues(cv2.Rodrigues(rvec)[0])[0] # NOT unique!!!739 #nrv = cv2.composeRT(neutral, np.zeros((3,1)),740 # drv, np.zeros((3,1)))[0]741 #print (rvec - nrv < 1E-12) # compare742 data['Pose'] = map(float, tuple(drv[:,0]))743# TODO: POSIT has to be tested and compared; draw both results!744 # POSIT: http://www.cfar.umd.edu/~daniel/daniel_papersfordownload/Pose25Lines.pdf745 if False:746 pywikibot.output("solvePnP:")747 pywikibot.output(str(rvec[:,0]))748 pywikibot.output(str(tvec[:,0]))749 pywikibot.output(str(err[:,0,:]))750 rvec, tvec, cm, err = self._util_get_Pose_POSIT(D3points, D2points)751 pywikibot.output("POSIT:")752 pywikibot.output(str(rvec[:,0]))753 pywikibot.output(str(tvec))754 pywikibot.output(str(np.array(err)[:,0,:]/max(self.image_size)))755 result.append( data )756 ## see '_drawRect'757 #if result:758 # #image_path_new = os.path.join(scriptdir, 'cache/0_DETECTED_' + self.image_filename)759 # image_path_new = self.image_path_JPEG.replace(u"cache/", u"cache/0_DETECTED_")760 # cv2.imwrite( image_path_new, img )761 #return faces.tolist()762 self._features['Faces'] += result763 return764 def _util_get_Pose_solvePnP(self, D3points, D2points, shape):765 """ Calculate pose from head model "little girl" w/o camera or other766 calibrations needed.767 D2points: left eye, right eye, nose, mouth768 """769 # howto (credits to "Roy"):770 # http://www.youtube.com/watch?v=ZDNH4BT5Do4771 # http://www.morethantechnical.com/2010/03/19/quick-and-easy-head-pose-estimation-with-opencv-w-code/772 # http://www.morethantechnical.com/2012/10/17/head-pose-estimation-with-opencv-opengl-revisited-w-code/773 # e.g. with head model "little girl" for use in "MeshLab":774 # http://www.turbosquid.com/FullPreview/Index.cfm/ID/302581775 # http://meshlab.sourceforge.net/776 # set-up camera matrix (no calibration needed!)777 max_d = max(shape)778 cameraMatrix = [[max_d, 0, shape[0]/2.0],779 [ 0, max_d, shape[1]/2.0],780 [ 0, 0, 1.0],]781 # calculate pose782 rvec, tvec = cv2.solvePnP(np.array(D3points).astype('float32'), np.array(D2points).astype('float32'), np.array(cameraMatrix).astype('float32'), None)783 # compare to 2D points784 err = []785 for i, vec in enumerate(np.array(D3points)):786 nvec = np.dot(cameraMatrix, (np.dot(cv2.Rodrigues(rvec)[0], vec) + tvec[:,0]))787 err.append(((D2points[i] - nvec[:2]/nvec[2]), D2points[i], nvec[:2]/nvec[2]))788 pywikibot.output(u'result for UN-calibrated camera:\n rot=%s' % rvec.transpose()[0])789 return rvec, tvec, np.array(cameraMatrix), (np.array(err)/max_d)790 #def _util_get_Pose_POSIT(self, D3points, D2points, shape):791 def _util_get_Pose_POSIT(self, D3points, D2points):792 """ Calculate pose from head model "little girl" w/o camera or other793 calibrations needed.794 Method similar to '_util_get_Pose_solvePnP', please compare.795 D2points: left eye, right eye, nose, mouth796 """797 # calculate pose798 import opencv799 #opencv.unit_test()800 (rmat, tvec, mdl) = opencv.posit(D3points, D2points, (100, 1.0e-4))801 rvec = cv2.Rodrigues(rmat)[0]802 # Project the model points with the estimated pose803 # http://opencv.willowgarage.com/documentation/cpp/camera_calibration_and_3d_reconstruction.html804 # intrinsic: camera matrix805 # extrinsic: rotation-translation matrix [R|t]806 # CV_32F, principal point in the centre of the image is (0, 0) instead of (self.image_size[0]*0.5)807 FOCAL_LENGTH = 760.0 # hard-coded in posit_python.cpp, should be changed...808 cameraMatrix = [[FOCAL_LENGTH, 0.0, 0.0],#shape[0]*0.0],809 [ 0.0, FOCAL_LENGTH, 0.0],#shape[1]*0.0],810 [ 0.0, 0.0, 1.0],]811 # compare to 2D points812 err = []813 for i, vec in enumerate(np.array(mdl)):814 nvec = np.dot(cameraMatrix, (np.dot(rmat, vec) + tvec))815 err.append(((D2points[i] - nvec[:2]/nvec[2]), D2points[i], nvec[:2]/nvec[2]))816 #pywikibot.output(u'result for UN-calibrated camera:\n rot=%s' % rvec.transpose()[0])817 return rvec, tvec, np.array(cameraMatrix), (np.array(err)/1.0)818 # https://pypi.python.org/pypi/xbob.flandmark819 # http://cmp.felk.cvut.cz/~uricamic/flandmark/820 def _detect_FaceLandmark_xBOB(self):821 """Prints the locations of any face landmark(s) found, respective822 converts them to usual face position data"""823 scale = 1.824 try:825 #video = bob.io.VideoReader(self.image_path_JPEG.encode('utf-8'))826 video = [cv2.imread( self.image_path_JPEG, cv.CV_LOAD_IMAGE_COLOR )]827 #if img == None:828 # raise IOError829 830 # !!! the 'scale' here IS RELEVANT FOR THE DETECTION RATE;831 # how small and how many features are detected as faces (or eyes)832 scale = max([1., np.average(np.array(video[0].shape)[0:2]/750.)])833 except IOError:834 pywikibot.warning(u'unknown file type [_detect_FaceLandmark_xBOB]')835 return836 except AttributeError:837 pywikibot.warning(u'unknown file type [_detect_FaceLandmark_xBOB]')838 return839 smallImg = np.empty( (cv.Round(video[0].shape[1]/scale), cv.Round(video[0].shape[0]/scale)), dtype=np.uint8 )840 video = [ cv2.resize( img, smallImg.shape, interpolation=cv2.INTER_LINEAR ) for img in video ]841 sys.path.append(os.path.join(scriptdir, 'dtbext'))842 import _bob as bob843 import xbob_flandmark as xbob844 localize = xbob.flandmark.Localizer()845 result = []846 for frame in video: # currently ALWAYS contains ONE (1!) entry847 frame = np.transpose(frame, (2,0,1))848 img = np.transpose(frame, (1,2,0))849 for i, flm in enumerate(localize(frame)):850 #for pi, point in enumerate(flm['landmark']):851 # cv2.circle(img, tuple(map(int, point)), 3, ( 0, 0, 255))852 # cv2.circle(img, tuple(map(int, point)), 5, ( 0, 255, 0))853 # cv2.circle(img, tuple(map(int, point)), 7, (255, 0, 0))854 # cv2.putText(img, str(pi), tuple(map(int, point)), cv2.FONT_HERSHEY_PLAIN, 1.0, (0,255,0))855 #cv2.rectangle(img, tuple(map(int, flm['bbox'][:2])), tuple(map(int, (flm['bbox'][0]+flm['bbox'][2], flm['bbox'][1]+flm['bbox'][3]))), (0, 255, 0))856 mat = np.array([flm['landmark'][3], flm['landmark'][4]])857 mi = np.min(mat, axis=0)858 mouth = tuple(mi.astype(int)) + tuple((np.max(mat, axis=0)-mi).astype(int))859 #cv2.rectangle(img, tuple(mi.astype(int)), tuple(np.max(mat, axis=0).astype(int)), (0, 255, 0))860 mat = np.array([flm['landmark'][5], flm['landmark'][1]])861 mi = np.min(mat, axis=0)862 leye = tuple(mi.astype(int)) + tuple((np.max(mat, axis=0)-mi).astype(int))863 #cv2.rectangle(img, tuple(mi.astype(int)), tuple(np.max(mat, axis=0).astype(int)), (0, 255, 0))864 mat = np.array([flm['landmark'][2], flm['landmark'][6]])865 mi = np.min(mat, axis=0)866 reye = tuple(mi.astype(int)) + tuple((np.max(mat, axis=0)-mi).astype(int))867 #cv2.rectangle(img, tuple(mi.astype(int)), tuple(np.max(mat, axis=0).astype(int)), (0, 255, 0))868 data = { 'ID': (i+1),869 'Position': flm['bbox'], 870 'Type': u'Landmark',871 'Eyes': [leye, reye],872 'Mouth': mouth,873 'Nose': tuple(np.array(flm['landmark'][7]).astype(int)) + (0, 0),874 'Ears': [],875 'Landmark': [tuple(lm) for lm in np.array(flm['landmark']).astype(int)], }876 data['Coverage'] = float(data['Position'][2]*data['Position'][3])/(self.image_size[0]*self.image_size[1])877 result.append(data)878 #img = img.astype('uint8')879 #cv2.imshow("people detector", img)880 #cv2.waitKey()881 self._features['Faces'] += result882 return883 # .../opencv/samples/cpp/peopledetect.cpp884 # + Haar/Cascade detection885 def _detect_People(self):886 # http://stackoverflow.com/questions/10231380/graphic-recognition-of-people887 # https://code.ros.org/trac/opencv/ticket/1298888 # http://opencv.itseez.com/modules/gpu/doc/object_detection.html889 # http://opencv.willowgarage.com/documentation/cpp/basic_structures.html890 # http://www.pygtk.org/docs/pygtk/class-gdkrectangle.html891 scale = 1.892 try:893 img = cv2.imread(self.image_path_JPEG, cv.CV_LOAD_IMAGE_COLOR)894 if (img == None) or (min(img.shape[:2]) < 100) or (not img.data) \895 or (self.image_size[0] is None):896 return897 # !!! the 'scale' here IS RELEVANT FOR THE DETECTION RATE;898 # how small and how many features are detected899 #scale = max([1., np.average(np.array(img.shape)[0:2]/500.)])900 scale = max([1., np.average(np.array(img.shape)[0:2]/400.)])901 #scale = max([1., np.average(np.array(img.shape)[0:2]/300.)])902 except IOError:903 pywikibot.warning(u'unknown file type [_detect_People]')904 return905 except AttributeError:906 pywikibot.warning(u'unknown file type [_detect_People]')907 return908 # similar to face detection909 smallImg = np.empty( (cv.Round(img.shape[1]/scale), cv.Round(img.shape[0]/scale)), dtype=np.uint8 )910 #gray = cv2.cvtColor( img, cv.CV_BGR2GRAY )911 gray = img912 smallImg = cv2.resize( gray, smallImg.shape, interpolation=cv2.INTER_LINEAR )913 #smallImg = cv2.equalizeHist( smallImg )914 img = smallImg915 916 hog = cv2.HOGDescriptor()917 hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector())918 #cv2.namedWindow("people detector", 1)919 920 found = found_filtered = []921 #t = time.time()922 # run the detector with default parameters. to get a higher hit-rate923 # (and more false alarms, respectively), decrease the hitThreshold and924 # groupThreshold (set groupThreshold to 0 to turn off the grouping completely).925 # detectMultiScale(img, hit_threshold=0, win_stride=Size(),926 # padding=Size(), scale0=1.05, group_threshold=2)927 enable_recovery() # enable recovery from hard crash928 found = list(hog.detectMultiScale(img, 0.25, (8,8), (32,32), 1.05, 2))929 disable_recovery() # disable since everything worked out fine930 # people haar/cascaded classifier931 # use 'haarcascade_fullbody.xml', ... also (like face detection)932 xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_fullbody.xml')933 #xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_lowerbody.xml')934 #xml = os.path.join(scriptdir, 'externals/opencv/haarcascades/haarcascade_upperbody.xml')935 if not os.path.exists(xml):936 raise IOError(u"No such file: '%s'" % xml)937 cascade = cv2.CascadeClassifier(xml)938 objects = list(cascade.detectMultiScale( smallImg,939 1.1, 3, 0940 #|cv.CV_HAAR_FIND_BIGGEST_OBJECT941 #|cv.CV_HAAR_DO_ROUGH_SEARCH942 |cv.CV_HAAR_SCALE_IMAGE,943 (30, 30) ))944 found += objects945 #t = time.time() - t946 #print("tdetection time = %gms\n", t*1000.)947 bbox = gtk.gdk.Rectangle(*(0,0,img.shape[1],img.shape[0]))948 # exclude duplicates (see also in 'classifyFeatures()')949 found_filtered = [gtk.gdk.Rectangle(*f) for f in self._util_merge_Regions(found, sub=True)[0]]950 result = []951 for i in range(len(found_filtered)):952 r = found_filtered[i]953 # the HOG detector returns slightly larger rectangles than the real objects.954 # so we slightly shrink the rectangles to get a nicer output.955 r.x += cv.Round(r.width*0.1)956 r.width = cv.Round(r.width*0.8)957 r.y += cv.Round(r.height*0.07)958 r.height = cv.Round(r.height*0.8)959 data = { 'ID': (i+1), }960 #'Center': (int(r.x + r.width*0.5), int(r.y + r.height*0.5)), }961 # crop to image size (because of the slightly bigger boxes)962 r = bbox.intersect(r)963 #cv2.rectangle(img, (r.x, r.y), (r.x+r.width, r.y+r.height), cv.Scalar(0,255,0), 3)964 data['Position'] = tuple(np.int_(np.array(r)*scale))965 data['Coverage'] = float(data['Position'][2]*data['Position'][3])/(self.image_size[0]*self.image_size[1])966 result.append( data )967 #cv2.imshow("people detector", img)968 #c = cv2.waitKey(0) & 255969 self._features['People'] = result970 return971 def _detect_Geometry(self):972 result = self._util_get_Geometry_CVnSCIPY()973 self._features['Geometry'] = [{'Lines': result['Lines'],974 'Circles': result['Circles'],975 'Corners': result['Corners'],}]976 return977 # https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/python/houghlines.py?rev=2770978 def _util_get_Geometry_CVnSCIPY(self):979 # http://docs.opencv.org/modules/imgproc/doc/feature_detection.html#cornerharris980 # http://docs.opencv.org/modules/imgproc/doc/feature_detection.html#houghcircles981 # http://docs.opencv.org/modules/imgproc/doc/feature_detection.html#houghlines982 # http://docs.opencv.org/modules/imgproc/doc/feature_detection.html#houghlinesp983 if hasattr(self, '_buffer_Geometry'):984 return self._buffer_Geometry985 self._buffer_Geometry = {'Lines': '-', 'Circles': '-', 'Edge_Ratio': '-', 'Corners': '-',986 'FFT_Peaks': '-'}987 scale = 1.988 try:989 img = cv2.imread(self.image_path_JPEG, cv.CV_LOAD_IMAGE_COLOR)990 if (img == None):991 raise IOError992 # !!! the 'scale' here IS RELEVANT FOR THE DETECTION RATE;993 # how small and how many features are detected994 scale = max([1., np.average(np.array(img.shape)[0:2]/500.)])995 except IOError:996 pywikibot.warning(u'unknown file type [_detect_Geometry]')997 return self._buffer_Geometry998 except AttributeError:999 pywikibot.warning(u'unknown file type [_detect_Geometry]')1000 return self._buffer_Geometry1001 # similar to face or people detection1002 smallImg = np.empty( (cv.Round(img.shape[1]/scale), cv.Round(img.shape[0]/scale)), dtype=np.uint8 )1003 _gray = cv2.cvtColor( img, cv.CV_BGR2GRAY )1004 # smooth it, otherwise a lot of false circles may be detected1005 #gray = cv2.GaussianBlur( _gray, (9, 9), 2 )1006 gray = cv2.GaussianBlur( _gray, (5, 5), 2 )1007 smallImg = cv2.resize( gray, smallImg.shape, interpolation=cv2.INTER_LINEAR )1008 #smallImg = cv2.equalizeHist( smallImg )1009 src = smallImg1010 # https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/python/houghlines.py?rev=27701011 #dst = cv2.Canny(src, 50, 200)1012 dst = cv2.Canny(src, 10, 10)1013 edges = cv2.Canny(src, 10, 10)1014 #color_dst = cv2.cvtColor(dst, cv.CV_GRAY2BGR)1015 # edges (in this sensitve form a meassure for color gradients)1016 data = {}1017 data['Edge_Ratio'] = float((edges != 0).sum())/(edges.shape[0]*edges.shape[1])1018 # lines1019 USE_STANDARD = True1020 if USE_STANDARD:1021 #lines = cv.HoughLines2(dst, storage, cv.CV_HOUGH_STANDARD, 1, pi / 180, 100, 0, 0)1022 #lines = cv2.HoughLines(dst, 1, math.pi / 180, 100)1023 lines = cv2.HoughLines(dst, 1, math.pi / 180, 200)1024 if (lines is not None) and len(lines):1025 lines = lines[0]1026 data['Lines'] = len(lines)1027 #for (rho, theta) in lines[:100]:1028 # a = math.cos(theta)1029 # b = math.sin(theta)1030 # x0 = a * rho 1031 # y0 = b * rho1032 # pt1 = (cv.Round(x0 + 1000*(-b)), cv.Round(y0 + 1000*(a)))1033 # pt2 = (cv.Round(x0 - 1000*(-b)), cv.Round(y0 - 1000*(a)))1034 # cv2.line(color_dst, pt1, pt2, cv.RGB(255, 0, 0), 3, 8)1035 else:1036 #lines = cv.HoughLines2(dst, storage, cv.CV_HOUGH_PROBABILISTIC, 1, pi / 180, 50, 50, 10)1037 lines = cv2.HoughLinesP(dst, 1, math.pi / 180, 100) 1038 #for line in lines:1039 # cv2.line(color_dst, line[0], line[1], cv.CV_RGB(255, 0, 0), 3, 8)1040 # circles1041 try:1042 #circles = cv2.HoughCircles(src, cv.CV_HOUGH_GRADIENT, 2, src.shape[0]/4)#, 200, 100 )1043 circles = cv2.HoughCircles(src, cv.CV_HOUGH_GRADIENT, 2, src.shape[0]/4, param2=200)1044 except cv2.error:1045 circles = None1046 if (circles is not None) and len(circles):1047 circles = circles[0]1048 data['Circles'] = len(circles)1049 #for c in circles:1050 # center = (cv.Round(c[0]), cv.Round(c[1]))1051 # radius = cv.Round(c[2])1052 # # draw the circle center1053 # cv2.circle( color_dst, center, 3, cv.CV_RGB(0,255,0), -1, 8, 0 )1054 # # draw the circle outline1055 # cv2.circle( color_dst, center, radius, cv.CV_RGB(0,0,255), 3, 8, 0 )1056 # corners1057 corner_dst = cv2.cornerHarris( edges, 2, 3, 0.04 )1058 # Normalizing1059 cv2.normalize( corner_dst, corner_dst, 0, 255, cv2.NORM_MINMAX, cv.CV_32FC1 )1060 #dst_norm_scaled = cv2.convertScaleAbs( corner_dst )1061 # Drawing a circle around corners1062 corner = []1063 for j in range(corner_dst.shape[0]):1064 for i in range(corner_dst.shape[1]):1065 if corner_dst[j,i] > 200:1066 #circle( dst_norm_scaled, Point( i, j ), 5, Scalar(0), 2, 8, 0 );1067 corner.append( (j,i) )1068 data['Corners'] = len(corner)1069 #cv2.imshow("people detector", color_dst)1070 #c = cv2.waitKey(0) & 2551071 # fft spectral/frequency/momentum analysis with svd peak detection1072 gray = cv2.resize( _gray, smallImg.shape, interpolation=cv2.INTER_LINEAR )1073 ##s = (self.image_size[1], self.image_size[0])1074 #s = gray.shape1075 fft = fftpack.fftn(gray)1076 #fft = np.fft.fftn(gray)1077 #Image.fromarray(fft.real).show()1078 # shift quadrants so that low spatial frequencies are in the center1079 fft = fftpack.fftshift(fft)1080 #Image.fromarray(fft.real).show()1081 ##Image.fromarray(fftpack.ifftn(fft).real).show()1082 ##Image.fromarray(fftpack.ifftn(fftpack.ifftshift(fft)).real).show()1083 ##Image.fromarray(fftpack.ifftn(fftpack.ifftshift(fft.real)).real).show()1084 # (scipy svd has more options...)1085 #U, S, Vh = linalg.svd(np.matrix(fft)) # scipy; unstable, crashes with C core dump1086 #U, S, Vh = np.linalg.svd(np.matrix(fft)) # numpy (full matrix); unstable, ----"-----1087 #U, S, Vh = np.linalg.svd(np.matrix(fft), full_matrices=False) # less memory; more stable1088 S = np.linalg.svd(np.matrix(fft), compute_uv=False) # less memory, faster; more stable1089 ma = 0.01*max(S)1090 count = sum([int(c > ma) for c in S])1091 #SS = np.zeros(s)1092 #ss = min(s)1093 #for i in range(0, len(S)-1, max( int(len(S)/100.), 1 )): # (len(S)==ss) -> else; problem!1094 # #SS = np.zeros(s)1095 # #SS[:(ss-i),:(ss-i)] = np.diag(S[:(ss-i)])1096 # SS[:(i+1),:(i+1)] = np.diag(S[:(i+1)])1097 # #Image.fromarray((np.dot(np.dot(U, SS), Vh) - fft).real).show()1098 # #Image.fromarray(fftpack.ifftn(fftpack.ifftshift(np.dot(np.dot(U, SS), Vh))).real - gray).show()1099 # print i, ((np.dot(np.dot(U, SS), Vh) - fft).real).max()1100 # print i, (fftpack.ifftn(fftpack.ifftshift(np.dot(np.dot(U, SS), Vh))).real - gray).max()1101 # #if ((np.dot(np.dot(U, SS), Vh) - fft).max() < (255/4.)):1102 # # break1103 #data['SVD_Comp'] = float(i)/ss1104 #data['SVD_Min'] = S[:(i+1)].min()1105 data['FFT_Peaks'] = float(count)/len(S)1106 #pywikibot.output( u'FFT_Peaks: %s' % data['FFT_Peaks'] )1107 # use wavelet transformation (FWT) from e.g. pywt, scipy signal or mlpy1108 # (may be other) in addition to FFT and compare the spectra with FFT...1109 # confer; "A Practical Guide to Wavelet Analysis" (http://journals.ametsoc.org/doi/pdf/10.1175/1520-0477%281998%29079%3C0061%3AAPGTWA%3E2.0.CO%3B2)1110 # on how to convert and adopt FFT and wavlet spectra frequency scales1111 if data:1112 self._buffer_Geometry.update(data)1113 return self._buffer_Geometry1114 # .../opencv/samples/cpp/bagofwords_classification.cpp1115 def _detectclassify_ObjectAll(self):1116 """Uses the 'The Bag of Words model' for detection and classification"""1117 # CAN ALSO BE USED FOR: TEXT, ...1118 # http://app-solut.com/blog/2011/07/the-bag-of-words-model-in-opencv-2-2/1119 # http://app-solut.com/blog/2011/07/using-the-normal-bayes-classifier-for-image-categorization-in-opencv/1120 # http://authors.library.caltech.edu/7694/1121 # http://www.vision.caltech.edu/Image_Datasets/Caltech256/1122 # http://opencv.itseez.com/modules/features2d/doc/object_categorization.html1123 1124 # http://www.morethantechnical.com/2011/08/25/a-simple-object-classifier-with-bag-of-words-using-opencv-2-3-w-code/1125 # source: https://github.com/royshil/FoodcamClassifier1126 # http://app-solut.com/blog/2011/07/using-the-normal-bayes-classifier-for-image-categorization-in-opencv/1127 # source: http://code.google.com/p/open-cv-bow-demo/downloads/detail?name=bowdemo.tar.gz&can=2&q=1128 # parts of code here should/have to be placed into e.g. a own1129 # class in 'dtbext/opencv/__init__.py' script/module1130 1131 trained = ['aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus',1132 'car', 'cat', 'chair', 'cow', 'diningtable', 'dog',1133 'horse', 'motorbike', 'person', 'pottedplant', 'sheep',1134 'sofa', 'train', 'tvmonitor',]1135 bowDescPath = os.path.join(scriptdir, 'dtbext/opencv/data/bowImageDescriptors/000000.xml.gz')1136 # https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/cpp/bagofwords_classification.cpp?rev=37141137 # stand-alone (in shell) for training e.g. with:1138 # BoWclassify /data/toolserver/pywikipedia/dtbext/opencv/VOC2007 /data/toolserver/pywikipedia/dtbext/opencv/data FAST SURF BruteForce | tee run.log1139 # BoWclassify /data/toolserver/pywikipedia/dtbext/opencv/VOC2007 /data/toolserver/pywikipedia/dtbext/opencv/data HARRIS SIFT BruteForce | tee run.log1140 # http://experienceopencv.blogspot.com/2011/02/object-recognition-bag-of-keypoints.html1141 import opencv1142 #opencv.unit_test()1143 if os.path.exists(bowDescPath):1144 os.remove(bowDescPath)1145 stdout = sys.stdout1146 sys.stdout = StringIO.StringIO()1147 #result = opencv.BoWclassify.main(0, '', '', '', '', '')1148 result = opencv.BoWclassify(6, 1149 os.path.join(scriptdir, 'dtbext/opencv/VOC2007'), 1150 os.path.join(scriptdir, 'dtbext/opencv/data'), 1151 'HARRIS', # not important; given by training1152 'SIFT', # not important; given by training1153 'BruteForce', # not important; given by training1154 [str(os.path.abspath(self.image_path).encode('latin-1'))])1155 #out = sys.stdout.getvalue()1156 sys.stdout = stdout1157 #print out1158 if not result:1159 raise ImportError("BoW did not resolve; no results found!")1160 os.remove(bowDescPath)1161 # now make the algo working; confer also1162 # http://www.xrce.xerox.com/layout/set/print/content/download/18763/134049/file/2004_010.pdf1163 # http://people.csail.mit.edu/torralba/shortCourseRLOC/index.html1164 self._features['Classify'] = [dict([ (trained[i], r) for i, r in enumerate(result) ])]1165 return1166 def _detectclassify_ObjectAll_PYWT(self):1167 """Uses the 'Fast Wavelet-Based Visual Classification' for detection1168 and classification"""1169 # Fast Wavelet-Based Visual Classification1170 # http://www.cmap.polytechnique.fr/~yu/publications/ICPR08Final.pdf1171 # CAN ALSO BE USED FOR: TEXT, AUDIO, (VIDEO), ...1172# TODO: for audio and video (time-based) also...!!!1173 import pywt # python-pywt1174# TODO: improve (honestly; truly apply) wavelet in a meaningful and USEFUL (correct) way/manner!1175# TODO: truly apply FFT and SVD (used before)1176 # wavelet transformation1177 # https://github.com/nigma/pywt/tree/master/demo1178 # image_blender, dwt_signal_decomposition.py, wp_scalogram.py, dwt_multidim.py, user_filter_banks.py:1179 #coeffs = pywt.dwtn(gray, 'db1') # Single-level n-dimensional Discrete Wavelet Transform1180 coeffs = pywt.dwt2(gray, 'db1') # 2D Discrete Wavelet Transform1181 #coeffs = pywt.wavedec2(gray, 'db1') # Multilevel 2D Discrete Wavelet Transform1182 pass1183 result = pywt.idwt2(coeffs, 'db1') # 2D Inverse Discrete Wavelet Transform1184 #result = pywt.waverec2(coeffs, 'db1') # Multilevel 2D Inverse Discrete Wavelet Transform1185 result = result[:gray.shape[0],:gray.shape[1]]1186 # consider 'swt' (2D Stationary Wavelet Transform) instead of 'dwt' too1187 pywikibot.output(u'%s' % coeffs)1188 pywikibot.output(u'%s' % np.abs(result - gray).max())1189 #data['Wavelet_Comp'] = coeffs1190 # https://github.com/nigma/pywt/blob/master/demo/image_blender.py1191 # http://www.ncbi.nlm.nih.gov/pubmed/187136751192 # https://github.com/nigma/pywt/blob/master/demo/wp_scalogram.py1193 # https://github.com/nigma/pywt/blob/master/demo/swt2.py1194 return1195 # a lot more paper and possible algos exist; (those with code are...)1196 # http://www.lix.polytechnique.fr/~schwander/python-srm/1197 # http://library.wolfram.com/infocenter/Demos/5725/#downloads1198 # http://code.google.com/p/pymeanshift/wiki/Examples1199 # (http://pythonvision.org/basic-tutorial, http://luispedro.org/software/mahotas, http://packages.python.org/pymorph/)1200 def _detect_SegmentColors(self): # may be SLIC other other too...1201 try:1202 #im = Image.open(self.image_path).convert(mode = 'RGB')1203 im = Image.open(self.image_path_JPEG)1204 ## crop 25% of the image in order to give the bot a more human eye1205 ## (needed for categorization only and thus should be done there/later)1206 #scale = 0.75 # crop 25% percent (area) bounding box1207 #(w, h) = ( self.image_size[0]*math.sqrt(scale), self.image_size[1]*math.sqrt(scale) )1208 #(l, t) = ( (self.image_size[0]-w)/2, (self.image_size[1]-h)/2 )1209 #i = im.crop( (int(l), int(t), int(l+w), int(t+h)) )1210 (l, t) = (0, 0)1211 i = im1212 except IOError:1213 pywikibot.warning(u'unknown file type [_detect_SegmentColors]')1214 return1215 result = []1216 try:1217 #h = i.histogram() # average over WHOLE IMAGE1218 (pic, scale) = self._util_detect_ColorSegments_JSEG(i) # split image into segments first1219 #(pic, scale) = self._util_detect_ColorSegments_SLIC(i) # split image into superpixel first1220 hist = self._util_get_ColorSegmentsHist_PIL(i, pic, scale) #1221 #pic = self._util_merge_ColorSegments(pic, hist) # iteratively in order to MERGE similar regions1222 #(pic, scale_) = self._util_detect_ColorSegments_JSEG(pic) # (final split)1223 ##(pic, scale) = self._util_detect_ColorSegments_JSEG(pic) # (final split)1224 #hist = self._util_get_ColorSegmentsHist_PIL(i, pic, scale) #1225 except TypeError:1226 pywikibot.warning(u'unknown file type [_detect_SegmentColors]')1227 return1228 i = 01229 # (may be do an additional region merge according to same color names...)1230 for (h, coverage, (center, bbox)) in hist:1231 if (coverage < 0.05): # at least 5% coverage needed (help for debugging/log_output)1232 continue1233 data = self._util_average_Color_colormath(h)1234 data['Coverage'] = float(coverage)1235 data['ID'] = (i+1)1236 data['Center'] = (int(center[0]+l), int(center[1]+t))1237 data['Position'] = (int(bbox[0]+l), int(bbox[1]+t), int(bbox[2]), int(bbox[3]))1238 data['Delta_R'] = math.sqrt( (self.image_size[0]/2 - center[0])**2 + \1239 (self.image_size[1]/2 - center[1])**2 )1240 result.append( data )1241 i += 11242 self._features['ColorRegions'] = result1243 return1244 # http://stackoverflow.com/questions/2270874/image-color-detection-using-python1245 # https://gist.github.com/12462681246 # colormath-1.0.8/examples/delta_e.py, colormath-1.0.8/examples/conversions.py1247 # http://code.google.com/p/python-colormath/1248 # http://en.wikipedia.org/wiki/Color_difference1249 # http://www.farb-tabelle.de/en/table-of-color.htm1250 def _detect_AverageColor(self):1251 try:1252 # we need to have 3 channels (but e.g. grayscale 'P' has only 1)1253 #i = Image.open(self.image_path).convert(mode = 'RGB')1254 i = Image.open(self.image_path_JPEG)1255 h = i.histogram()1256 except IOError:1257 pywikibot.warning(u'unknown file type [_detect_AverageColor]')1258 return1259 result = self._util_average_Color_colormath(h)1260 result['Gradient'] = self._util_get_Geometry_CVnSCIPY().get('Edge_Ratio', None) or '-'1261 result['FFT_Peaks'] = self._util_get_Geometry_CVnSCIPY().get('FFT_Peaks', None) or '-'1262 self._features['ColorAverage'] = [result]1263 return1264 # http://stackoverflow.com/questions/2270874/image-color-detection-using-python1265 # https://gist.github.com/12462681266 # colormath-1.0.8/examples/delta_e.py, colormath-1.0.8/examples/conversions.py1267 # http://code.google.com/p/python-colormath/1268 # http://en.wikipedia.org/wiki/Color_difference1269 # http://www.farb-tabelle.de/en/table-of-color.htm1270 # http://www5.konicaminolta.eu/de/messinstrumente/color-light-language.html1271 def _util_average_Color_colormath(self, h):1272 # split into red, green, blue1273 r = h[0:256]1274 g = h[256:256*2]1275 b = h[256*2: 256*3]1276 1277 # perform the weighted average of each channel:1278 # the *index* 'i' is the channel value, and the *value* 'w' is its weight1279 rgb = (1280 sum( i*w for i, w in enumerate(r) ) / max(1, sum(r)),1281 sum( i*w for i, w in enumerate(g) ) / max(1, sum(g)),1282 sum( i*w for i, w in enumerate(b) ) / max(1, sum(b))1283 )1284 # count number of colors used more than 1% of maximum1285 ma = 0.01*max(h)1286 count = sum([int(c > ma) for c in h])1287# # TODO: peak detection (not supported by my local scipy version yet)1288# # http://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.find_peaks_cwt.html1289# peakind = signal.find_peaks_cwt(fft, np.arange(1,10))1290# print peaks1291# print len(peakind), peakind1292 data = { #'histogram': h,1293 'RGB': rgb,1294 'Peaks': float(count)/len(h), }1295 #colors = pycolorname.RAL.colors1296 #colors = pycolorname.pantone.Formula_Guide_Solid1297 colors = pycolorname.pantone.Fashion_Home_paper1298 1299 #print "=== RGB Example: RGB->LAB ==="1300 # Instantiate an Lab color object with the given values.1301 rgb = RGBColor(rgb[0], rgb[1], rgb[2], rgb_type='sRGB')1302 # Show a string representation.1303 #print rgb1304 # Convert RGB to LAB using a D50 illuminant.1305 lab = rgb.convert_to('lab', target_illuminant='D65')1306 #print lab1307 #print "=== End Example ===\n"1308 1309 # Reference color.1310 #color1 = LabColor(lab_l=0.9, lab_a=16.3, lab_b=-2.22)1311 # Color to be compared to the reference.1312 #color2 = LabColor(lab_l=0.7, lab_a=14.2, lab_b=-1.80)1313 color2 = lab1314 res = (1.E100, '')1315 for c in colors:1316 rgb = colors[c]1317 rgb = RGBColor(rgb[0], rgb[1], rgb[2], rgb_type='sRGB')1318 color1 = rgb.convert_to('lab', target_illuminant='D65')1319 #print "== Delta E Colors =="1320 #print " COLOR1: %s" % color11321 #print " COLOR2: %s" % color21322 #print "== Results =="1323 #print " CIE2000: %.3f" % color1.delta_e(color2, mode='cie2000')1324 ## Typically used for acceptability.1325 #print " CMC: %.3f (2:1)" % color1.delta_e(color2, mode='cmc', pl=2, pc=1)1326 ## Typically used to more closely model human percetion.1327 #print " CMC: %.3f (1:1)" % color1.delta_e(color2, mode='cmc', pl=1, pc=1)1328 r = color1.delta_e(color2, mode='cmc', pl=2, pc=1)1329 if (r < res[0]):1330 res = (r, c, colors[c])1331 data['Color'] = res[1]1332 data['Delta_E'] = res[0]1333 data['RGBref'] = res[2]1334 return data1335 def _util_detect_ColorSegments_JSEG(self, im):1336 tmpjpg = os.path.join(scriptdir, "cache/jseg_buf.jpg")1337 tmpgif = os.path.join(scriptdir, "cache/jseg_buf.gif")1338 # same scale func as in '_detect_Faces'1339 scale = max([1., np.average(np.array(im.size)[0:2]/200.)])1340 #print np.array(im.size)/scale, scale1341 try:1342 smallImg = im.resize( tuple(np.int_(np.array(im.size)/scale)), Image.ANTIALIAS )1343 except IOError:1344 pywikibot.warning(u'unknown file type [_util_detect_ColorSegments_JSEG]')1345 return1346 1347 #im.thumbnail(size, Image.ANTIALIAS) # size is 640x4801348 smallImg.convert('RGB').save(tmpjpg, "JPEG", quality=100, optimize=True)1349 1350 # Program limitation: The total number of regions in the image must be less1351 # than 256 before the region merging process. This works for most images1352 # smaller than 512x512.1353 1354 # Processing time will be about 10 seconds for an 192x128 image and 60 seconds1355 # for a 352x240 image. It will take several minutes for a 512x512 image.1356 # Minimum image size is 64x64.1357 1358 # ^^^ THUS RESCALING TO ABOUT 200px ABOVE ^^^1359 # sys.stdout handeled, but with freopen which could give issues1360 import jseg1361 # e.g. "segdist -i test3.jpg -t 6 -r9 test3.map.gif"1362 enable_recovery() # enable recovery from hard crash1363 jseg.segdist_cpp.main( [ item.encode('utf-8') for item in 1364 ("segdist -i %s -t 6 -r9 %s"%(tmpjpg, tmpgif)).split(" ") ] )1365 disable_recovery() # disable since everything worked out fine1366 #out = open((tmpgif + ".stdout"), "r").read() # reading stdout1367 #print out1368 os.remove(tmpgif + ".stdout")1369 1370 os.remove( tmpjpg )1371 1372 # http://stackoverflow.com/questions/384759/pil-and-numpy1373 pic = Image.open(tmpgif)1374 #pix = np.array(pic)1375 #Image.fromarray(10*pix).show()1376 1377 os.remove( tmpgif )1378 return (pic, scale)1379 # http://planet.scipy.org/1380 # http://peekaboo-vision.blogspot.ch/2012/05/superpixels-for-python-pretty-slic.html1381 # http://ivrg.epfl.ch/supplementary_material/RK_SLICSuperpixels/index.html1382 def _util_detect_ColorSegments_SLIC(self, img):1383 import slic1384 im = np.array(img)1385 image_argb = np.dstack([im[:, :, :1], im]).copy("C")1386 #region_labels = slic.slic_n(image_argb, 1000, 10)1387 region_labels = slic.slic_n(image_argb, 1000, 50)1388 slic.contours(image_argb, region_labels, 10)1389 #import matplotlib.pyplot as plt1390 #plt.imshow(image_argb[:, :, 1:].copy())1391 #plt.show()1392 #pic = Image.fromarray(region_labels)1393 #pic.show()1394 #return (pic, 1.)1395 return (region_labels, 1.)1396 def _util_get_ColorSegmentsHist_PIL(self, im, pic, scale):1397 if not (type(np.ndarray(None)) == type(pic)):1398 pix = np.array(pic)1399 #Image.fromarray(10*pix).show()1400 else:1401 pix = pic1402 #Image.fromarray(255*pix/np.max(pix)).show()1403 try:1404 smallImg = im.resize( tuple(np.int_(np.array(im.size)/scale)), Image.ANTIALIAS )1405 except IOError:1406 pywikibot.warning(u'unknown file type [_util_get_ColorSegmentsHist_PIL]')1407 return1408 imgsize = float(smallImg.size[0]*smallImg.size[1])1409 hist = []1410 for i in range(np.max(pix)+1):1411 mask = np.uint8(pix == i)*2551412 (y, x) = np.where(mask != 0)1413 center = (np.average(x)*scale, np.average(y)*scale)1414 bbox = (np.min(x)*scale, np.min(y)*scale, 1415 (np.max(x)-np.min(x))*scale, (np.max(y)-np.min(y))*scale)1416 #coverage = np.count_nonzero(mask)/imgsize1417 coverage = (mask != 0).sum()/imgsize # count_nonzero is missing in older numpy1418 mask = Image.fromarray( mask )1419 h = smallImg.histogram(mask)1420 #smallImg.show()1421 #dispImg = Image.new('RGBA', smallImg.size)1422 #dispImg.paste(smallImg, mask)1423 #dispImg.show()1424 if (len(h) == 256):1425 pywikibot.output(u"gray scale image, try to fix...")1426 h = h*31427 if (len(h) == 256*4):1428 pywikibot.output(u"4-ch. image, try to fix (exclude transparency)...")1429 h = h[0:(256*3)]1430 hist.append( (h, coverage, (center, bbox)) )1431 1432 return hist1433 # http://www.scipy.org/SciPyPackages/Ndimage1434 # http://www.pythonware.com/library/pil/handbook/imagefilter.htm1435 def _util_merge_ColorSegments(self, im, hist):1436 # merge regions by simplifying through average color and re-running1437 # JSEG again...1438 if not (type(np.ndarray(None)) == type(im)):1439 pix = np.array(im)1440 else:1441 pix = im1442 im = Image.fromarray(255*pix/np.max(pix))1443 im = im.convert('RGB')1444 for j, (h, coverage, (center, bbox)) in enumerate(hist):1445 # split into red, green, blue1446 r = h[0:256]1447 g = h[256:256*2]1448 b = h[256*2: 256*3]1449 1450 # perform the weighted average of each channel:1451 # the *index* 'i' is the channel value, and the *value* 'w' is its weight1452 rgb = (1453 sum( i*w for i, w in enumerate(r) ) / max(1, sum(r)),1454 sum( i*w for i, w in enumerate(g) ) / max(1, sum(g)),1455 sum( i*w for i, w in enumerate(b) ) / max(1, sum(b))1456 )1457 # color frequency analysis; do not average regions with high fluctations1458 #rgb2 = (1459 # sum( i*i*w for i, w in enumerate(r) ) / max(1, sum(r)),1460 # sum( i*i*w for i, w in enumerate(g) ) / max(1, sum(g)),1461 # sum( i*i*w for i, w in enumerate(b) ) / max(1, sum(b))1462 #)1463 #if ( 500. < np.average( (1464 # rgb2[0] - rgb[0]**2,1465 # rgb2[1] - rgb[1]**2,1466 # rgb2[2] - rgb[2]**2, ) ) ):1467 # continue1468 mask = np.uint8(pix == j)*2551469 mask = Image.fromarray( mask )1470 #dispImg = Image.new('RGB', im.size)1471 #dispImg.paste(rgb, mask=mask)1472 #dispImg.show()1473 im.paste(rgb, mask=mask)1474 pix = np.array(im)1475 pix[:,:,0] = ndimage.gaussian_filter(pix[:,:,0], .5)1476 pix[:,:,1] = ndimage.gaussian_filter(pix[:,:,1], .5)1477 pix[:,:,2] = ndimage.gaussian_filter(pix[:,:,2], .5)1478 im = Image.fromarray( pix, mode='RGB' )1479 #im = im.filter(ImageFilter.BLUR) # or 'SMOOTH'1480 return im1481 # Category:... (several; look at self.gatherFeatures for more hints)1482 def _detect_Trained(self, info_desc, cascade_file, maxdim=500.):1483 # general (self trained) classification (e.g. people, ...)1484 # http://www.computer-vision-software.com/blog/2009/11/faq-opencv-haartraining/1485 # Can be used with haar classifier (use: opencv_haartraining) and1486 # cascaded classifier (use: opencv_traincascade), both should work.1487 # !!! train a own cascade classifier like for face detection used1488 # !!! with 'opencv_haartraing' -> xml (file to use like in face/eye detection)1489 # analogue to face detection:1490 # http://tutorial-haartraining.googlecode.com/svn/trunk/data/haarcascades/1491 # or own xml files trained onto specific file database/set1492 xml = os.path.join(scriptdir, ('externals/opencv/haarcascades/' + cascade_file))1493 if not os.path.exists(xml):1494 raise IOError(u"No such file: '%s'" % xml)1495 cascade = cv2.CascadeClassifier(xml)1496 scale = 1.1497 try:1498 img = cv2.imread( self.image_path_JPEG, cv.CV_LOAD_IMAGE_COLOR )1499 if (img == None) or (self.image_size[0] is None):1500 raise IOError1501 # !!! the 'scale' here IS RELEVANT FOR THE DETECTION RATE;1502 # how small and how many features are detected1503 scale = max([1., np.average(np.array(img.shape)[0:2]/maxdim)])1504 except IOError:1505 pywikibot.warning(u'unknown file type [_detect_Trained]')1506 return1507 except AttributeError:1508 pywikibot.warning(u'unknown file type [_detect_Trained]')1509 return1510 # similar to face detection1511 smallImg = np.empty( (cv.Round(img.shape[1]/scale), cv.Round(img.shape[0]/scale)), dtype=np.uint8 )1512 gray = cv2.cvtColor( img, cv.CV_BGR2GRAY )1513 smallImg = cv2.resize( gray, smallImg.shape, interpolation=cv2.INTER_LINEAR )1514 smallImg = cv2.equalizeHist( smallImg )1515 objects = list(cascade.detectMultiScale( smallImg,1516 1.1, 5, 01517 #|cv.CV_HAAR_FIND_BIGGEST_OBJECT1518 #|cv.CV_HAAR_DO_ROUGH_SEARCH1519 |cv.CV_HAAR_SCALE_IMAGE,1520 (30, 30) ))1521 result = []1522 for i, r in enumerate(objects):1523 data = { 'Position': tuple(np.int_(np.array(r)*scale)) }1524 data['Coverage'] = float(data['Position'][2]*data['Position'][3])/(self.image_size[0]*self.image_size[1])1525 result.append( data )1526 # generic detection ...1527 self._features[info_desc] = result1528 return1529 def _recognize_OpticalCodes(self):1530 # barcode and Data Matrix recognition (libdmtx/pydmtx, zbar, gocr?)1531 # http://libdmtx.wikidot.com/libdmtx-python-wrapper1532 # http://blog.globalstomp.com/2011/09/decoding-qr-code-code-128-code-39.html1533 # http://zbar.sourceforge.net/1534 # http://pypi.python.org/pypi/zbar1535 1536 # DataMatrix1537 from pydmtx import DataMatrix # linux distro package (fedora) / TS (debian)1538 ## Write a Data Matrix barcode1539 #dm_write = DataMatrix()1540 #dm_write.encode("Hello, world!")1541 #dm_write.save("hello.png", "png")1542 scale = 1.1543 try:1544 # Read a Data Matrix barcode1545 dm_read = DataMatrix()1546 img = Image.open(self.image_path_JPEG)1547 #if (img == None) or (self.image_size[0] is None):1548 if (self.image_size[0] is None):1549 raise IOError1550 # http://libdmtx.wikidot.com/libdmtx-python-wrapper1551 if img.mode != 'RGB':1552 img = img.convert('RGB')1553 scale = max([1., np.average(np.array(img.size)/200.)])1554 except IOError:1555 pywikibot.warning(u'unknown file type [_recognize_OpticalCodes]')1556 return1557 smallImg = img.resize( (int(img.size[0]/scale), int(img.size[1]/scale)) )1558 img = smallImg1559 enable_recovery() # enable recovery from hard crash1560 #res = dm_read.decode(img.size[0], img.size[1], buffer(img.tostring()))1561 disable_recovery() # disable since everything worked out fine1562 #print res1563 result = []1564 i = -11565 for i in range(dm_read.count()):1566 data, bbox = dm_read.stats(i+1)1567 bbox = np.array(bbox)1568 x, y = bbox[:,0], bbox[:,1]1569 pos = (np.min(x), np.min(y), np.max(x)-np.min(x), np.max(y)-np.min(y))1570 result.append({ 'ID': (i+1),1571 #'Data': dm_read.message(i+1),1572 'Data': data,1573 'Position': pos,1574 'Type': u'DataMatrix',1575 'Quality': 10, })1576 1577 self._features['OpticalCodes'] = result1578 # supports many popular symbologies1579 try:1580 import zbar # TS (debian)1581 except:1582 import _zbar as zbar # other distros (fedora)1583 1584 try:1585 img = Image.open(self.image_path_JPEG).convert('L')1586 width, height = img.size1587 except IOError:1588 pywikibot.warning(u'unknown file type [_recognize_OpticalCodes]')1589 return1590 1591 scanner = zbar.ImageScanner()1592 scanner.parse_config('enable')1593 zbar_img = zbar.Image(width, height, 'Y800', img.tostring())1594 1595 # scan the image for barcodes1596 # http://zbar.sourceforge.net/api/zbar_8h.html1597 scanner.scan(zbar_img)1598 for symbol in zbar_img:1599 i += 11600 p = np.array(symbol.location) # list of points within code region/area1601 p = (min(p[:,0]), min(p[:,1]), (max(p[:,0])-min(p[:,0])), (max(p[:,1])-min(p[:,1])))1602 result.append({ #'components': symbol.components,1603 'ID': (i+1),1604 #'Count': symbol.count, # 'ID'?1605 'Data': symbol.data or u'-',1606 'Position': p, # (left, top, width, height)1607 'Quality': symbol.quality, # usable for 'Confidence'1608 'Type': symbol.type, })1609 1610 # further detection ?1611 self._features['OpticalCodes'] = result1612 return1613 def _detect_Chessboard(self):1614 # Chessboard (opencv reference detector)1615 # http://www.c-plusplus.de/forum/273920-full1616 # http://www.youtube.com/watch?v=bV-jAnQ-tvw1617 # http://nullege.com/codes/show/src%40o%40p%40opencvpython-HEAD%40samples%40chessboard.py/12/cv.FindChessboardCorners/python1618 scale = 1.1619 try:1620 #im = cv.LoadImage(self.image_path_JPEG, cv.CV_LOAD_IMAGE_COLOR)1621 im = cv2.imread( self.image_path_JPEG, cv2.CV_LOAD_IMAGE_GRAYSCALE )1622 #im = cv2.imread( 'Mutilated_checkerboard_3_1.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE )1623 #im = cv2.imread( 'Jogo_de_Damas_-_Acatabul.JPG', cv2.CV_LOAD_IMAGE_GRAYSCALE )1624 chessboard_dim = ( 7, 7 )1625 if im == None:1626 raise IOError1627 scale = max([1., np.average(np.array(im.shape)[0:2]/1000.)])1628 #scale = max([1., np.average(np.array(im.shape)[0:2]/500.)])1629 #scale = max([1., np.average(np.array(im.shape)[0:2]/450.)])1630 except IOError:1631 pywikibot.warning(u'unknown file type [_detect_Chessboard]')1632 return1633 except AttributeError:1634 pywikibot.warning(u'unknown file type [_detect_Chessboard]')1635 return1636 smallImg = np.empty( (cv.Round(im.shape[1]/scale), cv.Round(im.shape[0]/scale)), dtype=np.uint8 )1637 #gray = cv2.cvtColor( im, cv.CV_BGR2GRAY )1638 smallImg = cv2.resize( im, smallImg.shape, interpolation=cv2.INTER_LINEAR )1639 #smallImg = cv2.equalizeHist( smallImg )1640 im = smallImg1641 found_all = False1642 corners = None1643 try:1644 #found_all, corners = cv.FindChessboardCorners( im, chessboard_dim )1645 found_all, corners = cv2.findChessboardCorners( im, chessboard_dim )1646 except cv2.error:1647 pywikibot.exception(tb=True)1648 #cv2.drawChessboardCorners( im, chessboard_dim, corners, found_all )1649 ##cv2.imshow("win", im)1650 ##cv2.waitKey()1651 result = {}1652 if corners is not None:1653 result = { 'Corners': [tuple(item[0]) for item in corners], }1654 self._features['Chessboard'] = [result]1655# TODO: improve chessboard detection (make it more tolerant)1656# ## http://stackoverflow.com/questions/7624765/converting-an-opencv-image-to-black-and-white1657# #im_gray = im1658# #im_gray_mat = cv.fromarray(im_gray)1659# #im_bw = cv.CreateImage(cv.GetSize(im_gray_mat), cv.IPL_DEPTH_8U, 1)1660# #im_bw_mat = cv.GetMat(im_bw)1661# #cv.Threshold(im_gray_mat, im_bw_mat, 0, 255, cv.CV_THRESH_BINARY | cv.CV_THRESH_OTSU)1662# #im = np.asarray(im_bw_mat)1663#1664# # chess board recognition (more tolerant)1665# # http://codebazaar.blogspot.ch/2011/08/chess-board-recognition-project-part-1.html1666# # https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/python/houghlines.py?rev=27701667# # http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/canny_detector/canny_detector.html1668# dst = im.copy()1669# color_dst = cv2.cvtColor(dst, cv.CV_GRAY2BGR)1670# dst = cv2.GaussianBlur(dst, (3, 3), 5)1671# thr = 1501672# dst = cv2.Canny(dst, thr, 3*thr)1673# cv2.imshow("win", dst)1674# cv2.waitKey()1675# # lines to find grid1676# # http://dsp.stackexchange.com/questions/2420/alternatives-to-hough-transform-for-detecting-a-grid-like-structure1677# USE_STANDARD = True1678# if USE_STANDARD:1679# #lines = cv.HoughLines2(dst, storage, cv.CV_HOUGH_STANDARD, 1, pi / 180, 100, 0, 0)1680# #lines = cv2.HoughLines(dst, 1, math.pi / 180, 100)1681# lines = cv2.HoughLines(dst, 1, math.pi / 180, 150)1682# if (lines is not None) and len(lines):1683# lines = lines[0]1684# #data['Lines'] = len(lines)1685#1686# ls = np.array(lines)1687# import pylab1688# (n, bins, patches) = pylab.hist(ls[:,1])1689# print n, bins, patches1690# pylab.grid(True)1691# pylab.show()1692#1693# for (rho, theta) in lines:1694# #if theta > 0.3125: continue1695# a = math.cos(theta)1696# b = math.sin(theta)1697# x0 = a * rho 1698# y0 = b * rho1699# pt1 = (cv.Round(x0 + 1000*(-b)), cv.Round(y0 + 1000*(a)))1700# pt2 = (cv.Round(x0 - 1000*(-b)), cv.Round(y0 - 1000*(a)))1701# cv2.line(color_dst, pt1, pt2, cv.RGB(255, 0, 0), 3, 8)1702# else:1703# #lines = cv.HoughLines2(dst, storage, cv.CV_HOUGH_PROBABILISTIC, 1, pi / 180, 50, 50, 10)1704# lines = cv2.HoughLinesP(dst, 1, math.pi / 180, 100) 1705#1706# for line in lines[0]:1707# print line1708# cv2.line(color_dst, tuple(line[0:2]), tuple(line[2:4]), cv.CV_RGB(255, 0, 0), 3, 8)1709# cv2.imshow("win", color_dst)1710# cv2.waitKey()1711 if found_all:1712 # pose detection1713 # http://docs.opencv.org/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html1714 # http://stackoverflow.com/questions/10022568/opencv-2-3-camera-calibration1715 d = shelve.open( os.path.join(scriptdir, 'externals/opencv/camera_virtual_default') )1716 if ('retval' not in d):1717 # http://commons.wikimedia.org/wiki/File:Mutilated_checkerboard_3.jpg1718 pywikibot.output(u"Doing (virtual) camera calibration onto reference image 'File:Mutilated_checkerboard_3.jpg'")1719 im3 = cv2.imread( 'Mutilated_checkerboard_3.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE )1720 im3 = cv2.resize( im3, (cv.Round(im3.shape[1]/scale), cv.Round(im3.shape[0]/scale)), interpolation=cv2.INTER_LINEAR )1721 # Compute the the three dimensional world-coordinates1722 tmp = []1723 for h in range(chessboard_dim[0]):1724 for w in range(chessboard_dim[1]):1725 tmp.append( (float(h), float(w), 0.0) )1726 objectPoints = np.array(tmp)1727 # Compute matrices1728 _found_all, _corners = cv2.findChessboardCorners( im3, chessboard_dim, flags=cv.CV_CALIB_CB_ADAPTIVE_THRESH | cv.CV_CALIB_CB_FILTER_QUADS )1729 #cv2.drawChessboardCorners( im3, chessboard_dim, _corners, _found_all )1730 retval, cameraMatrix, distCoeffs, rvecs, tvecs = cv2.calibrateCamera([objectPoints.astype('float32')], [_corners.astype('float32')], im3.shape, np.eye(3), np.zeros((5, 1)))1731 fovx, fovy, focalLength, principalPoint, aspectRatio = cv2.calibrationMatrixValues(cameraMatrix, im3.shape, 1.0, 1.0)1732 d['objectPoints'] = [objectPoints.astype('float32')] # shape: (49, 3) in a list of 1 item1733 d['imagePoints'] = [_corners.astype('float32')] # shape: (49, 1, 2) in a list of 1 item1734 d['cameraMatrix'] = cameraMatrix1735 d['distCoeffs'] = distCoeffs1736 d['rvecs'] = rvecs1737 d['tvecs'] = tvecs1738 d['imageSize'] = im3.shape1739 d['apertureWidth'] = 1.01740 d['apertureHeight'] = 1.01741 d['fovx'] = fovx1742 d['fovy'] = fovy1743 d['focalLength'] = focalLength1744 d['principalPoint'] = principalPoint1745 d['aspectRatio'] = aspectRatio1746 d['retval'] = retval1747 else:1748 objectPoints = d['objectPoints'][0]1749 cameraMatrix, distCoeffs = d['cameraMatrix'], d['distCoeffs']1750 # would be nice to use these:1751 #cameraMatrix, distCoeffs = np.eye(3), np.zeros((5,1))1752 # ..,since they are simple... else other have to be documented as "used calibration" !!!1753 d.close()1754 # http://answers.opencv.org/question/1073/what-format-does-cv2solvepnp-use-for-points-in/1755 rvec, tvec = cv2.solvePnP(objectPoints, corners, cameraMatrix, distCoeffs)1756 #rvec, tvec = cv2.solvePnP(objectPoints, corners, cameraMatrix, None)1757 # http://www.opencv.org.cn/opencvdoc/2.3.2/html/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html1758 # -> what about using POSIT ??? (see docs on enwiki)1759 # http://opencv.willowgarage.com/wiki/Posit1760 #(cv2.findFundamentalMat, cv2.findHomography or from 'pose', cv2.estimateAffine3D)1761 # (todo) draw the rotated 3D object (projected down to 2D)1762 im = cv2.cvtColor(im, cv2.COLOR_GRAY2BGR)1763 ## debug: axis-cross(es) - gives strange/wrong results1764 #for k in range(3): # row1765 # for j in range(5): # column1766 # rmat = cv2.Rodrigues(2*3.14/5.*j*np.array(np.eye(3)[:,k]))[0]1767 # mat, perp = self._util_getD2coords_proj( np.dot(rmat, np.eye(3)), cameraMatrix, None, None, distCoeffs=distCoeffs, sign=-1 )1768 # self._util_drawAxes(mat, 50+100*j, k*100+50, im)1769 ## debug: rotated axis-cross1770 #mat, perp = self._util_getD2coords_proj( np.eye(3), cameraMatrix, rvec, tvec, distCoeffs=distCoeffs )1771 #self._util_drawAxes(mat, 50, 350, im)1772 ## debug: self-calculated rotated axis-cross - gives strange/wrong results1773 #mat = np.dot((cameraMatrix), np.dot(cv2.Rodrigues(rvec)[0], np.eye(3)))1774 ##mat, perp = self._util_getD2coords_proj( mat, np.eye(3), None, None, distCoeffs=distCoeffs, sign=-1 )1775 #mat, perp = self._util_getD2coords_proj( mat, np.eye(3), None, None, distCoeffs=np.zeros((5,1)) )1776 #self._util_drawAxes(mat, 150, 350, im)1777 # debug: self-calculated rotated axis-cross - results looks good: OK1778 # (and can be calculated in order to give numerical results)1779 #rvec = np.zeros(3)1780 rot = rvec1781 mat, perp = self._util_getD2coords_calc(np.eye(3), cameraMatrix, rvec, tvec)1782 ortho = mat[:2,2]1783 ortho = ortho/np.linalg.norm(ortho)1784 #self._util_drawAxes(mat, 250, 350, im)1785 #self._util_drawAxes(mat, 50, 50, im)1786# TODO: compare face and chessboard pose estimations and unify them, then document everything (template in wiki, ...)1787 pywikibot.output(u'result for calibrated camera:\n rot=%s\n perp=%s\n perp2D=%s' % (rot.transpose()[0], perp[:,2], ortho))1788 pywikibot.output(u'nice would be to do the same for uncalibrated/default cam settings')1789 result.update({ 'Rotation': tuple(rot.transpose()[0]),1790 'Perp_Dir' : tuple(perp[:,2]),1791 'Perp_Dir_2D': tuple(ortho), })1792 self._features['Chessboard'] = [result]1793 #cv2.imshow("win", im)1794 #cv2.waitKey()1795 return1796# def _util_getD2coords_proj(self, D3coords, cameraMatrix, rvec=None, tvec=None, distCoeffs=np.zeros((5,1)), sign=1):1797# """Project 3D points down to 2D by using OpenCV functions."""1798# if rvec is None:1799# rvec = np.zeros((3,1))1800# if tvec is None:1801# tvec = np.zeros((3,1))1802# mat = np.zeros((2,D3coords.shape[0]))1803# matnorm = np.zeros((1,D3coords.shape[0]))1804# for i in range(D3coords.shape[0]):1805# D2raw, jacobian = cv2.projectPoints(np.array([[0.,0.,5.],[D3coords[0,i],D3coords[1,i],D3coords[2,i]+5.]]), rvec, tvec, cameraMatrix, distCoeffs)1806# D2norm = (D2raw[1][0]-D2raw[0][0])1807# #D2norm[1] *= sign # usual 2D coords <-> pixel/picture coords1808# D2norm[0] *= sign # usual 2D coords <-> pixel/picture coords1809# D2norm *= sign # invert all1810# mat[:,i] = D2norm1811# matnorm[:,i] = np.linalg.norm(D2norm)1812# mat = mat/max(matnorm[0])1813# return (mat, D3coords)1814 def _util_getD2coords_calc(self, D3coords, cameraMatrix, rvec, tvec, hacky=True):1815 """Calculate s m' = A [R|t] M' in order to project 3D points down to 2D.1816 m' = (u, v, 1)^T, M' = (X, Y, Z, 1)^T, A: camera m. and [R|t]: rotation-1817 translation matrix.1818 @see http://docs.opencv.org/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html1819 """1820 # cv2.decomposeProjectionMatrix(...)1821 cm = cameraMatrix.copy()1822 cm[0:2,2] = [0., 0.]1823 rmat = np.zeros((3,4))1824 # http://en.wikipedia.org/wiki/Rodrigues%27_rotation_formula1825 rmat[:,0:3] = cv2.Rodrigues(rvec)[0]1826 #rmat[:,0:3] = np.eye(3)1827 rmat[:,3] = tvec[:,0]1828 origin = np.dot(rmat, cv2.convertPointsToHomogeneous(np.zeros((3,3)).astype('float32')).transpose()[:,0,:])1829 origin2D = np.dot((cm), origin) # np.linalg.inv(cm)1830 #coords = np.dot(cv2.Rodrigues(rvec)[0], D3coords)1831 coords = np.dot(rmat, cv2.convertPointsToHomogeneous(D3coords.astype('float32')).transpose()[:,0,:])1832 coords2D = np.dot((cm), coords)1833 perp = coords - origin1834 if hacky:1835 # for '_detect_Chessboard' but looks a bit strange ... may be wrong?!1836 mat = coords2D - origin2D1837 mat = mat/max([np.linalg.norm(mat[:,i]) for i in range(3)])1838 else:1839 for i in range(3): # rescale with s1840 coords2D[:,i] /= coords2D[2,i]1841 origin2D[:,i] /= origin2D[2,i]1842 mat = coords2D - origin2D1843 # simple'n'fast solution, if just 2D results are needed1844 #mat, jacobian = cv2.projectPoints(np.append(np.zeros((1,3)), 1845 # D3coords, 1846 # axis=0),1847 # rvec, tvec, cm, np.zeros((5,1)))1848 #mat = mat[:,0,:]1849 #mat = (mat[1:,:] - mat[0,:]).transpose()1850 return (mat, perp)1851# def _util_drawAxes(self, mat, x, y, im):1852# color = [(0., 0., 255.), (0., 255., 0.), (255., 0., 0.)]1853# label = ['x', 'y', 'z']1854# for i in range(3):1855# D2norm = 40*mat[:,i]1856# cv2.line(im, (x,y), (x+D2norm[0].astype(int),y+D2norm[1].astype(int)), color[i], 1)1857# cv2.putText(im, label[i], (x+D2norm[0].astype(int),y+D2norm[1].astype(int)), cv2.FONT_HERSHEY_PLAIN, 1., color[i])1858 def _detect_Faces_EXIF(self):1859 res = self._util_get_DataTags_EXIF()1860 1861 # http://u88.n24.queensu.ca/exiftool/forum/index.php?topic=3156.01862 # http://u88.n24.queensu.ca/pub/facetest.pl1863 # ( all scaling stuff ignored (!) and some strongly simplified (!) )1864 # Example: 'File:Annagrah-2 041.JPG' (canon)1865 if 'Make' in res:1866 make = res['Make'].lower()1867 else:1868 make = ''1869 found = set(res.keys())1870 data = []1871 if 'ImageWidth' in res:1872 (width, height) = (str(res['ImageWidth']), str(res['ImageHeight']))1873 (width, height) = (re.sub(u'p[tx]', u'', width), re.sub(u'p[tx]', u'', height))1874 try:1875 (width, height) = (int(float(width)+0.5), int(float(height)+0.5))1876 except ValueError:1877 pywikibot.warning(u'%s contains incompatible unit(s), skipped' % ((width, height),))1878 return1879 else:1880 (width, height) = self.image_size1881 wasRotated = (height > width)1882 1883 if True in [item in make for item in ['sony', 'nikon', 'panasonic', 'casio', 'ricoh']]:1884 # UNTESTED: ['sony', 'nikon', 'casio', 'ricoh']1885 # TESTED: ['panasonic']1886 if set(['FacesDetected', 'Face1Position']).issubset(found):1887 i = 11888 if 'FaceOrientation' in res:1889 pywikibot.output(res['FaceOrientation']) # for rotation 'rot'1890 # 'crop' for 'casio' omitted here...1891 aspect = float(height)/width1892 if (aspect <= 3./4):1893 (fw, fh) = (320, 320 * aspect)1894 else:1895 (fw, fh) = (240 / aspect, 240)1896 #(sx, sy) = (1./width, 1./height)1897 (sx, sy) = (1./fw, 1./fh)1898 if 'FaceDetectFrameSize' in res:1899 (width, height) = map(int, res['FaceDetectFrameSize'].split(' '))1900 (sx, sy) = (1./width, 1./height)1901 while (('Face%iPosition'%i) in res) and (i <= int(res['FacesDetected'])):1902 buf = map(int, res['Face%iPosition'%i].split(' '))1903 (x1, y1) = ((buf[0]-buf[2]/2)*sx, (buf[1]-buf[3]/2)*sy) # 'panasonic'1904 (x2, y2) = (x1+buf[2]*sx, y1+buf[3]*sy) #1905 #(x1, y1) = (buf[1]*sx, buf[0]*sy)1906 #(x2, y2) = (x1+buf[3]*sx, y1+buf[2]*sy)1907 data.append({ 'Position': (x1, y1, x2, y2) })1908 if ('RecognizedFace%iName'%i) in res:1909 pywikibot.output(str((res['RecognizedFace%iName'%i], res['RecognizedFace%iAge'%i])))1910 i += 11911 elif 'fujifilm' in make:1912 # UNTESTED: 'fujifilm'1913 if set(['FacesDetected', 'FacePositions']).issubset(found):1914 buf = map(int, res['FacePositions'].split(' '))1915 (sx, sy) = (1./width, 1./height)1916 for i in range(int(res['FacesDetected'])):1917 data.append({ 'Position': [buf[i*4]*sx, buf[i*4+1]*sy, 1918 buf[i*4+2]*sx, buf[i*4+3]*sy] })1919 if ('Face%iName'%i) in res:1920 pywikibot.output(str((res['Face%iName'%i], res['Face%iCategory'%i], res['Face%iBirthday'%i])))1921 elif 'olympus' in make:1922 # UNTESTED: 'olympus'1923 if set(['FacesDetected', 'FaceDetectArea']).issubset(found):1924 buf = map(int, res['FacesDetected'].split(' '))1925 if buf[0] or buf[1]:1926 buf = map(int, res['FaceDetectArea'].split(' '))1927 for i in range(int(res['MaxFaces'])):1928 data.append({ 'Position': [buf[i*4], buf[i*4+1], buf[i*4+2], buf[i*4+3]] })1929 elif True in [item in make for item in ['pentax', 'sanyo']]:1930 # UNTESTED: ['pentax', 'sanyo']1931 if set(['FacesDetected']).issubset(found):1932 i = 11933 (sx, sy) = (1./width, 1./height)1934 while ('Face%iPosition'%i) in res:1935 buf = map(int, res['Face%iPosition'%i].split(' ') + \1936 res['Face%iSize'%i].split(' '))1937 (x1, y1) = ((buf[0] - buf[2]/2.)*sx, (buf[1] - buf[3]/2.)*sy)1938 (x2, y2) = (x1+buf[2]*sx, y1+buf[3]*sy)1939 data.append({ 'Position': (x1, y1, x2, y2) })1940 i += 11941 if 'FacePosition' in res:1942 buf = map(int, res['FacePosition'].split(' ') + ['100', '100']) # how big is the face?1943 (x1, y1) = (buf[0]*sx, buf[1]*sy)1944 (x2, y2) = (buf[2]*sx, buf[3]*sy)1945 data.append({ 'Position': (x1, y1, x2, y2) })1946 elif 'canon' in make:1947 if set(['FacesDetected', 'FaceDetectFrameSize']).issubset(found) \1948 and (int(res['FacesDetected'])):1949 # TESTED: older models store face detect information1950 (width, height) = map(int, res['FaceDetectFrameSize'].split(' ')) # default: (320,240)1951 (sx, sy) = (1./width, 1./height)1952 fw = res['FaceWidth'] or 351953 i = 11954 while ('Face%iPosition'%i) in res:1955 buf = map(int, res['Face%iPosition'%i].split(' '))1956 (x1, y1) = ((buf[0] + width/2. - fw)*sx, (buf[1] + height/2. - fw)*sy)1957 (x2, y2) = (x1 + fw*2*sx, y1 + fw*2*sy)1958 data.append({ 'Position': (x1, y1, x2, y2) })1959 i += 11960 elif set(['ValidAFPoints', 'AFImageWidth', 'AFImageHeight',1961 'AFAreaXPositions', 'AFAreaYPositions', 'PrimaryAFPoint']).issubset(found):1962 # TESTED: newer models use AF points1963 (width, height) = (int(res['AFImageWidth']), int(res['AFImageHeight']))1964 if ('AFAreaMode' in res) and ('Face' in res['AFAreaMode']):1965 buf_x = res['AFAreaXPositions'].split(' ')1966 buf_y = res['AFAreaYPositions'].split(' ')1967 buf_w = buf_h = [100] * len(buf_x) # how big is the face? (else)1968 if 'AFAreaWidths' in res:1969 buf_w = map(int, res['AFAreaWidths'].split(' '))1970 buf_h = map(int, res['AFAreaHeights'].split(' '))1971 elif 'AFAreaWidth' in res:1972 buf_w = [int(res['AFAreaWidth'])] * len(buf_x)1973 buf_h = [int(res['AFAreaHeight'])] * len(buf_x)1974 else:1975 pywikibot.output(u'No AF area size')1976 # conversion to positive coordinates1977 buf_x = [ int(x) + width/2. for x in buf_x ]1978 buf_y = [ int(y) + height/2. for y in buf_y ]1979 # EOS models have Y flipped1980 if ('Model' in res) and ('EOS' in res['Model']):1981 buf_y = [ height - y for y in buf_y ]1982 (sx, sy) = (1./width, 1./height)1983 for i in range(int(res['ValidAFPoints'])):1984 (x1, y1) = ((buf_x[i]-buf_w[i]/2)*sx, (buf_y[i]-buf_h[i]/2)*sy)1985 (x2, y2) = (x1+buf_w[i]*sx, y1+buf_h[i]*sy)1986 data.append({ 'Position': (x1, y1, x2, y2) })1987 else:1988 # not supported (yet...)1989 available = [item in res for item in ['FacesDetected', 'ValidAFPoints']]1990 unknown = ['face' in item.lower() for item in res.keys()]1991 if make and (True in (available+unknown)):1992 pywikibot.warning(u"skipped '%s' since not supported (yet) [_detect_Faces_EXIF]" % make)1993 pywikibot.warning(u"FacesDetected: %s - ValidAFPoints: %s" % tuple(available))1994 1995 # finally, rotate face coordinates if image was rotated1996 if wasRotated:1997 rot = 2701998 # variable rotation omitted here... ($$faceInfo{Rotation})1999 for i, d in enumerate(data):2000 # rotate face coordinates2001 p = data[i]['Position']2002 if wasRotated:2003 if (rot == 90):2004 p = (p[1], 1-p[0], p[3], 1-p[2])2005 else:2006 p = (1-p[1], p[0], 1-p[3], p[2])2007 if 'Rotation' in data[i]:2008 data[i]['Rotation'] -= rot2009 data[i]['Rotation'] += 360 if data[i]['Rotation'] < 0 else 02010 # rescale relative sizes to real pixel values2011 p = (p[0]*self.image_size[0] + 0.5, p[1]*self.image_size[1] + 0.5, 2012 p[2]*self.image_size[0] + 0.5, p[3]*self.image_size[1] + 0.5)2013 # change from (x1, y1, x2, y2) to (x, y, w, h)2014 #data[i]['Position'] = (p[0], p[1], p[0]-p[2], p[3]-p[1])2015 data[i]['Position'] = (min(p[0],p[2]), min(p[1],p[3]), 2016 abs(p[0]-p[2]), abs(p[3]-p[1]))2017 data[i] = { 'Position': tuple(map(int, data[i]['Position'])),2018 'ID': (i+1),2019 'Type': u'Exif',2020 'Eyes': [],2021 'Mouth': (),2022 'Nose': (), }2023 data[i]['Coverage'] = float(data[i]['Position'][2]*data[i]['Position'][3])/(self.image_size[0]*self.image_size[1])2024 # (exclusion of duplicates is done later by '_util_merge_Regions')2025 self._features['Faces'] += data2026 return2027 def _util_merge_Regions(self, regs, sub=False, overlap=False, close=False):2028 # sub=False, overlap=False, close=False ; level 0 ; similar regions, similar position (default)2029 # sub=True, overlap=False, close=False ; level 1 ; region contained in other, any shape/size2030 # sub=False, overlap=True, close=False ; level 2 ; center of region conatained in other2031 # sub=False, overlap=False, close=True ; level 3 ; regions placed close together2032 if not regs:2033 return ([], [])2034 dmax = np.linalg.norm(self.image_size)2035 #thsr = 1.0 # strict: if it is contained completely2036 thsr = 0.95 # a little bit tolerant: nearly completly contained (or 0.9)2037 drop = []2038 for i1, r1i in enumerate(regs):2039 r1 = np.float_(r1i)2040 (xy1, wh1) = (r1[0:2], r1[2:4])2041 c1 = xy1 + wh1/22042 a1 = wh1[0]*wh1[1]2043 # check for duplicates (e.g. similar regions in similar position)2044 i2 = 02045 while (i2 < i1):2046 r2i, r2 = regs[i2], np.float_(regs[i2])2047 (xy2, wh2) = (r2[0:2], r2[2:4])2048 c2 = xy2 + wh2/22049 a2 = wh2[0]*wh2[1]2050 dr = np.linalg.norm(c1-c2)/dmax2051 intersect = gtk.gdk.Rectangle(*r1i).intersect(gtk.gdk.Rectangle(*r2i))2052 area = intersect.width*intersect.height2053 ar1, ar2 = area/a1, area/a22054 check = [(1-dr), ar1, ar2]2055 # (I assume the 1. condition (1-dr) to be always true if the 2.2056 # and 3. are - so it's obsolete... how is the analytic relation?)2057 # add the first match (first is assumed to be the best one) / drop second one2058 #print check, np.average(check), np.std(check)2059 if (np.average(check) >= 0.9) and (np.std(check) <= 0.1):2060 #if (np.average(check) >= 0.85) and (np.std(check) <= 0.1):2061 drop.append( i1 )2062 # remove all sub-rect/-regions (all regions fully contained in other)2063 if sub:2064 #drop.append( [i1, i2][check[0:2].index(1.0)] )2065 if (ar1 >= thsr) and (i2 not in drop):2066 drop.append( i1 )2067 elif (ar2 >= thsr) and (i1 not in drop):2068 drop.append( i2 )2069 # from '_detect_Faces()'2070 if overlap:2071 if (r2[0] <= c1[0] <= (r2[0] + r2[2])) and \2072 (r2[1] <= c1[1] <= (r2[1] + r2[3])) and (i2 not in drop):2073 drop.append( i1 )2074 if close:2075 if (check[0] >= 0.985) and (i2 not in drop): # at least (!)2076 drop.append( i1 )2077 i2 += 12078 drop = sorted(list(set(drop)))2079 drop.reverse()2080 for i in drop:2081 del regs[i]2082 return (regs, drop)2083class _PngFile(_JpegFile):2084 pass2085class _GifFile(_JpegFile):2086 pass2087class _TiffFile(_JpegFile):2088 pass2089class _XcfFile(_JpegFile):2090 def _convert(self):2091 # Very few programs other than GIMP read XCF files. This is by design2092 # from the GIMP developers, the format is not really documented or2093 # supported as a general-purpose file format.2094 # Commons uses ImageMagick, thus we have EXACTLY THE SAME support!2095 # (can also be a drawback, e.g. when the library is buggy...)2096 proc = Popen("convert %s %s" % (self.image_path, self.image_path_JPEG),2097 shell=True, stderr=PIPE)#.stderr.read()2098 proc.wait()2099 if proc.returncode != 0:2100 raise ImportError("convert (ImageMagick) not found (may be other error occured)!")2101 elif proc.returncode:2102 self.image_path_JPEG = self.image_path2103 #data = Popen("identify -verbose info: %s" % self.image_path,2104 # shell=True, stderr=PIPE).stderr.read()2105 #print data2106 if not os.path.exists(self.image_path_JPEG):2107 # xcf can have more than 1 layer/page like gif, tiff, and movies...2108 self.image_path_JPEG = self.image_path_JPEG.replace('.jpg', '-0.jpg')2109 self.image_size = Image.open(self.image_path_JPEG).size2110 # MIME: 'image/x-xcf; charset=binary'2111 def _detect_Properties(self):2112 """Retrieve as much file property info possible, especially the same2113 as commons does in order to compare if those libraries (ImageMagick,2114 ...) are buggy (thus explicitely use other software for independence)"""2115 result = { 'Format': u'%s' % self.file_mime[1].upper(),2116 # DO NOT use ImageMagick (identify) instead of PIL to get these info !!2117 'Pages': 0,2118 'Dimensions': self.image_size,2119 'Filesize': os.path.getsize(self.file_name),2120 'MIME': u'%s/%s' % tuple(self.file_mime[:2]), }2121 #self._properties['Properties'] = [result]2122 self._properties['Properties'][0].update(result)2123 return2124class _SvgFile(_JpegFile):2125 def _convert(self):2126 # SVG: rasterize the SVG to bitmap (MAY BE GET FROM WIKI BY DOWNLOAD?...)2127 # (Mediawiki uses librsvg too: http://commons.wikimedia.org/wiki/SVG#SVGs_in_MediaWiki)2128 # http://stackoverflow.com/questions/6589358/convert-svg-to-png-in-python2129 # http://cairographics.org/pythoncairopil/2130 # http://cairographics.org/pyrsvg/2131 # http://stackoverflow.com/questions/9166400/convert-rgba-png-to-rgb-with-pil2132 try:2133 svg = rsvg.Handle(self.image_path)2134 img = cairo.ImageSurface(cairo.FORMAT_ARGB32, svg.props.width, svg.props.height)2135 ctx = cairo.Context(img)2136 svg.render_cairo(ctx)2137 #img.write_to_png("svg.png")2138 #Image.frombuffer("RGBA",( img.get_width(),img.get_height() ),2139 # img.get_data(),"raw","RGBA",0,1).save(self.image_path_JPEG, "JPEG")2140 png = Image.frombuffer("RGBA",( img.get_width(),img.get_height() ),2141 img.get_data(),"raw","RGBA",0,1)2142 background = Image.new("RGB", png.size, (255, 255, 255))2143 background.paste(png, mask=png.split()[3]) # 3 is the alpha channel2144 background.save(self.image_path_JPEG, "JPEG")2145 self.image_size = (svg.props.width, svg.props.height)2146 except MemoryError:2147 self.image_path_JPEG = self.image_path2148 except SystemError:2149 self.image_path_JPEG = self.image_path2150 # MIME: 'application/xml; charset=utf-8'2151 def _detect_Properties(self):2152 """Retrieve as much file property info possible, especially the same2153 as commons does in order to compare if those libraries (ImageMagick,2154 ...) are buggy (thus explicitely use other software for independence)"""2155 result = {'Format': u'-', 'Pages': 0}2156 # similar to PDF page count OR use BeautifulSoup2157 svgcountpages = re.compile("<page>")2158 pc = len(svgcountpages.findall( file(self.image_path,"r").read() ))2159 #svg = rsvg.Handle(self.image_path)2160 # http://validator.w3.org/docs/api.html#libs2161 # http://pypi.python.org/pypi/py_w3c/2162 vld = HTMLValidator()2163 valid = u'SVG'2164 try:2165 vld.validate(self.image.fileUrl())2166 valid = (u'Valid SVG' if vld.result.validity == 'true' else u'Invalid SVG')2167 except urllib2.URLError:2168 pass2169 except ValidationFault:2170 pass2171 #print vld.errors, vld.warnings2172 #self.image_size = (svg.props.width, svg.props.height)2173 result.update({ 'Format': valid,2174 'Mode': u'-',2175 'Palette': u'-',2176 'Pages': pc,2177 # may be set {{validSVG}} also or do something in bot template to2178 # recognize 'Format=SVG (valid)' ...2179 'Dimensions': self.image_size,2180 'Filesize': os.path.getsize(self.file_name),2181 'MIME': u'%s/%s' % tuple(self.file_mime[:2]), })2182 #self._properties['Properties'] = [result]2183 self._properties['Properties'][0].update(result)2184 return2185class _PdfFile(_JpegFile):2186 def getFeatures(self):2187 # optical and other text recognition (tesseract & ocropus, ...)2188 self._detect_EmbeddedText()2189# self._recognize_OpticalText()2190 # (may be just classify as 'contains text', may be store text, e.g. to wikisource)2191 return self._features2192 def _convert(self):2193# self._wikidata = self.image._latestInfo # all info wikimedia got from content (mime, sha1, ...)2194 # PDF: support extract text and images2195 # (Mediawiki uses ghostscript: https://www.mediawiki.org/wiki/Extension:PdfHandler#Pre-requisites)2196 # http://vermeulen.ca/python-pdf.html2197 # http://code.activestate.com/recipes/511465-pure-python-pdf-to-text-converter/2198 # http://stackoverflow.com/questions/25665/python-module-for-converting-pdf-to-text2199 if os.path.splitext(self.image_filename)[1].lower() == u'.pdf':2200 pass2201 # MIME: 'application/pdf; charset=binary'2202 def _detect_Properties(self):2203 """Retrieve as much file property info possible, especially the same2204 as commons does in order to compare if those libraries (ImageMagick,2205 ...) are buggy (thus explicitely use other software for independence)"""2206 # http://code.activestate.com/recipes/496837-count-pdf-pages/2207 #rxcountpages = re.compile(r"$\s*/Type\s*/Page[/\s]", re.MULTILINE|re.DOTALL)2208 rxcountpages = re.compile(r"/Type\s*/Page([^s]|$)", re.MULTILINE|re.DOTALL) # PDF v. 1.3,1.4,1.5,1.62209 pc = len(rxcountpages.findall( file(self.image_path,"rb").read() ))2210 result = { 'Format': u'PDF',2211 'Mode': u'-',2212 'Palette': u'-',2213 'Pages': pc,2214 'Dimensions': self.image_size,2215 'Filesize': os.path.getsize(self.file_name),2216 'MIME': u'%s/%s' % tuple(self.file_mime[:2]), }2217 #self._properties['Properties'] = [result]2218 self._properties['Properties'][0].update(result)2219 return2220 # ./run-test (ocropus/ocropy)2221 # (in fact all scripts/executables used here are pure python scripts!!!)2222 def _recognize_OpticalText(self):2223 # optical text recognition (tesseract & ocropus, ...)2224 # (no full recognition but - at least - just classify as 'contains text')2225 # http://www.claraocr.org/de/ocr/ocr-software/open-source-ocr.html2226 # https://github.com/edsu/ocropy2227 # http://de.wikipedia.org/wiki/Benutzer:DrTrigonBot/Doku#Categorization2228 # Usage:tesseract imagename outputbase [-l lang] [configfile [[+|-]varfile]...]2229 # tesseract imagename.tif output2230 # (it's simpler to run the scripts/executables in own environment/interpreter...)2231 path = os.path.join(scriptdir, 'dtbext/_ocropus/ocropy')2232 curdir = os.path.abspath(os.curdir)2233 os.chdir(path)2234 # binarization2235 if os.path.exists(os.path.join(path, "temp")):2236 shutil.rmtree(os.path.join(path, "temp"))2237 if os.system("ocropus-nlbin %s -o %s" % (self.image_path_JPEG, os.path.join(path, "temp"))):2238 raise ImportError("ocropus not found!")2239 2240 # page level segmentation2241 if os.system("ocropus-gpageseg --minscale 6.0 '%s'" % os.path.join(path, "temp/????.bin.png")):2242 # detection error2243 return2244 2245 # raw text line recognition2246 if os.system("ocropus-lattices --writebestpath '%s'" % os.path.join(path, "temp/????/??????.bin.png")):2247 # detection error2248 return2249 2250 # language model application2251 # (optional - improve the raw results by applying a pretrained model)2252 os.environ['OCROPUS_DATA'] = os.path.join(path, "models/")2253 if os.system("ocropus-ngraphs '%s'" % os.path.join(path, "temp/????/??????.lattice")):2254 # detection error2255 return2256 2257 # create hOCR output2258 if os.system("ocropus-hocr '%s' -o %s" % (os.path.join(path, "temp/????.bin.png"), os.path.join(path, "temp.html"))):2259 # detection error2260 return2261 2262 ## 'create HTML for debugging (use "firefox temp/index.html" to view)'2263 ## (optional - generate human readable debug output)2264 #if os.system("ocropus-visualize-results %s" % os.path.join(path, "temp")):2265 # # detection error2266 # return2267 2268 # "to see recognition results, type: firefox temp.html"2269 # "to see details on the recognition process, type: firefox temp/index.html"2270 tmpfile = open(os.path.join(path, "temp.html"), 'r')2271 data = tmpfile.read()2272 tmpfile.close()2273 shutil.rmtree(os.path.join(path, "temp"))2274 os.remove(os.path.join(path, "temp.html"))2275 os.chdir(curdir)2276 #print data2277 pywikibot.output(data)2278 2279 def _detect_EmbeddedText(self):2280 # may be also: http://www.reportlab.com/software/opensource/rl-toolkit/2281 # poppler pdftotext/pdfimages2282 # (similar as in '_util_get_DataTags_EXIF' but with stderr and no json output)2283 # http://poppler.freedesktop.org/2284 # http://www.izzycode.com/bash/how-to-install-pdf2text-on-centos-fedora-redhat.html2285 # MIGHT BE BETTER TO USE AS PYTHON MODULE:2286 # https://launchpad.net/poppler-python/2287 # http://stackoverflow.com/questions/2732178/extracting-text-from-pdf-with-poppler-c2288 # http://stackoverflow.com/questions/25665/python-module-for-converting-pdf-to-text2289 #proc = Popen("pdftotext -layout %s %s" % (self.image_path, self.image_path+'.txt'), 2290 proc = Popen("pdftotext %s %s" % (self.image_path, self.image_path+'.txt'), 2291 shell=True, stderr=PIPE)#.stderr.readlines()2292 proc.wait()2293 if proc.returncode:2294 raise ImportError("pdftotext not found!")2295 data = open(self.image_path+'.txt', 'r').readlines()2296 os.remove( self.image_path+'.txt' )2297# self._content_text = data2298 (s1, l1) = (len(u''.join(data)), len(data))2299 tmp_path = os.path.join(os.environ.get('TMP', '/tmp'), 'DrTrigonBot/')2300 os.mkdir( tmp_path )2301# switch this part off since 'pdfimages' (on toolserver) is too old; TS-14492302# proc = Popen("pdfimages -p %s %s/" % (self.image_path, tmp_path), 2303 proc = Popen("pdfimages %s %s/" % (self.image_path, tmp_path), 2304 shell=True, stderr=PIPE)#.stderr.readlines()2305 proc.wait()2306 if proc.returncode:2307 raise ImportError("pdfimages not found!")2308 images = os.listdir( tmp_path )2309# pages = set()2310 for f in images:2311# pages.add( int(f.split('-')[1]) )2312 os.remove( os.path.join(tmp_path, f) )2313 os.rmdir( tmp_path )2314 2315 ## pdfminer (tools/pdf2txt.py)2316 ## http://denis.papathanasiou.org/?p=343 (for layout and images)2317 #debug = 02318 #laparams = layout.LAParams()2319 ##2320 #pdfparser.PDFDocument.debug = debug2321 #pdfparser.PDFParser.debug = debug2322 #cmapdb.CMapDB.debug = debug2323 #pdfinterp.PDFResourceManager.debug = debug2324 #pdfinterp.PDFPageInterpreter.debug = debug2325 #pdfdevice.PDFDevice.debug = debug2326 ##2327 #rsrcmgr = pdfinterp.PDFResourceManager(caching=True)2328 #outfp = StringIO.StringIO()2329 #device = converter.TextConverter(rsrcmgr, outfp, codec='utf-8', laparams=laparams)2330 ##device = converter.XMLConverter(rsrcmgr, outfp, codec='utf-8', laparams=laparams, outdir=None)2331 ##device = converter.HTMLConverter(rsrcmgr, outfp, codec='utf-8', scale=1,2332 ## layoutmode='normal', laparams=laparams, outdir=None)2333 ##device = pdfdevice.TagExtractor(rsrcmgr, outfp, codec='utf-8')2334 #fp = file(self.image_path, 'rb')2335 #try:2336 # pdfinterp.process_pdf(rsrcmgr, device, fp, set(), maxpages=0, password='',2337 # caching=True, check_extractable=False)2338 #except AssertionError:2339 # pywikibot.warning(u'pdfminer missed, may be corrupt [_detect_EmbeddedText]')2340 # return2341 #except TypeError:2342 # pywikibot.warning(u'pdfminer missed, may be corrupt [_detect_EmbeddedText]')2343 # return2344 #fp.close()2345 #device.close()2346 #data = outfp.getvalue().splitlines(True)2347 #2348 #(s2, l2) = (len(u''.join(data)), len(data))2349 result = { 'Size': s1,2350 'Lines': l1,2351 #'Data': data,2352 #'Position': pos,2353# 'Images': u'%s (on %s page(s))' % (len(images), len(list(pages))), # pages containing images2354 'Images': u'%s' % len(images),2355 'Type': u'-', } # 'Type' could be u'OCR' above...2356 self._features['Text'] = [result]2357 return2358#class DjvuFile(_JpegFile):2359# pass2360class _OggFile(_JpegFile):2361 def getFeatures(self):2362 # general handling of all audio and video formats2363 self._detect_Streams()2364 # general audio feature extraction2365# self._detect_AudioFeatures()2366 return self._features2367 # MIME: 'application/ogg; charset=binary'2368 def _detect_Properties(self):2369 """Retrieve as much file property info possible, especially the same2370 as commons does in order to compare if those libraries (ImageMagick,2371 ...) are buggy (thus explicitely use other software for independence)"""2372 # 'ffprobe' (ffmpeg); audio and video streams files (ogv, oga, ...)2373 d = self._util_get_DataStreams_FFMPEG()2374 #print d2375 result = { 'Format': u'%s' % d['format']['format_name'].upper(),2376 'Pages': 0,2377 'Dimensions': self.image_size,2378 'Filesize': os.path.getsize(self.file_name),2379 'MIME': u'%s/%s' % tuple(self.file_mime[:2]), }2380 #self._properties['Properties'] = [result]2381 self._properties['Properties'][0].update(result)2382 return2383 def _detect_Streams(self):2384 # audio and video streams files (ogv, oga, ...)2385 d = self._util_get_DataStreams_FFMPEG()2386 if not d:2387 return2388 result = []2389 for s in d['streams']:2390 #print s2391 if (s["codec_type"] == "video"):2392 rate = s["avg_frame_rate"]2393 dim = (int(s["width"]), int(s["height"]))2394 #asp = s["display_aspect_ratio"]2395 elif (s["codec_type"] == "audio"):2396# switch this part off since 'ffprobe' (on toolserver) is too old2397# rate = u'%s/%s/%s' % (s["channels"], s["sample_fmt"], s["sample_rate"])2398 rate = u'%s/%s/%s' % (s["channels"], u'-', int(float(s["sample_rate"])))2399 dim = None2400 elif (s["codec_type"] == "data"):2401 rate = None2402 dim = None2403 result.append({ 'ID': int(s["index"]) + 1,2404 'Format': u'%s/%s' % (s["codec_type"], s.get("codec_name",u'?')),2405 'Rate': rate or u'-',2406 'Dimensions': dim or (None, None),2407 })2408 if 'image' in d["format"]["format_name"]:2409 result = []2410 self._features['Streams'] = result2411 return2412 def _util_get_DataStreams_FFMPEG(self):2413 if hasattr(self, '_buffer_FFMPEG'):2414 return self._buffer_FFMPEG2415 # (similar as in '_util_get_DataTags_EXIF')2416# switch this part off since 'ffprobe' (on toolserver) is too old; TS-14492417# data = Popen("ffprobe -v quiet -print_format json -show_format -show_streams %s" % self.image_path, 2418 proc = Popen("ffprobe -v quiet -show_format -show_streams %s" % self.image_path,#.replace('%', '%%'), 2419 shell=True, stdout=PIPE)#.stdout.read()2420 proc.wait()2421 if proc.returncode == 127:2422 raise ImportError("ffprobe (ffmpeg) not found!")2423 data = proc.stdout.read().strip()2424# self._buffer_FFMPEG = json.loads(data)2425 res, key, cur = {}, '', {}2426 for item in data.splitlines():2427 if (item[0] == '['):2428 if not (item[1] == '/'):2429 key = item[1:-1]2430 cur = {}2431 if key not in res:2432 res[key] = []2433 else:2434 res[key].append( cur )2435 else:2436 val = item.split('=')2437 cur[val[0].strip()] = val[1].strip()2438 if res:2439 res = { 'streams': res['STREAM'], 'format': res['FORMAT'][0] }2440 self._buffer_FFMPEG = res2441 2442 return self._buffer_FFMPEG2443 def _detect_AudioFeatures(self):2444 # http://yaafe.sourceforge.net/manual/tools.html2445 # http://yaafe.sourceforge.net/manual/quickstart.html - yaafe.py2446 # ( help: yaafe.py -h / features: yaafe.py -l )2447 #2448 # compile yaafe on fedora:2449 # 1.) get and compile 'argtable2' (2-13)2450 # 1.1 download from http://argtable.sourceforge.net/2451 # 1.2 unpack and cd to directory2452 # 1.3 $ ccmake .2453 # 1.4 set: CMAKE_BUILD_TYPE = Release2454 # 1.5 press: c, g (in order to configure and generate)2455 # 1.6 $ make2456 # 2.) get and compile 'yaafe'2457 # 1.1 download from http://yaafe.sourceforge.net/2458 # 1.2 unpack and cd to directory2459 # 1.3 $ ccmake .2460 # 1.4 set: ARGTABLE2_INCLUDE_DIR = /home/ursin/Desktop/argtable2-13/src2461 # ARGTABLE2_LIBRARY = /home/ursin/Desktop/argtable2-13/src/libargtable2.a2462 # ...2463 # DL_INCLUDE_DIR = /usr/include2464 # DL_LIBRARY = /usr/lib64/libdl.so2465 # FFTW3_INCLUDE_DIR = /usr/include2466 # FFTW3_LIBRARY = /usr/lib64/libfftw3.so2467 # HDF5_HL_LIBRARY = /usr/lib64/libhdf5_hl.so2468 # HDF5_INCLUDE_DIR = /usr/include2469 # HDF5_LIBRARY = /usr/lib64/libhdf5.so2470 # LAPACK_LIBRARY = /usr/lib64/liblapack.so2471 # MATLAB_ROOT = MATLAB_ROOT-NOTFOUND2472 # MPG123_INCLUDE_DIR = /usr/include2473 # MPG123_LIBRARY = /usr/lib64/libmpg123.so2474 # RT_LIBRARY = /usr/lib64/librt.so2475 # SNDFILE_INCLUDE_DIR = /usr/include2476 # SNDFILE_LIBRARY = /usr/lib64/libsndfile.so2477 # ...2478 # WITH_FFTW3 = ON2479 # WITH_HDF5 = ON2480 # WITH_LAPACK = ON2481 # WITH_MATLAB_MEX = OFF2482 # WITH_MPG123 = ON2483 # WITH_SNDFILE = ON2484 # WITH_TIMERS = ON2485 # (use t to toggle to more advanced options)2486 # CMAKE_CXX_FLAGS = -fpermissive2487 # CMAKE_C_FLAGS = -fpermissive2488 # (install all needed dependencies/packages into the OS also)2489 # 1.5 press: c, g (in order to configure and generate)2490 # 1.6 $ make2491 # 1.7 $ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/ursin/Desktop/yaafe-v0.64/src_cpp/yaafe-python/:/home/ursin/Desktop/yaafe-v0.64/src_cpp/yaafe-io/:/home/ursin/Desktop/yaafe-v0.64/src_cpp/yaafe-core/:/home/ursin/Desktop/yaafe-v0.64/src_cpp/yaafe-components/2492 # $ export YAAFE_PATH=/home/ursin/Desktop/yaafe-v0.64/src_python/2493 # $ export PYTHONPATH=/home/ursin/Desktop/yaafe-v0.64/src_python2494 import yaafelib as yaafe2495 # use WAV, OGG, MP3 (and others) audio file formats2496 #audiofile = '/home/ursin/data/09Audio_UNS/Amy MacDonald - This Is The Life (2007) - Pop/01-amy_macdonald-mr_rock_and_roll.mp3'2497 audiofile = self.image_path2498 yaafe.setVerbose(True)2499 #print 'Yaafe v%s'%yaafe.getYaafeVersion()2500 # Load important components2501 if (yaafe.loadComponentLibrary('yaafe-io')!=0):2502 pywikibot.warning(u'cannot load yaafe-io component library !') # ! needed, else it will crash !2503 # Build a DataFlow object using FeaturePlan2504 fp = yaafe.FeaturePlan(sample_rate=44100, normalize=0.98, resample=False)2505 #fp.addFeature('am: AmplitudeModulation blockSize=512 stepSize=256')2506 #fp.addFeature('ac: AutoCorrelation blockSize=512 stepSize=256')2507 #fp.addFeature('cdod: ComplexDomainOnsetDetection blockSize=512 stepSize=256')2508 #fp.addFeature('erg: Energy blockSize=512 stepSize=256')2509 #fp.addFeature('e: Envelope blockSize=512 stepSize=256')2510 fp.addFeature('ess: EnvelopeShapeStatistics blockSize=512 stepSize=256')2511 #fp.addFeature('f: Frames blockSize=512 stepSize=256')2512 #fp.addFeature('lpc: LPC blockSize=512 stepSize=256')2513 #fp.addFeature('lsf: LSF blockSize=512 stepSize=256')2514 #fp.addFeature('l: Loudness blockSize=512 stepSize=256')2515 #fp.addFeature('mfcc: MFCC blockSize=512 stepSize=256')2516 ## features: AutoCorrelationPeaksIntegrator, Cepstrum, Derivate, HistogramIntegrator, SlopeIntegrator, StatisticalIntegrator2517 #fp.addFeature('mfcc_d1: MFCC blockSize=512 stepSize=256 > Derivate DOrder=1')2518 #fp.addFeature('mfcc_d2: MFCC blockSize=512 stepSize=256 > Derivate DOrder=2')2519 #fp.addFeature('mas: MagnitudeSpectrum blockSize=512 stepSize=256')2520 #fp.addFeature('mes: MelSpectrum blockSize=512 stepSize=256')2521 #fp.addFeature('obsi: OBSI blockSize=512 stepSize=256')2522 #fp.addFeature('obsir: OBSIR blockSize=512 stepSize=256')2523 #fp.addFeature('psh: PerceptualSharpness blockSize=512 stepSize=256')2524 #fp.addFeature('psp: PerceptualSpread blockSize=512 stepSize=256')2525 #fp.addFeature('scfpb: SpectralCrestFactorPerBand blockSize=512 stepSize=256')2526 #fp.addFeature('sd: SpectralDecrease blockSize=512 stepSize=256')2527 #fp.addFeature('sfa: SpectralFlatness blockSize=512 stepSize=256')2528 #fp.addFeature('sfpb: SpectralFlatnessPerBand blockSize=512 stepSize=256')2529 #fp.addFeature('sfu: SpectralFlux blockSize=512 stepSize=256')2530 #fp.addFeature('sr: SpectralRolloff blockSize=512 stepSize=256')2531 fp.addFeature('sss: SpectralShapeStatistics blockSize=512 stepSize=256')2532 #fp.addFeature('ss: SpectralSlope blockSize=512 stepSize=256')2533 #fp.addFeature('sv: SpectralVariation blockSize=512 stepSize=256')2534 fp.addFeature('tss: TemporalShapeStatistics blockSize=512 stepSize=256')2535 fp.addFeature('zcr: ZCR blockSize=512 stepSize=256')2536 df = fp.getDataFlow()2537 ## or load a DataFlow from dataflow file.2538 #df = DataFlow()2539 #df.load(dataflow_file)2540 #fp.getDataFlow().save('')2541 #print df.display()2542 # configure an Engine2543 engine = yaafe.Engine()2544 engine.load(df)2545 # extract features from an audio file using AudioFileProcessor2546 afp = yaafe.AudioFileProcessor()2547 #afp.setOutputFormat('csv','',{}) # ! needed, else it will crash ! (but now produces file output)2548 #afp.processFile(engine,audiofile)2549 #feats = engine.readAllOutputs()2550 ## and play with your features2551 #print feats2552 # extract features from an audio file and write results to csv files2553 afp.setOutputFormat('csv','output',{'Precision':'8'})2554 afp.processFile(engine,audiofile)2555 # this creates output/myaudio.wav.mfcc.csv, .mfcc_d1.csv and .mfcc_d2.csv files.2556 ## extract features from a numpy array2557 #audio = np.random.randn(1,100000)2558 #feats = engine.processAudio(audio)2559 ## and play with your features2560 #print feats2561 import csv2562 data = {}2563 for ext in ['ess', 'sss', 'tss', 'zcr']:2564 fn = 'output' + audiofile + ('.%s.csv' % ext)2565 with open(fn, 'rb') as csvfile:2566 reader = csv.reader(csvfile, delimiter=',')2567 d = [row for row in reader]2568 d = np.array(d[5:]) # cut header and convert to numpy2569 d = np.float_(d)2570 d = tuple(np.average(d, axis=0))2571 pywikibot.output(ext)2572 #if ext in ['ess', 'sss', 'tss']:2573 # pywikibot.output(u"centroid: %s\nspread: %s\nskewness: %s\nkurtosis: %s\n" % d)2574 #elif ext in ['zcr']:2575 # pywikibot.output(u"zero-crossing rate: %s\n" % d)2576 data[ext.upper()] = d2577 os.remove(fn)2578 # remove folder too...2579 self._features['Audio'] = [data]2580 return2581class _MidiFile(_UnknownFile):2582 def getFeatures(self):2583 self._detect_AudioFeatures() # Audio2584 return self._features2585 def _detect_HeaderAndMetadata(self):2586 #_UnknownFile._detect_HeaderAndMetadata(self)2587 #result = {'Desc': self._properties['Metadata'][0]['Desc'].splitlines()}2588 result = {'Desc': []}2589 # extract data from midi file2590 # http://valentin.dasdeck.com/midi/midifile.htm2591 # http://stackoverflow.com/questions/3943149/reading-and-interpreting-data-from-a-binary-file-in-python2592 ba = bytearray(open(self.file_name, 'rb').read())2593 i = -12594 for key, data in [('Text', '\x01'), ('Copyright', '\x02')]:#, ('Lyrics', '\x05')]:2595 key = 'Desc'2596 #result[key] = []2597 while True:2598 i = ba.find('\xff%s' % data, i+1)2599 if i < 0: # something found?2600 break2601 e = (i+3+ba[i+2])2602 if ba[e] != 0: # length match with string end (00)?2603 e = ba.find('\x00', (i+3+ba[i+2]))2604 result[key].append(ba[i+3:e].decode('latin-1').strip())2605 #result[key] = u'\n'.join(result[key])2606 result[key] = u'\n'.join(result[key])2607 if not result['Desc']:2608 result['Desc'] = u'-'2609 ## find specific info in extracted data2610 #print [item.strip() for item in re.findall('Generated .*?\n', result['Text'])]2611 ##u"Cr'eateur: GNU LilyPond 2.0.1"2612 #import dateutil.parser2613 #dates = []2614 #for line in result['Text'].splitlines():2615 # # http://stackoverflow.com/questions/3276180/extracting-date-from-a-string-in-python2616 # try:2617 # dates.append(dateutil.parser.parse(line, fuzzy=True).isoformat(' ').decode('utf-8'))2618 # except ValueError:2619 # pass2620 #print dates2621 import _music21 as music212622 try:2623 s = music21.converter.parse(self.file_name)2624 if s.metadata:2625 pywikibot.output(unicode(s.metadata))2626 result.update(s.metadata)2627 except music21.midi.base.MidiException:2628 pass2629 self._properties['Metadata'] = [result]2630 return2631 # MIME: 'audio/midi; charset=binary'2632 def _detect_Properties(self):2633 """Retrieve as much file property info possible, especially the same2634 as commons does in order to compare if those libraries (ImageMagick,2635 ...) are buggy (thus explicitely use other software for independence)"""2636 result = { 'Format': u'%s' % self.file_mime[1].upper(),2637 'Pages': 0,2638 'Dimensions': self.image_size,2639 'Filesize': os.path.getsize(self.file_name),2640 'MIME': u'%s/%s' % tuple(self.file_mime[:2]), }2641 #self._properties['Properties'] = [result]2642 self._properties['Properties'][0].update(result)2643 return2644 # midi audio feature extraction2645 def _detect_AudioFeatures(self):2646 import _music21 as music212647 #music21.features.jSymbolic.getCompletionStats()2648 try:2649 #audiofile = '/home/ursin/Desktop/3_Ships.mid'2650 #s = music21.midi.translate.midiFilePathToStream(self.file_name)2651 s = music21.converter.parse(self.file_name)2652 except music21.midi.base.MidiException:2653 pywikibot.warning(u'unknown file type [_detect_AudioFeatures]')2654 return2655 #fs = music21.features.jSymbolic.extractorsById2656 #for k in fs:2657 # for i in range(len(fs[k])):2658 # if fs[k][i] is not None:2659 # n = fs[k][i].__name__2660 # if fs[k][i] not in music21.features.jSymbolic.featureExtractors:2661 # n += " (not implemented)"2662 # print k, i, n2663 # else:2664 # fe = fs[k][i](s)2665 # print k, i, n,2666 # try:2667 # f = fe.extract()2668 # print f.name, f.vector2669 # except AttributeError:2670 # print "ERROR"2671 data = {'RegisterImportance': (music21.features.jSymbolic.ImportanceOfBassRegisterFeature(s).extract().vector[0],2672 music21.features.jSymbolic.ImportanceOfMiddleRegisterFeature(s).extract().vector[0],2673 music21.features.jSymbolic.ImportanceOfHighRegisterFeature(s).extract().vector[0],),2674 'NoteDuration': (music21.features.jSymbolic.AverageNoteDurationFeature(s).extract().vector[0],2675 music21.features.jSymbolic.MaximumNoteDurationFeature(s).extract().vector[0],),2676 'IndependentVoices': (music21.features.jSymbolic.AverageNumberOfIndependentVoicesFeature(s).extract().vector[0],2677 music21.features.jSymbolic.MaximumNumberOfIndependentVoicesFeature(s).extract().vector[0],),2678 'MostCommonPitch': music21.features.jSymbolic.MostCommonPitchFeature(s).extract().vector[0],2679 'Tempo': music21.features.jSymbolic.InitialTempoFeature(s).extract().vector[0],2680 'Duration': s.highestTime,2681 #'Metadata': s.metadata if s.metadata else u'',2682 'Lyrics': s.lyrics(recurse=True) if s.lyrics(recurse=True) else u'',}2683 #print music21.text.assembleLyrics(s)2684 #print s.duration2685 #print s.offsetMap2686 #print s.measureOffsetMap()2687 #print s.seconds2688 #print s.secondsMap2689 self._features['Audio'] = [data]2690 return2691# http://commons.wikimedia.org/wiki/File_formats2692_FILETYPES = { '*': _UnknownFile,2693 ( 'image', 'jpeg'): _JpegFile,2694 ( 'image', 'png'): _PngFile,2695 ( 'image', 'gif'): _GifFile,2696 ( 'image', 'tiff'): _TiffFile,2697 ( 'image', 'x-xcf'): _XcfFile,2698 ( 'image', 'svg+xml'): _SvgFile, # unify/merge them?2699 ('application', 'xml'): _SvgFile, #2700 ('application', 'pdf'): _PdfFile,2701# djvu: python-djvulibre or python-djvu for djvu support2702# http://pypi.python.org/pypi/python-djvulibre/0.3.92703# ( 'image', 'vnd.djvu'): DjvuFile,2704 ( 'audio', 'midi'): _MidiFile,2705 ('application', 'ogg'): _OggFile,}2706# ( '?', '?'): _WebMFile,}2707def GenericFile(file_name):2708 # 'magic' (libmagic)2709 m = magic.open(magic.MAGIC_MIME) # or 'magic.MAGIC_NONE'2710 m.load()2711 file_mime = re.split('[/;\s]', m.file(file_name))2712 mime = mimetypes.guess_all_extensions('%s/%s' % tuple(file_mime[0:2]))2713 if mime and (os.path.splitext(file_name)[1].lower() not in mime):2714 pywikibot.warning(u'File extension does not match MIME type! File extension should be %s.' % mime)2715 # split detection and extraction according to file types; _JpegFile, ...2716 GenericFile = _FILETYPES.get(tuple(file_mime[:2]), _FILETYPES['*'])2717 return GenericFile(file_name, file_mime)2718# all classification and categorization methods and definitions - default variation2719# use simplest classification I can think of (self-made) and do categorization2720# mostly based on filtered/reported features2721class CatImages_Default(object):2722 #ignore = []2723 ignore = ['color']2724 2725 _thrhld_group_size = 42726 #_thrshld_guesses = 0.12727 _thrshld_default = 0.752728 # for '_detect_Trained'2729 cascade_files = [(u'Legs', 'haarcascade_lowerbody.xml'),2730 (u'Torsos', 'haarcascade_upperbody.xml'),2731 (u'Ears', 'haarcascade_mcs_leftear.xml'),2732 (u'Ears', 'haarcascade_mcs_rightear.xml'),2733 (u'Eyes', 'haarcascade_lefteye_2splits.xml'), # (http://yushiqi.cn/research/eyedetection)2734 (u'Eyes', 'haarcascade_righteye_2splits.xml'), # (http://yushiqi.cn/research/eyedetection)2735 #externals/opencv/haarcascades/haarcascade_mcs_lefteye.xml2736 #externals/opencv/haarcascades/haarcascade_mcs_righteye.xml2737 # (others include indifferent (left and/or right) and pair)2738 (u'Automobiles', 'cars3.xml'), # http://www.youtube.com/watch?v=c4LobbqeKZc2739 (u'Hands', '1256617233-2-haarcascade-hand.xml', 300.),] # http://www.andol.info/2740 # ('Hands' does not behave very well, in fact it detects any kind of skin and other things...)2741 #(u'Aeroplanes', 'haarcascade_aeroplane.xml'),] # e.g. for 'Category:Unidentified aircraft'2742 # very simple / rought / poor-man's min. thresshold classification2743 # (done by guessing, does not need to be trained)2744 # replace/improve this with RTrees, KNearest, Boost, SVM, MLP, NBayes, ...2745 def classifyFeatures(self):2746 # classification of detected features (should use RTrees, KNearest, Boost, SVM, MLP, NBayes, ...)2747 # ??? (may be do this in '_cat_...()' or '_filter_...()' ?!?...)2748 # Faces and eyes (opencv pre-trained haar and extracted EXIF data)2749 for i in range(len(self._info['Faces'])):2750 if self._info['Faces'][i]['Type'] == u'Exif':2751 c = self._thrshld_default2752 else:2753 c = (len(self._info['Faces'][i]['Eyes']) + 2.) / 4.2754 self._info['Faces'][i]['Confidence'] = c2755 self._info['Faces'][i]['ID'] = i+12756 # Segments and colors / Average color2757 #max_dim = max(self.image_size)2758 for i in range(len(self._info['ColorRegions'])):2759 data = self._info['ColorRegions'][i]2760 # has to be in descending order since only 1 resolves (!)2761 #if (data['Coverage'] >= 0.40) and (data['Delta_E'] <= 5.0):2762 # c = 1.02763 ##elif (data['Coverage'] >= 0.20) and (data['Delta_E'] <= 15.0):2764 ##elif (data['Coverage'] >= 0.20) and (data['Delta_E'] <= 10.0):2765 #elif (data['Coverage'] >= 0.25) and (data['Delta_E'] <= 10.0):2766 # c = 0.752767 #elif (data['Coverage'] >= 0.10) and (data['Delta_E'] <= 20.0):2768 # c = 0.52769 #else:2770 # c = 0.12771 ca = (data['Coverage'])**(1./7) # 0.15 -> ~0.752772 #ca = (data['Coverage'])**(1./6) # 0.20 -> ~0.752773 #ca = (data['Coverage'])**(1./5) # 0.25 -> ~0.752774 #ca = (data['Coverage'])**(1./4) # 0.35 -> ~0.752775 ##cb = (0.02 * (50. - data['Delta_E']))**(1.2) # 10.0 -> ~0.752776 #cb = (0.02 * (50. - data['Delta_E']))**(1./2) # 20.0 -> ~0.752777 ##cb = (0.02 * (50. - data['Delta_E']))**(1./3) # 25.0 -> ~0.752778 #cc = (1. - (data['Delta_R']/max_dim))**(1.) # 0.25 -> ~0.752779 #c = ( 3*ca + cb ) / 42780 #c = ( cc + 6*ca + 2*cb ) / 92781 c = ca2782 self._info['ColorRegions'][i]['Confidence'] = c2783 # People/Pedestrian (opencv pre-trained hog and haarcascade)2784 for i in range(len(self._info['People'])):2785 data = self._info['People'][i]2786 if (data['Coverage'] >= 0.20):2787 c = 0.752788 if (data['Coverage'] >= 0.10): # at least 10% coverage needed2789 c = 0.52790 else:2791 c = 0.12792 self._info['People'][i]['Confidence'] = c2793 # general (opencv pre-trained, third-party and self-trained haar2794 # and cascade) classification2795 for cf in self.cascade_files:2796 cat = cf[0]2797 for i in range(len(self._info[cat])):2798 data = self._info[cat][i]2799 # detect images with this as one of the main contents only thus2800 # high coverage requested as a minimal confidence estimation2801 self._info[cat][i]['Confidence'] = (data['Coverage'])**(1./5) # 0.25 -> ~0.752802 # barcode and Data Matrix recognition (libdmtx/pydmtx, zbar, gocr?)2803 for i in range(len(self._info['OpticalCodes'])):2804 self._info['OpticalCodes'][i]['Confidence'] = min(0.75*self._info['OpticalCodes'][i]['Quality']/10., 1.)2805 # Chessboard (opencv reference detector)2806 for i in range(len(self._info['Chessboard'])):2807 self._info['Chessboard'][i]['Confidence'] = len(self._info['Chessboard'][i]['Corners'])/49.2808 ## Geometric object (opencv hough line, circle, edges, corner, ...)2809 #if self._info['Geometry']:2810 # self._info['Geometry'][0]['Confidence'] = 1. - self._info['Geometry'][0]['Edge_Ratio']2811 # Category:Unidentified people2812 def _cat_people_People(self):2813 #relevance = bool(self._info_filter['People'])2814 relevance = self._cat_people_Groups()[1]2815 return (u'Unidentified people', relevance)2816 # Category:Unidentified people2817 #def _cat_multi_People(self):2818 def _cat_face_People(self):2819 relevance = bool(self._info_filter['Faces'])2820 #relevance = bool(self._info_filter['People']) or relevance2821 return (u'Unidentified people', relevance)2822 # Category:Groups2823 def _cat_people_Groups(self):2824 result = self._info_filter['People']2825 relevance = (len(result) >= self._thrhld_group_size) and \2826 (not self._cat_coloraverage_Graphics()[1])2827 return (u'Groups', relevance)2828 # Category:Groups2829 def _cat_face_Groups(self):2830 result = self._info_filter['Faces']2831 #if not (len(result) > 1): # 5 should give 0.75 and get reported2832 # relevance = 0.2833 #else:2834 # relevance = 1 - 1./(len(result)-1)2835 relevance = (len(result) >= self._thrhld_group_size)2836 return (u'Groups', relevance)2837 # Category:Faces2838 def _cat_face_Faces(self):2839 result = self._info_filter['Faces']2840 #return (u'Faces', ((len(result) == 1) and (result[0]['Coverage'] >= .50)))2841 return (u'Faces', ((len(result) == 1) and (result[0]['Coverage'] >= .40)))2842 # Category:Portraits2843 def _cat_face_Portraits(self):2844 result = self._info_filter['Faces']2845 #return (u'Portraits', ((len(result) == 1) and (result[0]['Coverage'] >= .25)))2846 return (u'Portraits', ((len(result) == 1) and (result[0]['Coverage'] >= .20)))2847 # Category:Barcode2848 def _cat_code_Barcode(self):2849 relevance = bool(self._info_filter['OpticalCodes'])2850 return (u'Barcode', relevance)2851 # Category:Chessboards2852 def _cat_chess_Chessboards(self):2853 relevance = bool(self._info_filter['Chessboard'])2854 return (u'Chessboards', relevance)2855 # Category:Books (literature) in PDF2856 def _cat_text_BooksPDF(self):2857 pdf = u'PDF' in self._info_filter['Properties'][0]['Format']2858 result = self._info_filter['Text']2859 relevance = pdf and len(result) and \2860 (self._info_filter['Properties'][0]['Pages'] >= 10) and \2861 (result[0]['Size'] >= 5E4) and (result[0]['Lines'] >= 1000)2862 return (u'Books (literature) in PDF', relevance)2863 # Category:Animated GIF2864 # Category:Animated PNG‎2865 # (Category:Animated SVG‎)2866 def _cat_prop_Animated_general(self):2867 result = self._info_filter['Properties']2868 relevance = result and (result[0]['Pages'] > 1) and \2869 (result[0]['Format'] in [u'GIF', u'PNG'])2870 return (u'Animated %s' % result[0]['Format'], relevance)2871 # Category:Human ears2872 def _cat_ears_HumanEars(self):2873 relevance = bool(self._info_filter['Ears'])2874 return (u'Human ears', relevance)2875 # Category:Human eyes2876 def _cat_eyes_HumanEyes(self):2877 relevance = bool(self._info_filter['Eyes'])2878 return (u'Human eyes', relevance)2879 # Category:Ogg sound files2880 def _cat_streams_OggSoundFiles(self):2881 result = self._info_filter['Streams']2882 return (u'Ogg sound files', ((len(result) == 1) and (u'audio/' in result[0]['Format'])))2883 # Category:Videos2884 def _cat_streams_Videos(self):2885 result = self._info_filter['Streams']2886 return (u'Videos', (True in [u'video/' in s['Format'] for s in result]))2887 # Category:Graphics2888 def _cat_coloraverage_Graphics(self):2889 result = self._info_filter['ColorAverage']2890 relevance = (result and result[0]['Gradient'] < 0.1) and \2891 (0.005 < result[0]['Peaks'] < 0.1) # black/white texts are below that2892 #(result[0]['FFT_Peaks'] < 0.2) # has to be tested first !!!2893 return (u'Graphics', bool(relevance))2894 # Category:MIDI files created with GNU LilyPond2895 def _cat_meta_MIDIfilescreatedwithGNULilyPond(self):2896 result = self._info_filter['Metadata']2897 relevance = len(result) and ('Desc' in result[0]) and \2898 (u"Generated automatically by: GNU LilyPond" in2899 result[0]['Desc'])2900 return (u'MIDI files created with GNU LilyPond', bool(relevance))2901 # Category:Bitmap_from_Inkscape (png)2902 def _cat_meta_BitmapfromInkscape(self):2903 result = self._info_filter['Metadata']2904 relevance = len(result) and ('Software' in result[0]) and \2905 (u"www.inkscape.org" in2906 result[0]['Software'].lower())2907 return (u'Bitmap from Inkscape', bool(relevance))2908 # Category:Created_with_Inkscape (svg)2909 def _cat_meta_CreatedwithInkscape(self):2910 result = self._info_filter['Metadata']2911 relevance = len(result) and ('Output_Extension' in result[0]) and \2912 (u"org.inkscape.output.svg.inkscape" in2913 result[0]['Output_Extension'].lower())2914 return (u'Created with Inkscape', bool(relevance))2915 # Category:Created_with_MATLAB (png)2916 # Category:Created_with_MATLAB (svg)2917 def _cat_meta_CreatedwithMATLAB(self):2918 result = self._info_filter['Metadata']2919 relevance = len(result) and \2920 ((('Software' in result[0]) and \2921 (u"MATLAB, The Mathworks, Inc." in 2922 result[0]['Software'])) \2923 or \2924 (('Desc' in result[0]) and \2925 (u"Matlab Figure" in 2926 result[0]['Desc'])) )2927 return (u'Created with MATLAB', bool(relevance))2928 # Category:Created_with_PLOT2SVG (svg) [new]2929 def _cat_meta_CreatedwithPLOT2SVG(self):2930 result = self._info_filter['Metadata']2931 relevance = len(result) and ('Desc' in result[0]) and \2932 (u"Converted by PLOT2SVG" in2933 result[0]['Desc'])2934 return (u'Created with PLOT2SVG', bool(relevance))2935 # Category:Created_with_ImageMagick (jpg)2936 def _cat_meta_CreatedwithImageMagick(self):2937 result = self._info_filter['Metadata']2938 relevance = len(result) and ('Software' in result[0]) and \2939 (u"ImageMagick" in2940 result[0]['Software'])2941 return (u'Created with ImageMagick', bool(relevance))2942 # Category:Created_with_Adobe_ImageReady (png)2943 def _cat_meta_CreatedwithAdobeImageReady(self):2944 result = self._info_filter['Metadata']2945 relevance = len(result) and ('Software' in result[0]) and \2946 (u"Adobe ImageReady" in2947 result[0]['Software'])2948 return (u'Created with Adobe ImageReady', bool(relevance))2949 # Category:Created_with_Adobe_Photoshop (jpg)2950 def _cat_meta_CreatedwithAdobePhotoshop(self):2951 result = self._info_filter['Metadata']2952 relevance = len(result) and ('Software' in result[0]) and \2953 (u"Adobe Photoshop" in2954 result[0]['Software'])2955 return (u'Created with Adobe Photoshop', bool(relevance))2956 # Category:Created_with_Picasa (jpg)2957 def _cat_meta_CreatedwithPicasa(self):2958 result = self._info_filter['Metadata']2959 relevance = len(result) and ('Software' in result[0]) and \2960 (u"Picasa" in2961 result[0]['Software'])2962 return (u'Created with Picasa', bool(relevance))2963 # Category:Created_with_Qtpfsgui (jpg)2964 def _cat_meta_CreatedwithQtpfsgui(self):2965 result = self._info_filter['Metadata']2966 relevance = len(result) and ('Software' in result[0]) and \2967 (u"Created with opensource tool Qtpfsgui" in2968 result[0]['Software'])2969 return (u'Created with Qtpfsgui', bool(relevance))2970 # Category:Created_with_Autopano (jpg)2971 def _cat_meta_CreatedwithAutopano(self):2972 result = self._info_filter['Metadata']2973 relevance = len(result) and ('Software' in result[0]) and \2974 (u"Autopano" in2975 result[0]['Software'])2976 return (u'Created with Autopano', bool(relevance))2977 # Category:Created_with_Xmgrace (png)2978 def _cat_meta_CreatedwithXmgrace(self):2979 result = self._info_filter['Metadata']2980 relevance = len(result) and ('Software' in result[0]) and \2981 (u"Grace" in2982 result[0]['Software'])2983 return (u'Created with Xmgrace', bool(relevance))2984 # Category:Created_with_darktable (jpg)2985 def _cat_meta_Createdwithdarktable(self):2986 result = self._info_filter['Metadata']2987 relevance = len(result) and ('Software' in result[0]) and \2988 (u"darktable" in2989 result[0]['Software'].lower())2990 return (u'Created with darktable', bool(relevance))2991 # Category:Created_with_easyHDR (jpg)2992 def _cat_meta_CreatedwitheasyHDR(self):2993 result = self._info_filter['Metadata']2994 relevance = len(result) and \2995 ((('Software' in result[0]) and \2996 (u"easyHDR" in2997 result[0]['Software'])) \2998 or \2999 (('Comment' in result[0]) and \3000 (u"easyHDR" in3001 result[0]['Comment'])) )3002 return (u'Created with easyHDR', bool(relevance))3003 # Category:Created_with_GIMP (jpg) [new]3004 def _cat_meta_CreatedwithGIMP(self):3005 result = self._info_filter['Metadata']3006 relevance = len(result) and \3007 ((('Software' in result[0]) and \3008 (u"GIMP" in3009 result[0]['Software'])) \3010 or \3011 (('Comment' in result[0]) and \3012 (u"Created with GIMP" in3013 result[0]['Comment'])) )3014 return (u'Created with GIMP', bool(relevance))3015 # Category:Created_with_R (svg)3016 def _cat_meta_CreatedwithR(self):3017 result = self._info_filter['Metadata']3018 relevance = len(result) and ('Desc' in result[0]) and \3019 (u"R SVG" in3020 result[0]['Desc'])3021 return (u'Created with R', bool(relevance))3022 # Category:Created_with_VectorFieldPlot (svg)3023 def _cat_meta_CreatedwithVectorFieldPlot(self):3024 result = self._info_filter['Metadata']3025 relevance = len(result) and ('Desc' in result[0]) and \3026 (u"created with VectorFieldPlot" in3027 result[0]['Desc'])3028 return (u'Created with VectorFieldPlot', bool(relevance))3029 # Category:Created_with_Chemtool (svg)3030 def _cat_meta_CreatedwithChemtool(self):3031 result = self._info_filter['Metadata']3032 relevance = len(result) and ('Desc' in result[0]) and \3033 (u"Created with Chemtool" in3034 result[0]['Desc'])3035 return (u'Created with Chemtool', bool(relevance))3036 # Category:Created_with_GNU_Octave (svg)3037 def _cat_meta_CreatedwithGNUOctave(self):3038 result = self._info_filter['Metadata']3039 relevance = len(result) and ('Desc' in result[0]) and \3040 (u"Produced by GNUPLOT" in3041 result[0]['Desc'])3042 return (u'Created with GNU Octave', bool(relevance))3043 # Category:Created_with_GeoGebra (svg)3044 def _cat_meta_CreatedwithGeoGebra(self):3045 result = self._info_filter['Metadata']3046 relevance = len(result) and ('DescProducer' in result[0]) and \3047 (u"geogebra.d.W" in3048 result[0]['DescProducer']) #and \3049 #(u"FreeHEP Graphics2D Driver" in3050 # result[0]['DescCreator'])3051 return (u'Created with GeoGebra', bool(relevance))3052 # Category:Created_with_Stella (png)3053 def _cat_meta_CreatedwithStella(self):3054 result = self._info_filter['Metadata']3055 relevance = len(result) and ('Comment' in result[0]) and \3056 (u"Created using Stella4D" in3057 result[0]['Comment'])3058 return (u'Created with Stella', bool(relevance))3059 # Category:Created_with_PhotoStitch (jpg)3060 def _cat_meta_CreatedwithPhotoStitch(self):3061 result = self._info_filter['Metadata']3062 relevance = len(result) and ('Comment' in result[0]) and \3063 (u"LEAD Technologies Inc." in3064 result[0]['Comment'])3065 return (u'Created with PhotoStitch', bool(relevance))3066 # Category:Created_with_Scribus (pdf)3067 def _cat_meta_CreatedwithScribus(self):3068 result = self._info_filter['Metadata']3069 relevance = len(result) and ('Producer' in result[0]) and \3070 (u"Scribus PDF Library" in3071 result[0]['Producer'])3072 return (u'Created with Scribus', bool(relevance))3073 # Category:Created_with_OpenOffice.org (pdf)3074 def _cat_meta_CreatedwithOpenOfficeorg(self):3075 result = self._info_filter['Metadata']3076 relevance = len(result) and ('Producer' in result[0]) and \3077 (u"OpenOffice.org" in3078 result[0]['Producer'])3079 return (u'Created with OpenOffice.org', bool(relevance))3080 # Category:Created_with_Tux_Paint (pdf)3081 def _cat_meta_CreatedwithTuxPaint(self):3082 result = self._info_filter['Metadata']3083 relevance = len(result) and ('Software' in result[0]) and \3084 (u"Tux Paint" in3085 result[0]['Software'])3086 return (u'Created with Tux Paint', bool(relevance))3087 # Category:Created_with_Microsoft_Image_Composite_Editor (jpg)3088 def _cat_meta_CreatedwithMicrosoftImageCompositeEditor(self):3089 result = self._info_filter['Metadata']3090 relevance = len(result) and ('Software' in result[0]) and \3091 (u"Microsoft ICE" in3092 result[0]['Software'])3093 return (u'Created with Microsoft Image Composite Editor', bool(relevance))3094# TODO: make '_cat_meta_general(self)'3095 # Category:Categorized by DrTrigonBot3096 def _addcat_BOT(self):3097 # - ALWAYS -3098 return (u"Categorized by DrTrigonBot", True)3099 # (Category:BMP)3100 # (Category:PNG)3101 # (Category:JPEG)3102 # Category:TIFF files3103 # (may be more image formats/extensions according to PIL, e.g. SVG, ...)3104 # Category:PDF files3105 def _addcat_prop_general(self):3106 fmt = self._info_filter['Properties'][0]['Format']3107 if u'TIFF' in fmt:3108 fmt = u'TIFF images'3109 #elif u'SVG' in fmt:3110 # # additional to PIL (rsvg, ...)3111 # # should be added as template instead of category (!)3112 # fmt = u''3113 elif u'PDF' in fmt:3114 # additional to PIL (...)3115 fmt = u'PDF files'3116 else:3117 # disable ALL categorization, except the listed exceptions above3118 # (BMP, PNG, JPEG, OGG; no general catgeory available, ...)3119 fmt = u''3120 # PIL: http://www.pythonware.com/library/pil/handbook/index.htm3121 return (fmt, bool(fmt))3122# # TODO: add templates (conditional/additional like 'addcat')3123# # Category:SVG - Category:Valid SVG‎ - Category:Invalid SVG3124# # {{ValidSVG}} - {{InvalidSVG}}3125# def _addtempl_prop_SVN(self):3126# fmt = self._info_filter['Properties'][0]['Format']3127# d = { u'Valid SVG': u'{{ValidSVG}}',3128# u'Invalid SVG': u'{{InvalidSVG}}', }3129# fmt = d.get(fmt, u'')3130#3131# return (fmt, bool(fmt))3132# # Category:Unidentified people3133# def _guess_Classify_People(self):3134# pass3135# # Category:Unidentified maps3136# def _guess_Classify_Maps(self):3137# pass3138# # Category:Unidentified flags3139# def _guess_Classify_Flags(self):3140# pass3141# # Category:Unidentified plants3142# def _guess_Classify_Plants(self):3143# pass3144# # Category:Unidentified coats of arms3145# def _guess_Classify_CoatsOfArms(self):3146# pass3147# # Category:Unidentified buildings3148# def _guess_Classify_Buildings(self):3149# pass3150# # Category:Unidentified trains3151# def _guess_Classify_Trains(self):3152# pass3153# # Category:Unidentified automobiles3154# def _guess_Classify_Automobiles(self):3155# pass3156# # Category:Unidentified buses3157# def _guess_Classify_Buses(self):3158# pass3159 # Category:Human legs3160 def _guess_legs_HumanLegs(self):3161 result = self._info_filter['Legs']3162 3163 return (u'Human legs', ((len(result) == 1) and (result[0]['Coverage'] >= .40)))3164 # Category:Human torsos3165 def _guess_torsos_HumanTorsos(self):3166 result = self._info_filter['Torsos']3167 3168 return (u'Human torsos', ((len(result) == 1) and (result[0]['Coverage'] >= .40)))3169 # Category:Automobiles3170 def _guess_automobiles_Automobiles(self):3171 result = self._info_filter['Automobiles']3172 3173 return (u'Automobiles', ((len(result) == 1) and (result[0]['Coverage'] >= .40)))3174 ## Category:Hands3175 #def _guess_hands_Hands(self):3176 # result = self._info_filter['Hands']3177 #3178 # return (u'Hands', ((len(result) == 1) and (result[0]['Coverage'] >= .50)))3179 # Category:Black ( 0, 0, 0)3180 # Category:Blue‎ ( 0, 0, 255)3181 # Category:Brown (165, 42, 42)3182 # Category:Green ( 0, 255, 0)3183 # Category:Orange (255, 165, 0)3184 # Category:Pink‎ (255, 192, 203)3185 # Category:Purple (160, 32, 240)3186 # Category:Red‎ (255, 0, 0)3187 # Category:Turquoise ( 64, 224, 208)3188 # Category:White‎ (255, 255, 255)3189 # Category:Yellow (255, 255, 0)3190 # http://www.farb-tabelle.de/en/table-of-color.htm3191 #def _collectColor(self):3192 #def _cat_color_Black(self):3193 # info = self._info_filter['ColorRegions']3194 # for item in info:3195 # if (u'Black' == item[u'Color']):3196 # return (u'Black', True)3197 # return (u'Black', False)3198 def __cat_color_general(self, col):3199 info = self._info_filter['ColorRegions']3200 for item in info:3201 if (col == item[u'Color']):3202 return (col, True)3203 return (col, False)3204 _cat_color_Black = lambda self: self.__cat_color_general(u'Black')3205 _cat_color_Blue = lambda self: self.__cat_color_general(u'Blue')3206 _cat_color_Brown = lambda self: self.__cat_color_general(u'Brown')3207 _cat_color_Green = lambda self: self.__cat_color_general(u'Green')3208 _cat_color_Orange = lambda self: self.__cat_color_general(u'Orange')3209 _cat_color_Pink = lambda self: self.__cat_color_general(u'Pink')3210 _cat_color_Purple = lambda self: self.__cat_color_general(u'Purple')3211 _cat_color_Red = lambda self: self.__cat_color_general(u'Red')3212 _cat_color_Turquoise = lambda self: self.__cat_color_general(u'Turquoise')3213 _cat_color_White = lambda self: self.__cat_color_general(u'White')3214 _cat_color_Yellow = lambda self: self.__cat_color_general(u'Yellow')3215# all classification and categorization methods and definitions - SVM variation3216# use 'pyml' SVM (libsvm) classifier3217# may be 'scikit-learn' or 'opencv' (svm, a.o.) could be of some use too3218class CatImages_SVM(CatImages_Default):3219 trained_cat = [u'Human_ears', u'Male faces']3220 # dummy: deactivated3221 def classifyFeatures(self):3222 for key in self._info:3223 for i in range(len(self._info[key])):3224 self._info[key][i]['Confidence'] = 1.03225 3226 # (all trained categories)3227 # http://scipy-lectures.github.com/advanced/scikit-learn/index.html3228 # http://mlpy.sourceforge.net/docs/3.5/index.html3229 # http://docs.opencv.org/modules/ml/doc/ml.html3230 def _cat_multi_generic(self):3231 # IT LOOKS LIKE (MAY BE) scikit-learn IS BETTER AND HAS MORE OPTIONS THAN pyml ... ?!!!3232 # create classifier feature set3233 # !!!currently number of detected features is used only -> lots of room for improvements!!!3234 features = []3235 for key in sorted(self._info):3236 #print key, len(self._info[key]), self._info[key]3237 features.append( len(self._info[key]) )3238 features = np.array(features)3239 linear_svm = mlpy.LibSvm().load_model('cache/test.csf')3240 yp = linear_svm.pred(features)3241 cat = self.trained_cat[int(yp)-1]3242 #print linear_svm.labels()3243 # confidence of match?3244 3245 return (cat, True)3246# Image by content categorization derived from 'checkimages.py'.3247class CatImagesBot(checkimages.checkImagesBot, CatImages_Default):3248#class CatImagesBot(checkimages.checkImagesBot, CatImages_SVM):3249# def __init__(self, site, logFulNumber = 25000, sendemailActive = False,3250# duplicatesReport = False, logFullError = True): pass3251# def setParameters(self, imageName): pass3252 # or may be '__init__' ... ???3253 def load_licenses(self):3254 #pywikibot.output(u'\n\t...Listing the procedures available...\n')3255 pywikibot.output(u'\n\t...Listing the procedures used...\n')3256 3257 self._funcs = {'filter': [], 'cat': [], 'addcat': [], 'guess': []}3258 for item in dir(self):3259 s = item.split('_')3260 if (len(s) < 3) or (s[1] not in self._funcs) or (s[2] in self.ignore):3261 continue3262 pywikibot.output( item )3263 self._funcs[s[1]].append( item )3264 self.tmpl_available_spec = tmpl_available_spec3265 gen = pagegenerators.PrefixingPageGenerator(prefix = u'Template:FileContentsByBot/')3266 buf = []3267 for item in gen:3268 item = item.title()3269 if (item[-4:] == "/doc"): # all docs3270 continue3271 item = os.path.split(item)[1]3272 if (item[0].lower() == item[0]): # e.g. 'generic'3273 continue3274 buf.append( item )3275 if buf:3276 self.tmpl_available_spec = buf3277 pywikibot.output( u'\n\t...Following specialized templates found, check them since they are used now...\n' )3278 pywikibot.output( u'tmpl_available_spec = [ %s ]\n' % u", ".join(buf) )3279 return []3280 def downloadImage(self):3281 #print self.image_path3282 pywikibot.output(u'Processing media %s ...' % self.image.title(asLink=True))3283 image_filename = os.path.split(self.image.fileUrl())[-1]3284 self.image_path = urllib2.quote(os.path.join(scriptdir, ('cache/' + image_filename[-128:])))3285 self._wikidata = self.image._latestInfo # all info wikimedia got from content (mime, sha1, ...)3286 #print self._wikidata3287 #print self._wikidata['mime']3288 #print self._wikidata['sha1']3289 #print self._wikidata['metadata']3290 #for item in self._wikidata['metadata']:3291 # print item['name'], item['value']3292 if not os.path.exists(self.image_path):3293 pywikibot.get_throttle()3294 f_url, data = self.site.getUrl(self.image.fileUrl(), no_hostname=True, 3295 back_response=True)3296 # needed patch for 'getUrl' applied upstream in r104413297 # (allows to re-read from back_response)3298 data = f_url.read()3299 del f_url # free some memory (no need to keep a copy...)3300 f = open(self.image_path, 'wb')3301 f.write( data )3302 f.close()3303 # LOOK ALSO AT: checkimages.CatImagesBot.checkStep3304 # (and category scripts/bots too...)3305 def checkStep(self):3306 self.thrshld = self._thrshld_default3307 self._info = {} # used for LOG/DEBUG OUTPUT ONLY3308 self._info_filter = {} # used for CATEGORIZATION3309 self._result_check = []3310 self._result_add = []3311 self._result_guess = []3312 # flush internal buffers3313 for attr in ['_buffer_EXIF', '_buffer_FFMPEG', '_buffer_Geometry']:#, '_content_text']:3314 if hasattr(self, attr):3315 delattr(self, attr)3316 # gather all features (information) related to current image3317 self.gatherFeatures()3318 # classification of detected features (should use RTrees, KNearest, Boost, SVM, MLP, NBayes, ...)3319 # ??? (may be do this in '_cat_...()' or '_filter_...()' ?!?...)3320 # http://opencv.itseez.com/doc/tutorials/ml/introduction_to_svm/introduction_to_svm.html3321 # http://stackoverflow.com/questions/8687885/python-opencv-svm-implementation3322 # https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/python2/letter_recog.py?rev=64803323 self.classifyFeatures() # assign confidences3324 # replace/improve this with RTrees, KNearest, Boost, SVM, MLP, NBayes, ...3325 # information template: use filter to select from gathered features3326 # the ones that get reported3327 self._info_filter = {}3328 for item in self._funcs['filter']:3329 self._info_filter.update( getattr(self, item)() )3330 # categorization: use explicit searches for classification (rel = ?)3331 for item in self._funcs['cat']:3332 (cat, rel) = getattr(self, item)()3333 #print cat, result, len(result)3334 if rel:3335 self._result_check.append( cat )3336 self._result_check = list(set(self._result_check))3337 # categorization: conditional (only if the ones before are present)3338 # (does not trigger report to page)3339 for item in self._funcs['addcat']:3340 (cat, rel) = getattr(self, item)()3341 #print cat, result, len(result)3342 if rel:3343 self._result_add.append( cat )3344 self._result_add = list(set(self._result_add))3345 # categorization: use guesses for unreliable classification (rel = 0.1)3346 if not useGuesses:3347 return self._result_check3348 for item in self._funcs['guess']:3349 (cat, rel) = getattr(self, item)()3350 #print cat, result, len(result)3351 if rel:3352 self._result_guess.append( cat )3353 return self._result_check3354 def tag_image(self):3355 self.clean_cache()3356 #if not self._existInformation(self._info_filter): # information available?3357 if not (self._result_check + self._result_guess): # category available?3358 return False3359 pywikibot.get_throttle()3360 content = self.image.get()3361 # check the type of template used on page; Information, Artwork, ...3362 for temp in [u"Information", u"Artwork"]:3363 pos = content.find(u'{{%s' % temp) + 23364 if pos > 1:3365 break3366 if pos > 1:3367 # cosmetic changes: format the page well to have '\n\n' after the template3368 diff = content[:(pos-2)].count(u'{{') - content[:(pos-2)].count(u'}}')3369 while (content[:pos].count(u'{{') - content[:pos].count(u'}}')) != diff:3370 pos = content.find(u'}}', pos) + 23371 if content[pos:(pos+2)] != (u"\n"*2):3372 content = content[:pos] + (u"\n"*2) + content[pos:].lstrip()3373 else:3374 pywikibot.warning(u'Page layout issue; Information template could '3375 u'not be found and thus the data not appended!')3376 return False3377 # append template and fill it with data3378 content = self._append_to_template(content, temp, tmpl_FileContentsByBot)3379 for i, key in enumerate(self._info_filter):3380 item = self._info_filter[key]3381 info = self._make_infoblock(key, item)3382 if info:3383 content = self._append_to_template(content, u"FileContentsByBot", info)3384 # append categories3385 tags = set([])3386 for i, cat in enumerate(list(set(self._result_check + self._result_add))):3387 tags.add( u"[[:Category:%s]]" % cat )3388 content = pywikibot.replaceCategoryLinks(content, [cat], site=self.site, addOnly=True)3389 # cleanup double categories, remove obsolete ones and add templates3390 content = pywikibot.replaceCategoryLinks( content, 3391 list(set(pywikibot.getCategoryLinks(content, site=self.site))),3392 site=self.site )3393 content = self._remove_category_or_template(content, u"Uncategorized") # template3394 content = self._add_template(content, u"Check categories|year={{subst:#time:Y}}|month={{subst:#time:F}}|day={{subst:#time:j}}|category=[[Category:Categorized by DrTrigonBot]]", top=True)3395 # add category guesses3396 for i, cat in enumerate(self._result_guess):3397 content += u"\n<!--DrTrigonBot-guess-- [[Category:%s]] -->" % cat3398 # verbosely output info about changes and apply them3399 pywikibot.output(u"--- " * 20)3400 pywikibot.output(content)3401 pywikibot.output(u"--- " * 20)3402 pywikibot.put_throttle()3403 self.image.put( content, comment="bot automatic categorization; adding %s" % u", ".join(tags),3404 botflag=False )3405# TODO: (work-a-round if https://bugzilla.wikimedia.org/show_bug.cgi?id=6421 not solved)3406# if hasattr(self, '_content_text'):3407# textpage = pywikibot.Page(self.site, os.path.join(self.image.title(), u'Contents/Text'))3408# textpage.put( self._content_text, comment="bot adding content from %s" % textpage.title(asLink=True),3409# botflag=False )3410 return True3411 def log_output(self):3412 # ColorRegions always applies here since there is at least 1 (THE average) color...3413 ignore = ['Properties', 'Metadata', 'ColorAverage', 'ColorRegions', 'Geometry']3414 #if not self._existInformation(self._info): # information available?3415 # information available? AND/OR category available?3416 if not (self._existInformation(self._info, ignore = ignore) or self._result_check):3417 return u""3418 ret = []3419 ret.append( u"" )3420 ret.append( u"== [[:%s]] ==" % self.image.title() )3421 ret.append( u'{|' )3422 ret.append( u'|<div style="position:relative;">' )3423 ret.append( u"[[%s|200px]]" % self.image.title() )3424 ret.append( self._make_markerblock(self._info[u'Faces'], 200.,3425 structure=['Position', 'Eyes', 'Mouth', 'Nose']) )3426 ret.append( self._make_markerblock(self._info[u'People'], 200.,3427 line='dashed') )3428 ret.append( u"</div>" )3429 ret.append( u'|<div style="position:relative;">' )3430 ret.append( u"[[%s|200px]]" % self.image.title() )3431 ret.append( self._make_markerblock(self._info[u'ColorRegions'], 200.) )3432 ret.append( self._make_markerblock(self._info[u'OpticalCodes'], 200.,3433 line='dashed') )3434 ret.append( u"</div>" )3435 ret.append( u'|<div style="position:relative;">' )3436 ret.append( u"[[%s|200px]]" % self.image.title() )3437 ret.append( self._make_markerblock(self._info[u'Ears'], 200.) )3438 ret.append( self._make_markerblock(self._info[u'Eyes'], 200.) )3439 ret.append( self._make_markerblock(self._info[u'Legs'], 200.,3440 line='dashed') )3441 ret.append( self._make_markerblock(self._info[u'Torsos'], 200.,3442 line='dashed') )3443 ret.append( self._make_markerblock(self._info[u'Automobiles'], 200.,3444 line='dashed') )3445 #ret.append( self._make_markerblock(self._info[u'Hands'], 200.,3446 # line='dashed') )3447 ret.append( u"</div>" )3448 ret.append( u'|}' )3449 color = {True: "rgb(0,255,0)", False: "rgb(255,0,0)"}[bool(self._result_check + self._result_guess)]3450 ret.append( u"<div style='background:%s'>'''automatic categorization''': %s</div>" % (color, u", ".join(list(set(self._result_check + self._result_add)))) )3451 buf = []3452 for i, key in enumerate(self._info):3453 item = self._info[key]3454 info = self._make_infoblock(key, item, [])3455 if info:3456 buf.append( info )3457 ret.append( tmpl_FileContentsByBot[3:] + u"\n" + u"\n".join( buf ) + u"\n}}" )3458 return u"\n".join( ret )3459 def clean_cache(self):3460 if os.path.exists(self.image_path):3461 os.remove( self.image_path )3462 #if os.path.exists(self.image_path_JPEG):3463 # os.remove( self.image_path_JPEG )3464 ##image_path_new = self.image_path_JPEG.replace(u"cache/", u"cache/0_DETECTED_")3465 ##if os.path.exists(image_path_new):3466 ## os.remove( image_path_new )3467 # LOOK ALSO AT: checkimages.CatImagesBot.report3468 def report(self):3469 tagged = self.tag_image()3470 logged = self.log_output()3471 return (tagged, logged)3472 def _make_infoblock(self, cat, res, tmpl_available=None):3473 if not res:3474 return u''3475 if (tmpl_available == None):3476 tmpl_available = self.tmpl_available_spec3477 generic = (cat not in tmpl_available)3478 titles = res[0].keys()3479 if not titles:3480 return u''3481 result = []3482 #result.append( u'{{(!}}style="background:%s;"' % {True: 'green', False: 'red'}[report] )3483 if generic:3484 result.append( u"{{FileContentsByBot/generic|name=%s|" % cat )3485 buf = dict([ (key, []) for key in titles ])3486 for item in res:3487 for key in titles:3488 buf[key].append( self._output_format(item[key]) )3489 for key in titles:3490 result.append( u" {{FileContentsByBot/generic|name=%s|value=%s}}" % (key, u"; ".join(buf[key])) )3491 else:3492 result.append( u"{{FileContentsByBot/%s|" % cat )3493 for item in res:3494 result.append( u" {{FileContentsByBot/%s" % cat )3495 for key in titles:3496 if item[key]: # (work-a-round for empty 'Eyes')3497 result.append( self._output_format_flatten(key, item[key]) )3498 result.append( u" }}" )3499 result.append( u"}}" )3500 return u"\n".join( result )3501 def _output_format(self, value):3502 if (type(value) == type(float())):3503 # round/strip floats3504 return "%.3f" % value3505 else:3506 # output string representation of variable3507 return str(value)3508 def _output_format_flatten(self, key, value):3509 # flatten structured varible recursively3510 if (type(value) == type(tuple())) or (type(value) == type(list())):3511 buf = []3512 for i, t in enumerate(value):3513 buf.append( self._output_format_flatten(key + (u"-%02i" % i), t) )3514 return u"\n".join( buf )3515 else:3516 # end of recursion3517 return u" | %s = %s" % (key, self._output_format(value))3518 def _make_markerblock(self, res, size, structure=['Position'], line='solid'):3519 # same as in '_detect_Faces'3520 colors = [ (0,0,255),3521 (0,128,255),3522 (0,255,255),3523 (0,255,0),3524 (255,128,0),3525 (255,255,0),3526 (255,0,0),3527 (255,0,255) ]3528 result = []3529 for i, r in enumerate(res):3530 if ('RGB' in r):3531 color = list(np.array((255,255,255))-np.array(r['RGBref']))3532 else:3533 color = list(colors[i%8])3534 color.reverse()3535 color = u"%02x%02x%02x" % tuple(color)3536 3537 #scale = r['size'][0]/size3538 scale = self.image_size[0]/size3539 f = list(np.array(r[structure[0]])/scale)3540 3541 result.append( u'<div class="%s-marker" style="position:absolute; left:%ipx; top:%ipx; width:%ipx; height:%ipx; border:2px %s #%s;"></div>' % tuple([structure[0].lower()] + f + [line, color]) )3542 for ei in range(len(structure)-1):3543 data = r[structure[ei+1]]3544 if data and (not hasattr(data[0], '__iter__')): # Mouth and Nose are not lists3545 data = [ r[structure[ei+1]] ]3546 for e in data:3547 e = list(np.array(e)/scale)3548 3549 result.append( u'<div class="%s-marker" style="position:absolute; left:%ipx; top:%ipx; width:%ipx; height:%ipx; border:2px solid #%s;"></div>' % tuple([structure[ei+1].lower()] + e + [color]) )3550 return u"\n".join( result )3551 # place into 'textlib' (or else e.g. 'catlib'/'templib'...)3552 def _remove_category_or_template(self, text, name):3553 text = re.sub(u"[\{\[]{2}%s.*?[\}\]]{2}\n?" % name, u"", text)3554 return text3555 # place into 'textlib'3556 def _add_template(self, text, name, params={}, top=False, raw=False):3557 if top:3558 buf = [(u"{{%s}}" % name), text]3559 else:3560 if raw:3561 buf = [text, name]3562 else:3563 buf = [text, (u"{{%s}}" % name)]3564 return u"\n".join( buf )3565 # place into 'textlib' (or else e.g. 'catlib'/'templib'...)3566 def _append_to_template(self, text, name, append):3567 # mask/search template to append to3568 pattern = re.compile(u"(\{\{%s.*?\n)(\s*\}\}\n{2})" % name, flags=re.S)3569 template = pattern.search(text).groups()3570 # append to template3571 template = u"".join( [template[0], append, u"\n", template[1]] )3572 # apply changes3573 text = pattern.sub(template, text)3574 return text3575 # gather data from all information interfaces3576 def gatherFeatures(self):3577 # split detection and extraction according to file types; _JpegFile, ...3578 with GenericFile(self.image_path) as gf:3579 gf.image = self.image # patch for _SvgFile needing url3580 for func in ['getProperties', 'getFeatures']:3581 result = getattr(gf, func)()3582 self._info.update(result)3583 self.image_size = gf.image_size3584 def _existInformation(self, info, ignore = ['Properties', 'Metadata', 'ColorAverage']):3585 result = []3586 for item in info:3587 if item in ignore:3588 continue3589 if info[item]:3590 result.append( item )3591 return result3592 def _filter_Properties(self):3593 # >>> never drop <<<3594 result = self._info['Properties']3595 return {'Properties': result}3596 def _filter_Metadata(self):3597 # >>> never drop <<<3598 result = self._info['Metadata']3599 return {'Metadata': result}3600 def _filter_Faces(self):3601 result = self._info['Faces']3602 if (len(result) < self._thrhld_group_size):3603 buf = []3604 for item in self._info['Faces']:3605 # >>> drop if below thrshld <<<3606 if (item['Confidence'] >= self.thrshld):3607 buf.append( item )3608 result = buf3609 return {'Faces': result}3610 def _filter_People(self):3611 result = self._info['People']3612 if (len(result) < self._thrhld_group_size):3613 buf = []3614 for item in self._info['People']:3615 # >>> drop if below thrshld <<<3616 if (item['Confidence'] >= self.thrshld):3617 buf.append( item )3618 result = buf3619 return {'People': result}3620 def _filter_ColorRegions(self):3621 #result = {}3622 result = []3623 for item in self._info['ColorRegions']:3624 ## >>> drop wrost ones... (ignore all below 0.2) <<<3625 #if (result.get(item['Color'], {'Confidence': 0.2})['Confidence'] < item['Confidence']):3626 # result[item['Color']] = item3627 # >>> drop if below thrshld <<<3628 if (item['Confidence'] >= self.thrshld):3629 result.append( item )3630 #return {'ColorRegions': [result[item] for item in result]}3631 return {'ColorRegions': result}3632 def _filter_ColorAverage(self):3633 # >>> never drop <<<3634 result = self._info['ColorAverage']3635 return {'ColorAverage': result}3636 def _filter_OpticalCodes(self):3637 # use all, since detection should be very reliable3638 #result = self._info['OpticalCodes']3639 result = []3640 for item in self._info['OpticalCodes']:3641 # >>> drop if below thrshld <<<3642 if (item['Confidence'] >= self.thrshld):3643 result.append( item )3644 return {'OpticalCodes': result}3645 def _filter_Chessboard(self):3646 # use all, since detection should be very reliable3647 result = self._info['Chessboard']3648 return {'Chessboard': result}3649 def _filter_Text(self):3650 # use all, since detection should be very reliable3651 result = self._info['Text']3652 return {'Text': result}3653 def _filter_Legs(self):3654 result = []3655 for item in self._info['Legs']:3656 # >>> drop if below thrshld <<<3657 if (item['Confidence'] >= self.thrshld):3658 result.append( item )3659 return {'Legs': result}3660 def _filter_Torsos(self):3661 result = []3662 for item in self._info['Torsos']:3663 # >>> drop if below thrshld <<<3664 if (item['Confidence'] >= self.thrshld):3665 result.append( item )3666 return {'Torsos': result}3667 def _filter_Ears(self):3668 result = []3669 for item in self._info['Ears']:3670 # >>> drop if below thrshld <<<3671 if (item['Confidence'] >= self.thrshld):3672 result.append( item )3673 return {'Ears': result}3674 def _filter_Eyes(self):3675 result = []3676 for item in self._info['Eyes']:3677 # >>> drop if below thrshld <<<3678 if (item['Confidence'] >= self.thrshld):3679 result.append( item )3680 return {'Eyes': result}3681 def _filter_Automobiles(self):3682 result = []3683 for item in self._info['Automobiles']:3684 # >>> drop if below thrshld <<<3685 if (item['Confidence'] >= self.thrshld):3686 result.append( item )3687 return {'Automobiles': result}3688 def _filter_Streams(self):3689 # use all, (should be reliable)3690 result = self._info['Streams']3691 return {'Streams': result}3692# def _filter_Audio(self):3693# # use all, (should be reliable)3694# result = self._info['Audio']3695# return {'Audio': result}3696 #def _filter_Geometry(self):3697 # result = []3698 # for item in self._info['Geometry']:3699 # # >>> drop if below thrshld <<<3700 # if (item['Confidence'] >= self.thrshld):3701 # result.append( item )3702 # return {'Geometry': result}3703 #def _filter_Hands(self):3704 # result = []3705 # for item in self._info['Hands']:3706 # # >>> drop if below thrshld <<<3707 # if (item['Confidence'] >= self.thrshld):3708 # result.append( item )3709 # return {'Hands': result}3710# def _filter_Classify(self):3711# from operator import itemgetter3712# result = sorted(self._info['Classify'][0].items(), key=itemgetter(1))3713# result.reverse()3714# pywikibot.output(u' Best: %s' % result[:3] )3715# pywikibot.output(u'Worst: %s' % result[-3:] )3716#3717# # >>> dummy: drop all (not reliable yet since untrained) <<<3718# return {'Classify': []}3719def main():3720 """ Main function """3721 global useGuesses3722 # Command line configurable parameters3723 limit = 150 # How many images to check?3724# untagged = False # Use the untagged generator3725 sendemailActive = False # Use the send-email3726 train = False3727 generator = None3728 # default3729 if len(sys.argv) < 2:3730 sys.argv += ['-cat']3731 # debug: 'python catimages.py -debug'3732 # run/test: 'python catimages.py [-start:File:abc]'3733 sys.argv += ['-family:commons', '-lang:commons']3734 #sys.argv += ['-noguesses']3735 # try to resume last run and continue3736 if os.path.exists( os.path.join(scriptdir, 'cache/catimages_start') ):3737 shutil.copy2(os.path.join(scriptdir, 'cache/catimages_start'), os.path.join(scriptdir, 'cache/catimages_start.bak'))3738 posfile = open(os.path.join(scriptdir, 'cache/catimages_start'), "r")3739 firstPageTitle = posfile.read().decode('utf-8')3740 posfile.close()3741 else:3742 firstPageTitle = None3743 # Here below there are the parameters.3744 for arg in pywikibot.handleArgs():3745 if arg.startswith('-limit'):3746 if len(arg) == 7:3747 limit = int(pywikibot.input(u'How many files do you want to check?'))3748 else:3749 limit = int(arg[7:])3750# elif arg == '-sendemail':3751# sendemailActive = True3752 elif arg.startswith('-start'):3753 if len(arg) == 6:3754 firstPageTitle = None3755 elif len(arg) > 6:3756 firstPageTitle = arg[7:]3757 #firstPageTitle = firstPageTitle.split(":")[1:]3758 #generator = pywikibot.getSite().allpages(start=firstPageTitle, namespace=6)3759 elif arg.startswith('-cat'):3760 if len(arg) == 4:3761 catName = u'Media_needing_categories'3762 elif len(arg) > 4:3763 catName = str(arg[5:])3764 catSelected = catlib.Category(pywikibot.getSite(), 'Category:%s' % catName)3765 generator = pagegenerators.CategorizedPageGenerator(catSelected, recurse = True)3766# elif arg.startswith('-untagged'):3767# untagged = True3768# if len(arg) == 9:3769# projectUntagged = str(pywikibot.input(u'In which project should I work?'))3770# elif len(arg) > 9:3771# projectUntagged = str(arg[10:])3772 elif arg == '-noguesses':3773 useGuesses = False3774 elif arg.startswith('-single'):3775 if len(arg) > 7:3776 pageName = unicode(arg[8:])3777 if 'File:' not in pageName:3778 pageName = 'File:%s' % pageName3779 generator = [ pywikibot.Page(pywikibot.getSite(), pageName) ]3780 firstPageTitle = None3781 elif arg.startswith('-train'):3782 train = True3783 generator = None3784 # Understand if the generator is present or not.3785 if not generator:3786 pywikibot.output(u'no generator defined... EXIT.')3787 sys.exit()3788 3789 # Define the site.3790 site = pywikibot.getSite()3791 # Block of text to translate the parameters set above.3792 image_old_namespace = u"%s:" % site.image_namespace()3793 image_namespace = u"File:"3794 # A little block-statement to ensure that the bot will not start with en-parameters3795 if site.lang not in project_inserted:3796 pywikibot.output(u"Your project is not supported by this script. You have to edit the script and add it!")3797 return3798 # Defing the Main Class.3799 Bot = CatImagesBot(site, sendemailActive = sendemailActive,3800 duplicatesReport = False, logFullError = False)3801# # Untagged is True? Let's take that generator3802# if untagged == True:3803# generator = Bot.untaggedGenerator(projectUntagged, limit)3804 # Ok, We (should) have a generator, so let's go on.3805 # Take the additional settings for the Project3806 Bot.takesettings()3807 # do classifier training on good (homgenous) commons categories3808 if train:3809 trainbot(generator, Bot, image_old_namespace, image_namespace)3810 return3811 # Not the main, but the most important loop.3812 outresult = []3813 for image in generator:3814 if firstPageTitle:3815 if (image.title() == firstPageTitle):3816 pywikibot.output( u"found last page '%s' ..." % image.title() )3817 firstPageTitle = None3818 continue3819 else:3820 #pywikibot.output( u"skipping page '%s' ..." % image.title() )3821 continue3822 # recover from hard crash in the run before, thus skip one more page3823 if os.path.exists( os.path.join(scriptdir, 'cache/catimages_recovery') ):3824 pywikibot.output( u"trying to recover from hard crash, skipping page '%s' ..." % image.title() )3825 disable_recovery()3826 # in case the next one has a hard-crash too...3827 posfile = open(os.path.join(scriptdir, 'cache/catimages_start'), "w")3828 posfile.write( image.title().encode('utf-8') )3829 posfile.close()3830 continue3831 #comment = None # useless, also this, let it here for further developments3832 try:3833 imageName = image.title().split(image_namespace)[1] # Deleting the namespace (useless here)3834 except IndexError:# Namespace image not found, that's not an image! Let's skip...3835 try:3836 imageName = image.title().split(image_old_namespace)[1]3837 except IndexError:3838 pywikibot.output(u"%s is not a file, skipping..." % image.title())3839 continue3840 Bot.setParameters(imageName) # Setting the image for the main class3841 try:3842 Bot.downloadImage()3843 except IOError, err:3844 # skip if download not possible3845 pywikibot.warning(u"%s, skipped..." % err)3846 continue3847 except:3848 # skip on any unexpected error, but report it3849 pywikibot.exception(tb=True)3850 pywikibot.error(u"was not able to process page %s !!!\n" %\3851 image.title(asLink=True))3852 continue3853 resultCheck = Bot.checkStep()3854 tagged = False3855 try:3856 (tagged, ret) = Bot.report()3857 if ret:3858 outresult.append( ret )3859 except AttributeError:3860 pywikibot.exception(tb=True)3861 pywikibot.error(u"was not able to process page %s !!!\n" %\3862 image.title(asLink=True))3863 limit += -13864 if not tagged:3865 posfile = open(os.path.join(scriptdir, 'cache/catimages_start'), "w")3866 posfile.write( image.title().encode('utf-8') )3867 posfile.close()3868 if limit <= 0:3869 break3870 if resultCheck:3871 continue3872 if outresult:3873 outpage = pywikibot.Page(site, u"User:DrTrigon/User:DrTrigonBot/logging")3874 #outresult = [ outpage.get() ] + outresult # append to page3875 outresult = u"\n".join(outresult)3876 pywikibot.output(u"Size of log page data: %s byte(s)" % len(outresult))3877 # work-a-round: write pages mutliple times if content is too large in order to circumvent3878 # "HTTPError: 504 Gateway Time-out" leading finally to "MaxTriesExceededError"3879 # (why is that...?!?? FIX THIS in the framework core e.g. 'postForm'!)3880 tmp = outresult3881 while tmp:3882 i = np.array([m.start() for m in re.finditer(u"\n\n==", tmp)]3883 + [len(tmp)])3884 #pos = i[ np.where((i - 2048*1024) <= 0)[0][-1] ] # $wgMaxArticleSize3885 pos = i[ np.where((i - 500*1024) <= 0)[0][-1] ]3886 pywikibot.output(u"Size of bunch to write: %s byte(s)" % len(tmp[:pos]))3887 outpage.put( tmp[:pos], comment="bot writing log for last run" )3888 tmp = tmp[pos:]3889 if pywikibot.simulate:3890 #print u"--- " * 203891 #print u"--- " * 203892 #print outresult3893 posfile = open(os.path.join(scriptdir, 'cache/catimages.log'), "a")3894 posfile.write( outresult )3895 posfile.close()3896# http://scipy-lectures.github.com/advanced/scikit-learn/index.html3897# http://mlpy.sourceforge.net/docs/3.5/index.html3898# http://docs.opencv.org/modules/ml/doc/ml.html3899# train pyml (svm), opencv BoW and haarcascade classifiers3900# choose a good and meaningful featureset from extracted (better than actual one)3901def trainbot(generator, Bot, image_old_namespace, image_namespace):3902 # IT LOOKS LIKE (MAY BE) scikit-learn IS BETTER AND HAS MORE OPTIONS THAN pyml ... ?!!!3903 # gather training dataset from wiki commons categories3904 trainset = []3905 for i, catName in enumerate(Bot.trained_cat):3906 catSelected = catlib.Category(pywikibot.getSite(), 'Category:%s' % catName)3907 generator = pagegenerators.CategorizedPageGenerator(catSelected)3908 for image in generator:3909 try:3910 imageName = image.title().split(image_namespace)[1] # Deleting the namespace (useless here)3911 except IndexError:# Namespace image not found, that's not an image! Let's skip...3912 try:3913 imageName = image.title().split(image_old_namespace)[1]3914 except IndexError:3915 pywikibot.output(u"%s is not a file, skipping..." % image.title())3916 continue3917 Bot.setParameters(imageName) # Setting the image for the main class3918 try:3919 Bot.downloadImage()3920 except IOError, err:3921 # skip if download not possible3922 pywikibot.warning(u"%s, skipped..." % err)3923 continue3924 except Exception, err:3925 # skip on any unexpected error, but report it3926 pywikibot.error(u"%s" % err)3927 pywikibot.error(u"was not able to process page %s !!!\n" %\3928 image.title(asLink=True))3929 continue3930 # gather all features (information) related to current image3931 Bot._info = {}3932 Bot.gatherFeatures()3933 3934 # create classifier feature set3935 # !!!currently number of detected features is used only -> lots of room for improvements!!!3936 # choose a good and meaningful featureset from extracted (better than actual one)3937 features = []3938 for key in sorted(Bot._info):3939 #print key, len(self._info[key]), self._info[key]3940 features.append( len(Bot._info[key]) )3941 features.append( i+1 ) # category id (returned by predictor later)3942 #print features3943 trainset.append( features )3944 trainset = np.array(trainset)3945 cols = trainset.shape[1]3946 # http://mlpy.sourceforge.net/docs/3.5/tutorial.html3947 import matplotlib.pyplot as plt # required for plotting3948 ##iris = np.loadtxt('iris.csv', delimiter=',')3949 ##x, y = iris[:, :4], iris[:, 4].astype(np.int) # x: (observations x attributes) matrix, y: classes (1: setosa, 2: versicolor, 3: virginica)3950 #trainset = np.loadtxt('cache/test.csv', delimiter=' ')3951 #cols = trainset.shape[1]3952 #print trainset3953 x, y = trainset[:, :(cols-1)], trainset[:, (cols-1)].astype(np.int) # x: (observations x attributes) matrix, y: classes (1: setosa, 2: versicolor, 3: virginica)3954 pywikibot.output(x.shape)3955 pywikibot.output(y.shape)3956 3957 # Dimensionality reduction by Principal Component Analysis (PCA)3958 pca = mlpy.PCA() # new PCA instance3959 pca.learn(x) # learn from data3960 z = pca.transform(x, k=2) # embed x into the k=2 dimensional subspace3961 pywikibot.output(z.shape)3962 3963 plt.set_cmap(plt.cm.Paired)3964 fig1 = plt.figure(1)3965 title = plt.title("PCA on dataset")3966 plot = plt.scatter(z[:, 0], z[:, 1], c=y)3967 labx = plt.xlabel("First component")3968 laby = plt.ylabel("Second component")3969 plt.show()3970 3971 # Learning by Kernel Support Vector Machines (SVMs) on principal components3972 linear_svm = mlpy.LibSvm(kernel_type='linear') # new linear SVM instance3973 linear_svm.learn(z, y) # learn from principal components3974 3975 # !!! train also BoW (bag-of-words) in '_detectclassify_ObjectAll' resp. 'opencv.BoWclassify.main' !!!3976 3977 xmin, xmax = z[:,0].min()-0.1, z[:,0].max()+0.13978 ymin, ymax = z[:,1].min()-0.1, z[:,1].max()+0.13979 xx, yy = np.meshgrid(np.arange(xmin, xmax, 0.01), np.arange(ymin, ymax, 0.01))3980 zgrid = np.c_[xx.ravel(), yy.ravel()]3981 yp = linear_svm.pred(zgrid)3982 3983 plt.set_cmap(plt.cm.Paired)3984 fig2 = plt.figure(2)3985 title = plt.title("SVM (linear kernel) on principal components")3986 plot1 = plt.pcolormesh(xx, yy, yp.reshape(xx.shape))3987 plot2 = plt.scatter(z[:, 0], z[:, 1], c=y)3988 labx = plt.xlabel("First component")3989 laby = plt.ylabel("Second component")3990 limx = plt.xlim(xmin, xmax)3991 limy = plt.ylim(ymin, ymax)3992 plt.show()3993 3994 linear_svm.save_model('cache/test.csf')3995 pywikibot.output(u'Linear SVM model stored to %s.' % 'cache/test.csf')3996# for functions in C/C++ that might crash hard without any exception throwed3997# e.g. an abort due to an assert or something else3998def enable_recovery():3999 recoveryfile = open(os.path.join(scriptdir, 'cache/catimages_recovery'), "w")4000 recoveryfile.write('')4001 recoveryfile.close()4002def disable_recovery():4003 if os.path.exists( os.path.join(scriptdir, 'cache/catimages_recovery') ):4004 os.remove( os.path.join(scriptdir, 'cache/catimages_recovery') )4005# Main loop will take all the (name of the) images and then i'll check them.4006if __name__ == "__main__":4007 old = datetime.datetime.strptime(str(datetime.datetime.utcnow()).split('.')[0], "%Y-%m-%d %H:%M:%S") #timezones are UTC4008 if sys.exc_info()[0]: # re-raise ImportError4009 raise #4010 try:4011 main()4012 finally:4013 final = datetime.datetime.strptime(str(datetime.datetime.utcnow()).split('.')[0], "%Y-%m-%d %H:%M:%S") #timezones are UTC4014 delta = final - old4015 secs_of_diff = delta.seconds4016 pywikibot.output("Execution time: %s" % secs_of_diff)...

Full Screen

Full Screen

Excel.py

Source:Excel.py Github

copy

Full Screen

...27 sheet = sheets[0]28 rows = sheet.rows29 for row in rows:30 row_value = [col.value for col in row]31 self._info_filter(row_value)32 if not self._header:33 self._header = self._info.pop(0)34 else:35 self._info.pop(0) # 删除开头的数据也即表头36 def _info_filter(self, info):37 pass38 def _write(self):39 if self._type == 'w':40 pass41 else:42 self._read()43 self._wb_write = openpyxl.Workbook()44 ws = self._wb_write.active45 ws.append(self._header)46 print('已经写入header了')47 print(len(self._info))48 for row in self._info:49 ws.append(row)50 print('全部写入')...

Full Screen

Full Screen

Copyfile.py

Source:Copyfile.py Github

copy

Full Screen

...8import openpyxl9from Classes.Excel import Excel10log_path = r'C:\Users\dell\Desktop\IV.xlsx'11class ReadLog(Excel):12 def _info_filter(self, info):13 self._info.append([info[-2] + '.png', info[-1] + '.png'])14 def _read(self):15 if not self._info:16 print('到这一步了')17 self._wb_read = openpyxl.load_workbook(self._path_read)18 sheets = self._wb_read.worksheets # 获取当前所有的sheet19 sheet = sheets[4]20 rows = sheet.rows21 for row in rows:22 row_value = [col.value for col in row]23 self._info_filter(row_value)24 if not self._header:25 self._header = self._info.pop(0)26 else:27 self._info.pop(0) # 删除开头的数据也即表头28excel = ReadLog(log_path, None, types='r')29excel.read()30paths = excel.info31base_path = r'G:\LearmonthData\learmonth_pics'32new_path = r'G:\LearmonthData\IV\5-12h'33for path in paths:34 full_path = os.path.join(base_path, path[-1])35 new_full_path = os.path.join(new_path, path[-2])36 try:37 shutil.copyfile(full_path, new_full_path)...

Full Screen

Full Screen

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.

LambdaTest Learning Hubs:

YouTube

You could also refer to video tutorials over LambdaTest YouTube channel to get step by step demonstration from industry experts.

Run autotest automation tests on LambdaTest cloud grid

Perform automation testing on 3000+ real desktop and mobile devices online.

Try LambdaTest Now !!

Get 100 minutes of automation test minutes FREE!!

Next-Gen App & Browser Testing Cloud

Was this article helpful?

Helpful

NotHelpful