def process(self):
"""
Performs the line segmentation.
"""
with PyTessBaseAPI(path=TESSDATA_PREFIX) as tessapi:
for (n, input_file) in enumerate(self.input_files):
pcgts = from_file(self.workspace.download_file(input_file))
image_url = pcgts.get_Page().imageFilename
for region in pcgts.get_Page().get_TextRegion():
log.debug("Detecting lines in %s with tesseract",
region.id)
image = self.workspace.resolve_image_as_pil(
image_url,
polygon_from_points(region.get_Coords().points))
tessapi.SetImage(image)
offset = xywh_from_points(region.get_Coords().points)
for (line_no, component) in enumerate(
tessapi.GetComponentImages(RIL.TEXTLINE, True)):
line_id = '%s_line%04d' % (region.id, line_no)
line_xywh = component[1]
line_xywh['x'] += offset['x']
line_xywh['y'] += offset['y']
line_points = points_from_xywh(line_xywh)
region.add_TextLine(
TextLineType(id=line_id,
Coords=CoordsType(line_points)))
ID = concat_padded(self.output_file_grp, n)
self.workspace.add_file(
ID=ID,
file_grp=self.output_file_grp,
basename=ID + '.xml',
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts).encode('utf-8'),
)
def process(self):
"""
Performs the (text) recognition.
"""
print(self.parameter)
with PyTessBaseAPI(path=TESSDATA_PREFIX, lang=DEFAULT_MODEL) as tessapi:
log.info("Using model %s in %s for recognition", get_languages()[0], get_languages()[1][-1])
for (n, input_file) in enumerate(self.input_files):
log.info("INPUT FILE %i / %s", n, input_file)
pcgts = from_file(self.workspace.download_file(input_file))
# TODO use binarized / gray
pil_image = self.workspace.resolve_image_as_pil(pcgts.get_Page().imageFilename)
tessapi.SetImage(pil_image)
# TODO slow
# tessapi.SetPageSegMode(PSM.SINGLE_LINE)
log.info("page %s", pcgts)
for region in pcgts.get_Page().get_TextRegion():
textlines = region.get_TextLine()
log.info("About to recognize text in %i lines of region '%s'", len(textlines), region.id)
for line in textlines:
log.debug("Recognizing text in line '%s'", line.id)
xywh = xywh_from_points(line.get_Coords().points)
tessapi.SetRectangle(xywh['x'], xywh['y'], xywh['w'], xywh['h'])
# log.debug("xywh: %s", xywh)
line.add_TextEquiv(TextEquivType(Unicode=tessapi.GetUTF8Text()))
# tessapi.G
# print(tessapi.AllWordConfidences())
ID = mets_file_id(self.output_file_grp, n)
self.add_output_file(
ID=ID,
file_grp=self.output_file_grp,
basename=ID + '.xml',
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts).encode('utf-8'),
)
def process(self):
"""
Performs the region segmentation.
"""
with tesserocr.PyTessBaseAPI(path=TESSDATA_PREFIX) as tessapi:
print(self.input_file_grp)
for (n, input_file) in enumerate(self.input_files):
pcgts = from_file(self.workspace.download_file(input_file))
image = self.workspace.resolve_image_as_pil(
pcgts.get_Page().imageFilename)
log.debug("Detecting regions with tesseract")
tessapi.SetImage(image)
for component in tessapi.GetComponentImages(
tesserocr.RIL.BLOCK, True):
points, index = points_from_xywh(
component[1]), component[2]
#
# the region reference in the reading order element
#
ID = "region%04d" % index
log.debug("Detected region '%s': %s", ID, points)
# <pg:ReadingOrder>
ro = pcgts.get_Page().get_ReadingOrder()
if ro is None:
ro = ReadingOrderType()
pcgts.get_Page().set_ReadingOrder(ro)
# <pg:OrderedGroup>
og = ro.get_OrderedGroup()
if og is None:
og = OrderedGroupType(id="reading-order")
ro.set_OrderedGroup(og)
# <pg:RegionRefIndexed>
og.add_RegionRefIndexed(
RegionRefIndexedType(regionRef=ID, index=index))
#
# text region
#
pcgts.get_Page().add_TextRegion(
TextRegionType(id=ID,
Coords=CoordsType(points=points)))
ID = concat_padded(self.output_file_grp, n)
self.workspace.add_file(
ID=ID,
file_grp=self.output_file_grp,
basename=ID + '.xml',
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts).encode('utf-8'),
)
def process(self):
"""
Performs the line segmentation.
"""
with PyTessBaseAPI(
psm=PSM.SINGLE_LINE,
path=TESSDATA_PREFIX,
) as tessapi:
for (n, input_file) in enumerate(self.input_files):
pcgts = from_file(self.workspace.download_file(input_file))
image_url = pcgts.get_Page().imageFilename
for region in pcgts.get_Page().get_TextRegion():
for line in region.get_TextLine():
log.debug("Detecting words in line '%s'", line.id)
image = self.workspace.resolve_image_as_pil(
image_url,
polygon_from_points(line.get_Coords().points))
tessapi.SetImage(image)
offset = xywh_from_points(line.get_Coords().points)
for (word_no, component) in enumerate(
tessapi.GetComponentImages(RIL.WORD, True)):
word_id = '%s_word%04d' % (line.id, word_no)
word_xywh = component[1]
word_xywh['x'] += offset['x']
word_xywh['y'] += offset['y']
line.add_Word(
WordType(id=word_id,
Coords=CoordsType(
points_from_xywh(word_xywh))))
ID = concat_padded(self.output_file_grp, n)
self.add_output_file(
ID=ID,
file_grp=self.output_file_grp,
basename=ID + '.xml',
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts).encode('utf-8'),
)
def write_alignment_to_xml(self):
"""
Write the alignments into new output-file-group.
The alignment is done by the master file (first index)
"""
self.log.info("writing alignment to %s", self.process.output_file_grp)
master = self.ifs[0]
pcgts = from_file(self.process.workspace.download_file(master))
ilist = iter(self.line_alignments)
for region in pcgts.get_Page().get_TextRegion():
for line in region.get_TextLine():
self.log.info("line: %s", line.get_TextEquiv()[0].Unicode)
line.get_TextEquiv()[0].set_index(0)
current = next(ilist)
self.add_line_alignments(line, current)
self.add_word_alignments(line, current)
self.log.debug("master basename: %s", master.basename)
self.process.add_output_file(
ID="{}_{}".format(master.ID, self.process.output_file_grp),
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts),
file_grp=self.process.output_file_grp,
basename=master.basename,
)
def process(self):
"""
Performs the (text) recognition.
"""
print(self.parameter)
if self.parameter['textequiv_level'] not in ['line', 'glyph']:
raise Exception("currently only implemented at the line/glyph level")
model = get_languages()[1][-1] # last installed model
if 'model' in self.parameter:
model = self.parameter['model']
if model not in get_languages()[1]:
raise Exception("configured model " + model + " is not installed")
with PyTessBaseAPI(path=TESSDATA_PREFIX, lang=model) as tessapi:
log.info("Using model %s in %s for recognition", model, get_languages()[0])
for (n, input_file) in enumerate(self.input_files):
log.info("INPUT FILE %i / %s", n, input_file)
pcgts = from_file(self.workspace.download_file(input_file))
# TODO use binarized / gray
pil_image = self.workspace.resolve_image_as_pil(pcgts.get_Page().imageFilename)
tessapi.SetImage(pil_image)
# TODO slow
# tessapi.SetPageSegMode(PSM.SINGLE_LINE)
log.info("page %s", pcgts)
for region in pcgts.get_Page().get_TextRegion():
textlines = region.get_TextLine()
log.info("About to recognize text in %i lines of region '%s'", len(textlines), region.id)
for line in textlines:
log.debug("Recognizing text in line '%s'", line.id)
xywh = xywh_from_points(line.get_Coords().points)
tessapi.SetRectangle(xywh['x'], xywh['y'], xywh['w'], xywh['h'])
tessapi.SetPageSegMode(PSM.SINGLE_LINE)
# log.debug("xywh: %s", xywh)
line.add_TextEquiv(TextEquivType(Unicode=tessapi.GetUTF8Text()))
# tessapi.G
# print(tessapi.AllWordConfidences())
if self.parameter['textequiv_level'] == 'glyph':
for word in line.get_Word():
log.debug("Recognizing text in word '%s'", word.id)
xywh = xywh_from_points(word.get_Coords().points)
tessapi.SetRectangle(xywh['x'], xywh['y'], xywh['w'], xywh['h'])
tessapi.SetPageSegMode(PSM.SINGLE_WORD)
word.add_TextEquiv(TextEquivType(Unicode=tessapi.GetUTF8Text()))
result_it = tessapi.GetIterator()
for (result_no, result) in enumerate(iterate_level(result_it, RIL.SYMBOL)):
#symb = result.GetUTF8Text(RIL.SYMBOL) # is first choice?
#conf = result.Confidence(RIL.SYMBOL) # is first choice?
bbox = result.BoundingBox(RIL.SYMBOL)
if bbox == None:
continue
glyph_id = '%s_glyph%04d' % (word.id, result_no)
log.debug("Recognizing text in glyph '%s'", glyph_id)
glyph = GlyphType(id=glyph_id, Coords=CoordsType(points_from_x0y0x1y1(bbox)))
word.add_Glyph(glyph)
choice_it = result.GetChoiceIterator()
for (choice_no, choice) in enumerate(choice_it):
alternative_symb = choice.GetUTF8Text()
alternative_conf = choice.Confidence()
glyph.add_TextEquiv(TextEquivType(index=choice_no, conf=alternative_conf, Unicode=alternative_symb))
ID = concat_padded(self.output_file_grp, n)
self.add_output_file(
ID=ID,
file_grp=self.output_file_grp,
basename=ID + '.xml',
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts),
)
def process(self):
"""
Performs the (text) recognition.
"""
# print(self.parameter)
log.debug("TESSDATA: %s, installed tesseract models: %s",
*get_languages())
maxlevel = self.parameter['textequiv_level']
model = get_languages()[1][-1] # last installed model
if 'model' in self.parameter:
model = self.parameter['model']
if model not in get_languages()[1]:
raise Exception("configured model " + model +
" is not installed")
with PyTessBaseAPI(path=TESSDATA_PREFIX, lang=model) as tessapi:
log.info("Using model '%s' in %s for recognition at the %s level",
model,
get_languages()[0], maxlevel)
tessapi.SetVariable(
"glyph_confidences",
"2") # populate GetChoiceIterator() with LSTM models, too
# tessapi.SetVariable("tessedit_single_match", "0")
#
# tessedit_load_sublangs
# tessedit_preserve_min_wd_len 2
# tessedit_prefer_joined_punct 0
# tessedit_write_rep_codes 0
# tessedit_parallelize 0
# tessedit_zero_rejection 0
# tessedit_zero_kelvin_rejection 0
# tessedit_reject_mode 0
# tessedit_use_reject_spaces 1
# tessedit_fix_fuzzy_spaces 1
# tessedit_char_blacklist
# tessedit_char_whitelist
# chs_leading_punct ('`"
# chs_trailing_punct1 ).,;:?!
# chs_trailing_punct2 )'`"
# numeric_punctuation .,
# unrecognised_char |
# ok_repeated_ch_non_alphanum_wds -?*=
# conflict_set_I_l_1 Il1[]
# preserve_interword_spaces 0
# tessedit_enable_dict_correction 0
# tessedit_enable_bigram_correction 1
# stopper_smallword_size 2
# wordrec_max_join_chunks 4
# suspect_space_level 100
# suspect_short_words 2
# language_model_ngram_on 0
# language_model_ngram_order 8
# language_model_min_compound_length 3
# language_model_penalty_non_freq_dict_word 0.1
# language_model_penalty_non_dict_word 0.15
# language_model_penalty_punc 0.2
# language_model_penalty_case 0.1
# language_model_penalty_script 0.5
# language_model_penalty_chartype 0.3
# language_model_penalty_spacing 0.05
# textord_max_noise_size 7
# enable_noise_removal 1
# classify_bln_numeric_mode 0
# lstm_use_matrix 1
# user_words_file
# user_patterns_file
for (n, input_file) in enumerate(self.input_files):
log.info("INPUT FILE %i / %s", n, input_file)
pcgts = from_file(self.workspace.download_file(input_file))
# TODO use binarized / gray
pil_image = self.workspace.resolve_image_as_pil(
pcgts.get_Page().imageFilename)
tessapi.SetImage(pil_image)
metadata = pcgts.get_Metadata() # ensured by from_file()
metadata.add_MetadataItem(
MetadataItemType(
type_="processingStep",
name=OCRD_TOOL['tools']['ocrd-tesserocr-recognize']
['steps'][0],
value='ocrd-tesserocr-recognize',
Labels=[
LabelsType(externalRef="parameters",
Label=[
LabelType(
type_=name,
value=self.parameter[name])
for name in self.parameter.keys()
])
]))
# TODO slow
# tessapi.SetPageSegMode(PSM.SINGLE_LINE)
log.info("Recognizing text in page '%s'", pcgts.get_pcGtsId())
page = pcgts.get_Page()
if maxlevel == 'page':
# not sure what to do here:
# - We cannot simply do GetUTF8Text(), because there is no TextEquiv on the page level.
# - We could GetComponentImages(RIL.BLOCK) and add a text region for each, then enter region level recognition below. But what if regions are already annotated? How to go about non-text blocks?
raise Exception(
"currently only implemented below the page level")
## region, line, word, or glyph level:
regions = page.get_TextRegion()
if not regions:
log.warning("Page contains no text regions")
self._process_regions(regions, maxlevel, tessapi)
ID = concat_padded(self.output_file_grp, n)
self.workspace.add_file(
ID=ID,
file_grp=self.output_file_grp,
basename=ID + '.xml',
mimetype=MIMETYPE_PAGE,
content=to_xml(pcgts),
)