|
|
|
@ -10,6 +10,7 @@ import numpy as np
|
|
|
|
from lxml import etree as ET
|
|
|
|
from lxml import etree as ET
|
|
|
|
from ocrd_utils import getLogger
|
|
|
|
from ocrd_utils import getLogger
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class Normalization(enum.Enum):
|
|
|
|
class Normalization(enum.Enum):
|
|
|
|
NFC = 1
|
|
|
|
NFC = 1
|
|
|
|
NFC_MUFI = 2 # TODO
|
|
|
|
NFC_MUFI = 2 # TODO
|
|
|
|
@ -18,7 +19,7 @@ class Normalization(enum.Enum):
|
|
|
|
|
|
|
|
|
|
|
|
def normalize(text, normalization):
|
|
|
|
def normalize(text, normalization):
|
|
|
|
if normalization == Normalization.NFC:
|
|
|
|
if normalization == Normalization.NFC:
|
|
|
|
return unicodedata.normalize('NFC', text)
|
|
|
|
return unicodedata.normalize("NFC", text)
|
|
|
|
if normalization == Normalization.NFC_MUFI:
|
|
|
|
if normalization == Normalization.NFC_MUFI:
|
|
|
|
raise NotImplementedError()
|
|
|
|
raise NotImplementedError()
|
|
|
|
if normalization == Normalization.NFC_SBB:
|
|
|
|
if normalization == Normalization.NFC_SBB:
|
|
|
|
@ -36,31 +37,31 @@ def unjoin_ligatures(s):
|
|
|
|
"""Unjoin ligatures, i.e. ff becomes ff."""
|
|
|
|
"""Unjoin ligatures, i.e. ff becomes ff."""
|
|
|
|
|
|
|
|
|
|
|
|
equivalences = {
|
|
|
|
equivalences = {
|
|
|
|
'': 'ſſ',
|
|
|
|
"": "ſſ",
|
|
|
|
"\ueba7": 'ſſi', # MUFI: LATIN SMALL LIGATURE LONG S LONG S I
|
|
|
|
"\ueba7": "ſſi", # MUFI: LATIN SMALL LIGATURE LONG S LONG S I
|
|
|
|
'': 'ch',
|
|
|
|
"": "ch",
|
|
|
|
'': 'ck',
|
|
|
|
"": "ck",
|
|
|
|
'': 'll',
|
|
|
|
"": "ll",
|
|
|
|
'': 'ſi',
|
|
|
|
"": "ſi",
|
|
|
|
'': 'ſt',
|
|
|
|
"": "ſt",
|
|
|
|
'fi': 'fi',
|
|
|
|
"fi": "fi",
|
|
|
|
'ff': 'ff',
|
|
|
|
"ff": "ff",
|
|
|
|
'fl': 'fl',
|
|
|
|
"fl": "fl",
|
|
|
|
'ffi': 'ffi',
|
|
|
|
"ffi": "ffi",
|
|
|
|
'': 'ct',
|
|
|
|
"": "ct",
|
|
|
|
'': 'tz', # MUFI: LATIN SMALL LIGATURE TZ
|
|
|
|
"": "tz", # MUFI: LATIN SMALL LIGATURE TZ
|
|
|
|
'\uf532': 'as', # eMOP: Latin small ligature as
|
|
|
|
"\uf532": "as", # eMOP: Latin small ligature as
|
|
|
|
'\uf533': 'is', # eMOP: Latin small ligature is
|
|
|
|
"\uf533": "is", # eMOP: Latin small ligature is
|
|
|
|
'\uf534': 'us', # eMOP: Latin small ligature us
|
|
|
|
"\uf534": "us", # eMOP: Latin small ligature us
|
|
|
|
'\uf535': 'Qu', # eMOP: Latin ligature capital Q small u
|
|
|
|
"\uf535": "Qu", # eMOP: Latin ligature capital Q small u
|
|
|
|
'ij': 'ij', # U+0133 LATIN SMALL LIGATURE IJ
|
|
|
|
"ij": "ij", # U+0133 LATIN SMALL LIGATURE IJ
|
|
|
|
'\uE8BF': 'q&',
|
|
|
|
"\uE8BF": "q&",
|
|
|
|
# MUFI: LATIN SMALL LETTER Q LIGATED WITH FINAL ET
|
|
|
|
# MUFI: LATIN SMALL LETTER Q LIGATED WITH FINAL ET
|
|
|
|
# XXX How to replace this correctly?
|
|
|
|
# XXX How to replace this correctly?
|
|
|
|
'\uEBA5': 'ſp', # MUFI: LATIN SMALL LIGATURE LONG S P
|
|
|
|
"\uEBA5": "ſp", # MUFI: LATIN SMALL LIGATURE LONG S P
|
|
|
|
'st': 'st', # U+FB06 LATIN SMALL LIGATURE ST
|
|
|
|
"st": "st", # U+FB06 LATIN SMALL LIGATURE ST
|
|
|
|
}
|
|
|
|
}
|
|
|
|
s = unicodedata.normalize('NFC', s)
|
|
|
|
s = unicodedata.normalize("NFC", s)
|
|
|
|
for fr, to in equivalences.items():
|
|
|
|
for fr, to in equivalences.items():
|
|
|
|
s = s.replace(fr, to)
|
|
|
|
s = s.replace(fr, to)
|
|
|
|
return s
|
|
|
|
return s
|
|
|
|
@ -70,20 +71,20 @@ def substitute_equivalences(s):
|
|
|
|
# These are for OCR-D GT vs Tesseract frk vs Calamari GT4HistOCR
|
|
|
|
# These are for OCR-D GT vs Tesseract frk vs Calamari GT4HistOCR
|
|
|
|
# It might make sense to use different rules for GT and for the different OCR
|
|
|
|
# It might make sense to use different rules for GT and for the different OCR
|
|
|
|
equivalences = {
|
|
|
|
equivalences = {
|
|
|
|
'': 'ü',
|
|
|
|
"": "ü",
|
|
|
|
'': 'ä',
|
|
|
|
"": "ä",
|
|
|
|
'==': '–', # → en-dash
|
|
|
|
"==": "–", # → en-dash
|
|
|
|
'—': '–', # em-dash → en-dash
|
|
|
|
"—": "–", # em-dash → en-dash
|
|
|
|
'': 'ö',
|
|
|
|
"": "ö",
|
|
|
|
'’': '\'',
|
|
|
|
"’": "'",
|
|
|
|
'⸗': '-',
|
|
|
|
"⸗": "-",
|
|
|
|
'aͤ': 'ä', # LATIN SMALL LETTER A, COMBINING LATIN SMALL LETTER E
|
|
|
|
"aͤ": "ä", # LATIN SMALL LETTER A, COMBINING LATIN SMALL LETTER E
|
|
|
|
'oͤ': 'ö', # LATIN SMALL LETTER O, COMBINING LATIN SMALL LETTER E
|
|
|
|
"oͤ": "ö", # LATIN SMALL LETTER O, COMBINING LATIN SMALL LETTER E
|
|
|
|
'uͤ': 'ü', # LATIN SMALL LETTER U, COMBINING LATIN SMALL LETTER E
|
|
|
|
"uͤ": "ü", # LATIN SMALL LETTER U, COMBINING LATIN SMALL LETTER E
|
|
|
|
'\uF50E': 'q́' # U+F50E LATIN SMALL LETTER Q WITH ACUTE ACCENT
|
|
|
|
"\uF50E": "q́", # U+F50E LATIN SMALL LETTER Q WITH ACUTE ACCENT
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
s = unicodedata.normalize('NFC', s)
|
|
|
|
s = unicodedata.normalize("NFC", s)
|
|
|
|
s = unjoin_ligatures(s)
|
|
|
|
s = unjoin_ligatures(s)
|
|
|
|
for fr, to in equivalences.items():
|
|
|
|
for fr, to in equivalences.items():
|
|
|
|
s = s.replace(fr, to)
|
|
|
|
s = s.replace(fr, to)
|
|
|
|
@ -115,13 +116,14 @@ class ExtractedText:
|
|
|
|
Objects of this class are guaranteed to be a. always in their normalization
|
|
|
|
Objects of this class are guaranteed to be a. always in their normalization
|
|
|
|
and b. in NFC.
|
|
|
|
and b. in NFC.
|
|
|
|
"""
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
|
|
segment_id = attr.ib(type=Optional[str])
|
|
|
|
segment_id = attr.ib(type=Optional[str])
|
|
|
|
|
|
|
|
|
|
|
|
@segment_id.validator
|
|
|
|
@segment_id.validator
|
|
|
|
def check(self, _, value):
|
|
|
|
def check(self, _, value):
|
|
|
|
if value is None:
|
|
|
|
if value is None:
|
|
|
|
return
|
|
|
|
return
|
|
|
|
if not re.match(r'[\w\d_-]+', value):
|
|
|
|
if not re.match(r"[\w\d_-]+", value):
|
|
|
|
raise ValueError('Malformed segment id "{}"'.format(value))
|
|
|
|
raise ValueError('Malformed segment id "{}"'.format(value))
|
|
|
|
|
|
|
|
|
|
|
|
# An object contains either
|
|
|
|
# An object contains either
|
|
|
|
@ -141,7 +143,7 @@ class ExtractedText:
|
|
|
|
def check(self, _, value):
|
|
|
|
def check(self, _, value):
|
|
|
|
if value is not None and self.segments is not None:
|
|
|
|
if value is not None and self.segments is not None:
|
|
|
|
raise ValueError("Can't have both segments and text")
|
|
|
|
raise ValueError("Can't have both segments and text")
|
|
|
|
if value is not None and unicodedata.normalize('NFC', value) != value:
|
|
|
|
if value is not None and unicodedata.normalize("NFC", value) != value:
|
|
|
|
raise ValueError('String "{}" is not in NFC.'.format(value))
|
|
|
|
raise ValueError('String "{}" is not in NFC.'.format(value))
|
|
|
|
if value is not None and normalize(value, self.normalization) != value:
|
|
|
|
if value is not None and normalize(value, self.normalization) != value:
|
|
|
|
raise ValueError('String "{}" is not normalized.'.format(value))
|
|
|
|
raise ValueError('String "{}" is not normalized.'.format(value))
|
|
|
|
@ -169,31 +171,24 @@ class ExtractedText:
|
|
|
|
seg_ids = [s.segment_id_for_pos(i) for i in range(len(s.text))]
|
|
|
|
seg_ids = [s.segment_id_for_pos(i) for i in range(len(s.text))]
|
|
|
|
segment_id_for_pos.extend(seg_ids)
|
|
|
|
segment_id_for_pos.extend(seg_ids)
|
|
|
|
segment_id_for_pos.extend(repeat(None, len(self.joiner)))
|
|
|
|
segment_id_for_pos.extend(repeat(None, len(self.joiner)))
|
|
|
|
segment_id_for_pos = segment_id_for_pos[:-len(self.joiner)]
|
|
|
|
segment_id_for_pos = segment_id_for_pos[: -len(self.joiner)]
|
|
|
|
|
|
|
|
|
|
|
|
# This is frozen, so we have to jump through the hoop:
|
|
|
|
# This is frozen, so we have to jump through the hoop:
|
|
|
|
object.__setattr__(self, '_segment_id_for_pos', segment_id_for_pos)
|
|
|
|
object.__setattr__(self, "_segment_id_for_pos", segment_id_for_pos)
|
|
|
|
assert self._segment_id_for_pos
|
|
|
|
assert self._segment_id_for_pos
|
|
|
|
|
|
|
|
|
|
|
|
return self._segment_id_for_pos[pos]
|
|
|
|
return self._segment_id_for_pos[pos]
|
|
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
@classmethod
|
|
|
|
def from_text_segment(cls, text_segment, nsmap, textequiv_level='region'):
|
|
|
|
def from_text_segment(cls, text_segment, nsmap, textequiv_level="region"):
|
|
|
|
"""Build an ExtractedText from a PAGE content text element"""
|
|
|
|
"""Build an ExtractedText from a PAGE content text element"""
|
|
|
|
|
|
|
|
|
|
|
|
localname_for_textequiv_level = {
|
|
|
|
localname_for_textequiv_level = {"region": "TextRegion", "line": "TextLine"}
|
|
|
|
'region': 'TextRegion',
|
|
|
|
|
|
|
|
'line': 'TextLine'
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
textequiv_level_for_localname = invert_dict(localname_for_textequiv_level)
|
|
|
|
textequiv_level_for_localname = invert_dict(localname_for_textequiv_level)
|
|
|
|
children_for_localname = {
|
|
|
|
children_for_localname = {"TextRegion": "TextLine"}
|
|
|
|
'TextRegion': 'TextLine'
|
|
|
|
joiner_for_textequiv_level = {"line": "\n"}
|
|
|
|
}
|
|
|
|
|
|
|
|
joiner_for_textequiv_level = {
|
|
|
|
|
|
|
|
'line': '\n'
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
segment_id = text_segment.attrib['id']
|
|
|
|
segment_id = text_segment.attrib["id"]
|
|
|
|
localname = ET.QName(text_segment).localname
|
|
|
|
localname = ET.QName(text_segment).localname
|
|
|
|
if localname == localname_for_textequiv_level[textequiv_level]:
|
|
|
|
if localname == localname_for_textequiv_level[textequiv_level]:
|
|
|
|
segment_text = None
|
|
|
|
segment_text = None
|
|
|
|
@ -201,19 +196,20 @@ class ExtractedText:
|
|
|
|
segment_text = get_textequiv_unicode(text_segment, nsmap)
|
|
|
|
segment_text = get_textequiv_unicode(text_segment, nsmap)
|
|
|
|
# FIXME hardcoded SBB normalization
|
|
|
|
# FIXME hardcoded SBB normalization
|
|
|
|
segment_text = normalize_sbb(segment_text)
|
|
|
|
segment_text = normalize_sbb(segment_text)
|
|
|
|
segment_text = segment_text or ''
|
|
|
|
segment_text = segment_text or ""
|
|
|
|
return cls(segment_id, None, None, segment_text)
|
|
|
|
return cls(segment_id, None, None, segment_text)
|
|
|
|
else:
|
|
|
|
else:
|
|
|
|
# Recurse
|
|
|
|
# Recurse
|
|
|
|
sub_localname = children_for_localname[localname]
|
|
|
|
sub_localname = children_for_localname[localname]
|
|
|
|
sub_textequiv_level = textequiv_level_for_localname[sub_localname]
|
|
|
|
sub_textequiv_level = textequiv_level_for_localname[sub_localname]
|
|
|
|
segments = []
|
|
|
|
segments = []
|
|
|
|
for sub_segment in text_segment.iterfind('./page:%s' % sub_localname,
|
|
|
|
for sub_segment in text_segment.iterfind(
|
|
|
|
namespaces=nsmap):
|
|
|
|
"./page:%s" % sub_localname, namespaces=nsmap
|
|
|
|
|
|
|
|
):
|
|
|
|
segments.append(
|
|
|
|
segments.append(
|
|
|
|
ExtractedText.from_text_segment(
|
|
|
|
ExtractedText.from_text_segment(
|
|
|
|
sub_segment, nsmap,
|
|
|
|
sub_segment, nsmap, textequiv_level=sub_textequiv_level
|
|
|
|
textequiv_level=sub_textequiv_level)
|
|
|
|
)
|
|
|
|
)
|
|
|
|
)
|
|
|
|
joiner = joiner_for_textequiv_level[sub_textequiv_level]
|
|
|
|
joiner = joiner_for_textequiv_level[sub_textequiv_level]
|
|
|
|
return cls(segment_id, segments, joiner, None)
|
|
|
|
return cls(segment_id, segments, joiner, None)
|
|
|
|
@ -231,24 +227,24 @@ def invert_dict(d):
|
|
|
|
|
|
|
|
|
|
|
|
def get_textequiv_unicode(text_segment, nsmap) -> str:
|
|
|
|
def get_textequiv_unicode(text_segment, nsmap) -> str:
|
|
|
|
"""Get the TextEquiv/Unicode text of the given PAGE text element."""
|
|
|
|
"""Get the TextEquiv/Unicode text of the given PAGE text element."""
|
|
|
|
segment_id = text_segment.attrib['id']
|
|
|
|
segment_id = text_segment.attrib["id"]
|
|
|
|
textequivs = text_segment.findall('./page:TextEquiv', namespaces=nsmap)
|
|
|
|
textequivs = text_segment.findall("./page:TextEquiv", namespaces=nsmap)
|
|
|
|
|
|
|
|
|
|
|
|
if not textequivs:
|
|
|
|
if not textequivs:
|
|
|
|
return ''
|
|
|
|
return ""
|
|
|
|
|
|
|
|
|
|
|
|
textequiv = get_first_textequiv(textequivs, segment_id)
|
|
|
|
textequiv = get_first_textequiv(textequivs, segment_id)
|
|
|
|
return textequiv.find('./page:Unicode', namespaces=nsmap).text or ''
|
|
|
|
return textequiv.find("./page:Unicode", namespaces=nsmap).text or ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_first_textequiv(textequivs, segment_id):
|
|
|
|
def get_first_textequiv(textequivs, segment_id):
|
|
|
|
"""Get the first TextEquiv based on index or conf order if index is not present."""
|
|
|
|
"""Get the first TextEquiv based on index or conf order if index is not present."""
|
|
|
|
log = getLogger('processor.OcrdDinglehopperEvaluate')
|
|
|
|
log = getLogger("processor.OcrdDinglehopperEvaluate")
|
|
|
|
if len(textequivs) == 1:
|
|
|
|
if len(textequivs) == 1:
|
|
|
|
return textequivs[0]
|
|
|
|
return textequivs[0]
|
|
|
|
|
|
|
|
|
|
|
|
# try ordering by index
|
|
|
|
# try ordering by index
|
|
|
|
indices = np.array([get_attr(te, 'index') for te in textequivs], dtype=float)
|
|
|
|
indices = np.array([get_attr(te, "index") for te in textequivs], dtype=float)
|
|
|
|
nan_mask = np.isnan(indices)
|
|
|
|
nan_mask = np.isnan(indices)
|
|
|
|
if np.any(~nan_mask):
|
|
|
|
if np.any(~nan_mask):
|
|
|
|
if np.any(nan_mask):
|
|
|
|
if np.any(nan_mask):
|
|
|
|
@ -256,10 +252,12 @@ def get_first_textequiv(textequivs, segment_id):
|
|
|
|
index = np.nanargmin(indices)
|
|
|
|
index = np.nanargmin(indices)
|
|
|
|
else:
|
|
|
|
else:
|
|
|
|
# try ordering by conf
|
|
|
|
# try ordering by conf
|
|
|
|
confidences = np.array([get_attr(te, 'conf') for te in textequivs], dtype=float)
|
|
|
|
confidences = np.array([get_attr(te, "conf") for te in textequivs], dtype=float)
|
|
|
|
if np.any(~np.isnan(confidences)):
|
|
|
|
if np.any(~np.isnan(confidences)):
|
|
|
|
log.info("No index attributes, use 'conf' attribute to sort TextEquiv in %s.",
|
|
|
|
log.info(
|
|
|
|
segment_id)
|
|
|
|
"No index attributes, use 'conf' attribute to sort TextEquiv in %s.",
|
|
|
|
|
|
|
|
segment_id,
|
|
|
|
|
|
|
|
)
|
|
|
|
index = np.nanargmax(confidences)
|
|
|
|
index = np.nanargmax(confidences)
|
|
|
|
else:
|
|
|
|
else:
|
|
|
|
# fallback to first entry in case of neither index or conf present
|
|
|
|
# fallback to first entry in case of neither index or conf present
|
|
|
|
|
