anglk/WordMord
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

new tool metallaksis

Browse Source
This commit is contained in:
Angeliki 2022-01-31 12:16:13 -06:00
parent bf438eb5be
commit 7c0acf3b81
5 changed files with 1163 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@ -1,3 +1,4 @@
pdf/* pdf/*
images/* images/*
audio/* audio/*
scripts/venv/*

500
scripts/hocrtransform.py Executable file
View File

@ -0,0 +1,500 @@
#!/usr/bin/env python3
#
# Copyright (c) 2010, Jonathan Brinley
# Original version from: https://github.com/jbrinley/HocrConverter
#
# Copyright (c) 2013-14, Julien Pfefferkorn
# Modifications
#
# Copyright (c) 2015-16, James R. Barlow
# Set text to transparent
#
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the
# "Software"), to deal in the Software without restriction, including
# without limitation the rights to use, copy, modify, merge, publish,
# distribute, sublicense, and/or sell copies of the Software, and to
# permit persons to whom the Software is furnished to do so, subject to
# the following conditions:
#
# The above copyright notice and this permission notice shall be included
# in all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
# CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
# SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
import argparse
import os
import re
from itertools import chain
from math import atan, cos, sin
from pathlib import Path
from typing import Any, NamedTuple, Optional, Tuple, Union
from xml.etree import ElementTree
from reportlab.lib.colors import black, cyan, magenta, red
from reportlab.lib.units import inch
from reportlab.pdfgen.canvas import Canvas
# According to Wikipedia these languages are supported in the ISO-8859-1 character
# set, meaning reportlab can generate them and they are compatible with hocr,
# assuming Tesseract has the necessary languages installed. Note that there may
# not be language packs for them.
HOCR_OK_LANGS = frozenset(
[
# Languages fully covered by Latin-1:
'afr', # Afrikaans
'alb', # Albanian
'ast', # Leonese
'baq', # Basque
'bre', # Breton
'cos', # Corsican
'eng', # English
'eus', # Basque
'fao', # Faoese
'gla', # Scottish Gaelic
'glg', # Galician
'glv', # Manx
'ice', # Icelandic
'ind', # Indonesian
'isl', # Icelandic
'ita', # Italian
'ltz', # Luxembourgish
'mal', # Malay Rumi
'mga', # Irish
'nor', # Norwegian
'oci', # Occitan
'por', # Portugeuse
'roh', # Romansh
'sco', # Scots
'sma', # Sami
'spa', # Spanish
'sqi', # Albanian
'swa', # Swahili
'swe', # Swedish
'tgl', # Tagalog
'wln', # Walloon
# Languages supported by Latin-1 except for a few rare characters that OCR
# is probably not trained to recognize anyway:
'cat', # Catalan
'cym', # Welsh
'dan', # Danish
'deu', # German
'dut', # Dutch
'est', # Estonian
'fin', # Finnish
'fra', # French
'hun', # Hungarian
'kur', # Kurdish
'nld', # Dutch
'wel', # Welsh
]
)
Element = ElementTree.Element
class Rect(NamedTuple): # pylint: disable=inherit-non-class
"""A rectangle for managing PDF coordinates."""
x1: Any
y1: Any
x2: Any
y2: Any
class HocrTransformError(Exception):
pass
class HocrTransform:
"""
A class for converting documents from the hOCR format.
For details of the hOCR format, see:
http://kba.cloud/hocr-spec/
"""
box_pattern = re.compile(r'bbox((\s+\d+){4})')
baseline_pattern = re.compile(
r'''
baseline \s+
([\-\+]?\d*\.?\d*) \s+ # +/- decimal float
([\-\+]?\d+) # +/- int''',
re.VERBOSE,
)
ligatures = str.maketrans(
{'ff': 'ff', 'ffi': 'f‌f‌i', 'ffl': 'f‌f‌l', 'fi': 'fi', 'fl': 'fl'}
)
def __init__(self, *, hocr_filename: Union[str, Path], dpi: float):
self.dpi = dpi
self.hocr = ElementTree.parse(os.fspath(hocr_filename))
# if the hOCR file has a namespace, ElementTree requires its use to
# find elements
matches = re.match(r'({.*})html', self.hocr.getroot().tag)
self.xmlns = ''
if matches:
self.xmlns = matches.group(1)
# get dimension in pt (not pixel!!!!) of the OCRed image
self.width, self.height = None, None
for div in self.hocr.findall(self._child_xpath('div', 'ocr_page')):
coords = self.element_coordinates(div)
pt_coords = self.pt_from_pixel(coords)
self.width = pt_coords.x2 - pt_coords.x1
self.height = pt_coords.y2 - pt_coords.y1
# there shouldn't be more than one, and if there is, we don't want
# it
break
if self.width is None or self.height is None:
raise HocrTransformError("hocr file is missing page dimensions")
def __str__(self): # pragma: no cover
"""
Return the textual content of the HTML body
"""
if self.hocr is None:
return ''
body = self.hocr.find(self._child_xpath('body'))
if body:
return self._get_element_text(body)
else:
return ''
def _get_element_text(self, element: Element):
"""
Return the textual content of the element and its children
"""
text = ''
if element.text is not None:
text += element.text
for child in element:
text += self._get_element_text(child)
if element.tail is not None:
text += element.tail
return text
@classmethod
def element_coordinates(cls, element: Element) -> Rect:
"""
Returns a tuple containing the coordinates of the bounding box around
an element
"""
out = Rect._make(0 for _ in range(4))
if 'title' in element.attrib:
matches = cls.box_pattern.search(element.attrib['title'])
if matches:
coords = matches.group(1).split()
out = Rect._make(int(coords[n]) for n in range(4))
return out
@classmethod
def baseline(cls, element: Element) -> Tuple[float, float]:
"""
Returns a tuple containing the baseline slope and intercept.
"""
if 'title' in element.attrib:
matches = cls.baseline_pattern.search(element.attrib['title'])
if matches:
return float(matches.group(1)), int(matches.group(2))
return (0.0, 0.0)
def pt_from_pixel(self, pxl) -> Rect:
"""
Returns the quantity in PDF units (pt) given quantity in pixels
"""
return Rect._make((c / self.dpi * inch) for c in pxl)
def _child_xpath(self, html_tag: str, html_class: Optional[str] = None) -> str:
xpath = f".//{self.xmlns}{html_tag}"
if html_class:
xpath += f"[@class='{html_class}']"
return xpath
@classmethod
def replace_unsupported_chars(cls, s: str) -> str:
"""
Given an input string, returns the corresponding string that:
* is available in the Helvetica facetype
* does not contain any ligature (to allow easy search in the PDF file)
"""
return s.translate(cls.ligatures)
def topdown_position(self, element):
pxl_line_coords = self.element_coordinates(element)
line_box = self.pt_from_pixel(pxl_line_coords)
# Coordinates here are still in the hocr coordinate system, so 0 on the y axis
# is the top of the page and increasing values of y will move towards the
# bottom of the page.
return line_box.y2
def to_pdf(
self,
*,
out_filename: Path,
image_filename: Optional[Path] = None,
show_bounding_boxes: bool = False,
fontname: str = "Helvetica",
invisible_text: bool = False,
interword_spaces: bool = False,
) -> None:
"""
Creates a PDF file with an image superimposed on top of the text.
Text is positioned according to the bounding box of the lines in
the hOCR file.
The image need not be identical to the image used to create the hOCR
file.
It can have a lower resolution, different color mode, etc.
Arguments:
out_filename: Path of PDF to write.
image_filename: Image to use for this file. If omitted, the OCR text
is shown.
show_bounding_boxes: Show bounding boxes around various text regions,
for debugging.
fontname: Name of font to use.
invisible_text: If True, text is rendered invisible so that is
selectable but never drawn. If False, text is visible and may
be seen if the image is skipped or deleted in Acrobat.
interword_spaces: If True, insert spaces between words rather than
drawing each word without spaces. Generally this improves text
extraction.
"""
# create the PDF file
# page size in points (1/72 in.)
pdf = Canvas(
os.fspath(out_filename),
pagesize=(self.width, self.height),
pageCompression=1,
)
# draw bounding box for each paragraph
# light blue for bounding box of paragraph
pdf.setStrokeColor(cyan)
# light blue for bounding box of paragraph
pdf.setFillColor(cyan)
pdf.setLineWidth(0) # no line for bounding box
for elem in self.hocr.iterfind(self._child_xpath('p', 'ocr_par')):
elemtxt = self._get_element_text(elem).rstrip()
if len(elemtxt) == 0:
continue
pxl_coords = self.element_coordinates(elem)
pt = self.pt_from_pixel(pxl_coords)
# draw the bbox border
if show_bounding_boxes: # pragma: no cover
pdf.rect(
pt.x1, self.height - pt.y2, pt.x2 - pt.x1, pt.y2 - pt.y1, fill=1
)
found_lines = False
for line in sorted(
chain(
self.hocr.iterfind(self._child_xpath('span', 'ocr_header')),
self.hocr.iterfind(self._child_xpath('span', 'ocr_line')),
self.hocr.iterfind(self._child_xpath('span', 'ocr_textfloat')),
),
key=self.topdown_position,
):
found_lines = True
self._do_line(
pdf,
line,
"ocrx_word",
fontname,
invisible_text,
interword_spaces,
show_bounding_boxes,
)
if not found_lines:
# Tesseract did not report any lines (just words)
root = self.hocr.find(self._child_xpath('div', 'ocr_page'))
self._do_line(
pdf,
root,
"ocrx_word",
fontname,
invisible_text,
interword_spaces,
show_bounding_boxes,
)
# put the image on the page, scaled to fill the page
if image_filename is not None:
pdf.drawImage(
os.fspath(image_filename), 0, 0, width=self.width, height=self.height
)
# finish up the page and save it
pdf.showPage()
pdf.save()
@classmethod
def polyval(cls, poly, x): # pragma: no cover
return x * poly[0] + poly[1]
def _do_line(
self,
pdf: Canvas,
line: Optional[Element],
elemclass: str,
fontname: str,
invisible_text: bool,
interword_spaces: bool,
show_bounding_boxes: bool,
):
if not line:
return
pxl_line_coords = self.element_coordinates(line)
line_box = self.pt_from_pixel(pxl_line_coords)
line_height = line_box.y2 - line_box.y1
slope, pxl_intercept = self.baseline(line)
if abs(slope) < 0.005:
slope = 0.0
angle = atan(slope)
cos_a, sin_a = cos(angle), sin(angle)
text = pdf.beginText()
intercept = pxl_intercept / self.dpi * inch
# Don't allow the font to break out of the bounding box. Division by
# cos_a accounts for extra clearance between the glyph's vertical axis
# on a sloped baseline and the edge of the bounding box.
fontsize = (line_height - abs(intercept)) / cos_a
text.setFont(fontname, fontsize)
if invisible_text:
text.setTextRenderMode(3) # Invisible (indicates OCR text)
# Intercept is normally negative, so this places it above the bottom
# of the line box
baseline_y2 = self.height - (line_box.y2 + intercept)
if show_bounding_boxes: # pragma: no cover
# draw the baseline in magenta, dashed
pdf.setDash()
pdf.setStrokeColor(magenta)
pdf.setLineWidth(0.5)
# negate slope because it is defined as a rise/run in pixel
# coordinates and page coordinates have the y axis flipped
pdf.line(
line_box.x1,
baseline_y2,
line_box.x2,
self.polyval((-slope, baseline_y2), line_box.x2 - line_box.x1),
)
# light green for bounding box of word/line
pdf.setDash(6, 3)
pdf.setStrokeColor(red)
text.setTextTransform(cos_a, -sin_a, sin_a, cos_a, line_box.x1, baseline_y2)
pdf.setFillColor(black) # text in black
elements = line.findall(self._child_xpath('span', elemclass))
for elem in elements:
elemtxt = self._get_element_text(elem).strip()
elemtxt = self.replace_unsupported_chars(elemtxt)
if elemtxt == '':
continue
pxl_coords = self.element_coordinates(elem)
box = self.pt_from_pixel(pxl_coords)
if interword_spaces:
# if `--interword-spaces` is true, append a space
# to the end of each text element to allow simpler PDF viewers
# such as PDF.js to better recognize words in search and copy
# and paste. Do not remove space from last word in line, even
# though it would look better, because it will interfere with
# naive text extraction. \n does not work either.
elemtxt += ' '
box = Rect._make(
(
box.x1,
line_box.y1,
box.x2 + pdf.stringWidth(' ', fontname, line_height),
line_box.y2,
)
)
box_width = box.x2 - box.x1
font_width = pdf.stringWidth(elemtxt, fontname, fontsize)
# draw the bbox border
if show_bounding_boxes: # pragma: no cover
pdf.rect(
box.x1, self.height - line_box.y2, box_width, line_height, fill=0
)
# Adjust relative position of cursor
# This is equivalent to:
# text.setTextOrigin(pt.x1, self.height - line_box.y2)
# but the former generates a full text reposition matrix (Tm) in the
# content stream while this issues a "offset" (Td) command.
# .moveCursor() is relative to start of the text line, where the
# "text line" means whatever reportlab defines it as. Do not use
# use .getCursor(), since moveCursor() rather unintuitively plans
# its moves relative to .getStartOfLine().
# For skewed lines, in the text transform we set up a rotated
# coordinate system, so we don't have to account for the
# incremental offset. Surprisingly most PDF viewers can handle this.
cursor = text.getStartOfLine()
dx = box.x1 - cursor[0]
dy = baseline_y2 - cursor[1]
text.moveCursor(dx, dy)
# If reportlab tells us this word is 0 units wide, our best seems
# to be to suppress this text
if font_width > 0:
text.setHorizScale(100 * box_width / font_width)
text.textOut(elemtxt)
pdf.drawText(text)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Convert hocr file to PDF')
parser.add_argument(
'-b',
'--boundingboxes',
action="store_true",
default=False,
help='Show bounding boxes borders',
)
parser.add_argument(
'-r',
'--resolution',
type=int,
default=300,
help='Resolution of the image that was OCRed',
)
parser.add_argument(
'-i',
'--image',
default=None,
help='Path to the image to be placed above the text',
)
parser.add_argument(
'--interword-spaces',
action='store_true',
default=False,
help='Add spaces between words',
)
parser.add_argument('hocrfile', help='Path to the hocr file to be parsed')
parser.add_argument('outputfile', help='Path to the PDF file to be generated')
args = parser.parse_args()
hocr = HocrTransform(hocr_filename=args.hocrfile, dpi=args.resolution)
hocr.to_pdf(
out_filename=args.outputfile,
image_filename=args.image,
show_bounding_boxes=args.boundingboxes,
interword_spaces=args.interword_spaces,
)

518
scripts/hocrtransformpdf.py Executable file
View File

@ -0,0 +1,518 @@
#!venv/bin python3
#
# Copyright (c) 2010, Jonathan Brinley
# Original version from: https://github.com/jbrinley/HocrConverter
#
# Copyright (c) 2013-14, Julien Pfefferkorn
# Modifications
#
# Copyright (c) 2015-16, James R. Barlow
# Set text to transparent
# Copyright (c) 2022, WordMord & Alex Roidl
# Set text back to visible and change bounding boxes
#
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the
# "Software"), to deal in the Software without restriction, including
# without limitation the rights to use, copy, modify, merge, publish,
# distribute, sublicense, and/or sell copies of the Software, and to
# permit persons to whom the Software is furnished to do so, subject to
# the following conditions:
#
# The above copyright notice and this permission notice shall be included
# in all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
# CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
# SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
import argparse
import os
import re
from itertools import chain
from math import atan, cos, sin
from pathlib import Path
from typing import Any, NamedTuple, Optional, Tuple, Union
from xml.etree import ElementTree
from reportlab.lib.colors import black, cyan, magenta, red
from reportlab.lib.units import inch
from reportlab.pdfgen.canvas import Canvas
from reportlab.pdfbase import pdfmetrics
from reportlab.pdfbase.ttfonts import TTFont
from reportlab.pdfbase.pdfmetrics import registerFontFamily
# According to Wikipedia these languages are supported in the ISO-8859-1 character
# set, meaning reportlab can generate them and they are compatible with hocr,
# assuming Tesseract has the necessary languages installed. Note that there may
# not be language packs for them.
HOCR_OK_LANGS = frozenset(
[
# Languages fully covered by Latin-1:
'afr', # Afrikaans
'alb', # Albanian
'ast', # Leonese
'baq', # Basque
'bre', # Breton
'cos', # Corsican
'eng', # English
'eus', # Basque
'fao', # Faoese
'gla', # Scottish Gaelic
'glg', # Galician
'glv', # Manx
'ice', # Icelandic
'ind', # Indonesian
'isl', # Icelandic
'ita', # Italian
'ltz', # Luxembourgish
'mal', # Malay Rumi
'mga', # Irish
'nor', # Norwegian
'oci', # Occitan
'por', # Portugeuse
'roh', # Romansh
'sco', # Scots
'sma', # Sami
'spa', # Spanish
'sqi', # Albanian
'swa', # Swahili
'swe', # Swedish
'tgl', # Tagalog
'wln', # Walloon
# Languages supported by Latin-1 except for a few rare characters that OCR
# is probably not trained to recognize anyway:
'cat', # Catalan
'cym', # Welsh
'dan', # Danish
'deu', # German
'dut', # Dutch
'est', # Estonian
'fin', # Finnish
'fra', # French
'hun', # Hungarian
'kur', # Kurdish
'nld', # Dutch
'wel', # Welsh
]
)
Element = ElementTree.Element
class Rect(NamedTuple): # pylint: disable=inherit-non-class
"""A rectangle for managing PDF coordinates."""
x1: Any
y1: Any
x2: Any
y2: Any
class HocrTransformError(Exception):
pass
class HocrTransform:
"""
A class for converting documents from the hOCR format.
For details of the hOCR format, see:
http://kba.cloud/hocr-spec/
"""
box_pattern = re.compile(r'bbox((\s+\d+){4})')
baseline_pattern = re.compile(
r'''
baseline \s+
([\-\+]?\d*\.?\d*) \s+ # +/- decimal float
([\-\+]?\d+) # +/- int''',
re.VERBOSE,
)
ligatures = str.maketrans(
{'ff': 'ff', 'ffi': 'f‌f‌i', 'ffl': 'f‌f‌l', 'fi': 'fi', 'fl': 'fl'}
)
def __init__(self, *, hocr_filename: Union[str, Path], dpi: float):
self.dpi = dpi
self.hocr = ElementTree.parse(os.fspath(hocr_filename))
# if the hOCR file has a namespace, ElementTree requires its use to
# find elements
matches = re.match(r'({.*})html', self.hocr.getroot().tag)
self.xmlns = ''
if matches:
self.xmlns = matches.group(1)
# get dimension in pt (not pixel!!!!) of the OCRed image
self.width, self.height = None, None
for div in self.hocr.findall(self._child_xpath('div', 'ocr_page')):
coords = self.element_coordinates(div)
pt_coords = self.pt_from_pixel(coords)
self.width = pt_coords.x2 - pt_coords.x1
self.height = pt_coords.y2 - pt_coords.y1
# there shouldn't be more than one, and if there is, we don't want
# it
break
if self.width is None or self.height is None:
raise HocrTransformError("hocr file is missing page dimensions")
def __str__(self): # pragma: no cover
"""
Return the textual content of the HTML body
"""
if self.hocr is None:
return ''
body = self.hocr.find(self._child_xpath('body'))
if body:
return self._get_element_text(body)
else:
return ''
def _get_element_text(self, element: Element):
"""
Return the textual content of the element and its children
"""
text = ''
if element.text is not None:
text += element.text
for child in element:
text += self._get_element_text(child)
if element.tail is not None:
text += element.tail
return text
@classmethod
def element_coordinates(cls, element: Element) -> Rect:
"""
Returns a tuple containing the coordinates of the bounding box around
an element
"""
out = Rect._make(0 for _ in range(4))
if 'title' in element.attrib:
matches = cls.box_pattern.search(element.attrib['title'])
if matches:
coords = matches.group(1).split()
out = Rect._make(int(coords[n]) for n in range(4))
return out
@classmethod
def baseline(cls, element: Element) -> Tuple[float, float]:
"""
Returns a tuple containing the baseline slope and intercept.
"""
if 'title' in element.attrib:
matches = cls.baseline_pattern.search(element.attrib['title'])
if matches:
return float(matches.group(1)), int(matches.group(2))
return (0.0, 0.0)
def pt_from_pixel(self, pxl) -> Rect:
"""
Returns the quantity in PDF units (pt) given quantity in pixels
"""
return Rect._make((c / self.dpi * inch) for c in pxl)
def _child_xpath(self, html_tag: str, html_class: Optional[str] = None) -> str:
xpath = f".//{self.xmlns}{html_tag}"
if html_class:
xpath += f"[@class='{html_class}']"
return xpath
@classmethod
def replace_unsupported_chars(cls, s: str) -> str:
"""
Given an input string, returns the corresponding string that:
* is available in the Helvetica facetype
* does not contain any ligature (to allow easy search in the PDF file)
"""
return s.translate(cls.ligatures)
def topdown_position(self, element):
pxl_line_coords = self.element_coordinates(element)
line_box = self.pt_from_pixel(pxl_line_coords)
# Coordinates here are still in the hocr coordinate system, so 0 on the y axis
# is the top of the page and increasing values of y will move towards the
# bottom of the page.
return line_box.y2
def to_pdf(
self,
*,
out_filename: Path,
image_filename: Optional[Path] = None,
show_bounding_boxes: bool = False,
fontname: str = "Helvetica",
invisible_text: bool = False,
interword_spaces: bool = False,
) -> None:
"""
Creates a PDF file with an image superimposed on top of the text.
Text is positioned according to the bounding box of the lines in
the hOCR file.
The image need not be identical to the image used to create the hOCR
file.
It can have a lower resolution, different color mode, etc.
Arguments:
out_filename: Path of PDF to write.
image_filename: Image to use for this file. If omitted, the OCR text
is shown.
show_bounding_boxes: Show bounding boxes around various text regions,
for debugging.
fontname: Name of font to use.
invisible_text: If True, text is rendered invisible so that is
selectable but never drawn. If False, text is visible and may
be seen if the image is skipped or deleted in Acrobat.
interword_spaces: If True, insert spaces between words rather than
drawing each word without spaces. Generally this improves text
extraction.
"""
# create the PDF file
# page size in points (1/72 in.)
pdfmetrics.registerFont(TTFont('Greek', '../styles/fonts/greek.ttf'))
pdfmetrics.registerFont(TTFont('GreekB', '../styles/fonts/greek-bold.ttf'))
registerFontFamily('Greek', normal='Greek', bold='GreekB')
pdf = Canvas(
os.fspath(out_filename),
pagesize=(self.width, self.height),
pageCompression=1,
)
if image_filename is not None:
pdf.drawImage(
os.fspath(image_filename), 0, 0, width=self.width, height=self.height
)
# draw bounding box for each paragraph
# light blue for bounding box of paragraph
pdf.setStrokeColor(black)
# light blue for bounding box of paragraph
pdf.setFillColor(black)
pdf.setLineWidth(1) # no line for bounding box
for elem in self.hocr.iterfind(self._child_xpath('p', 'ocr_par')):
elemtxt = self._get_element_text(elem).rstrip()
if len(elemtxt) == 0:
continue
pxl_coords = self.element_coordinates(elem)
pt = self.pt_from_pixel(pxl_coords)
# draw the bbox border
if show_bounding_boxes: # pragma: no cover
pdf.rect(
pt.x1, self.height - pt.y2, pt.x2 - pt.x1, pt.y2 - pt.y1, fill=1
)
found_lines = False
for line in sorted(
chain(
self.hocr.iterfind(self._child_xpath('span', 'ocr_header')),
self.hocr.iterfind(self._child_xpath('span', 'ocr_line')),
self.hocr.iterfind(self._child_xpath('span', 'ocr_textfloat')),
),
key=self.topdown_position,
):
found_lines = True
self._do_line(
pdf,
line,
"ocrx_word",
fontname,
invisible_text,
interword_spaces,
show_bounding_boxes,
)
if not found_lines:
# Tesseract did not report any lines (just words)
root = self.hocr.find(self._child_xpath('div', 'ocr_page'))
self._do_line(
pdf,
root,
"ocrx_word",
fontname,
invisible_text,
interword_spaces,
show_bounding_boxes,
)
# put the image on the page, scaled to fill the page
# finish up the page and save it
pdf.showPage()
pdf.save()
@classmethod
def polyval(cls, poly, x): # pragma: no cover
return x * poly[0] + poly[1]
def _do_line(
self,
pdf: Canvas,
line: Optional[Element],
elemclass: str,
fontname: str,
invisible_text: bool,
interword_spaces: bool,
show_bounding_boxes: bool,
):
if not line:
return
pxl_line_coords = self.element_coordinates(line)
line_box = self.pt_from_pixel(pxl_line_coords)
line_height = line_box.y2 - line_box.y1
slope, pxl_intercept = self.baseline(line)
if abs(slope) < 0.005:
slope = 0.0
angle = atan(slope)
cos_a, sin_a = cos(angle), sin(angle)
text = pdf.beginText()
intercept = pxl_intercept / self.dpi * inch
# Don't allow the font to break out of the bounding box. Division by
# cos_a accounts for extra clearance between the glyph's vertical axis
# on a sloped baseline and the edge of the bounding box.
fontsize = (line_height - abs(intercept)) / cos_a * 1.2
#fontsize = 10.5
text.setFont('Greek', fontsize)
#if invisible_text:
# text.setTextRenderMode(3) # Invisible (indicates OCR text)
# Intercept is normally negative, so this places it above the bottom
# of the line box
baseline_y2 = self.height - (line_box.y2 + intercept)
if False: # pragma: no cover
# draw the baseline in magenta, dashed
pdf.setDash()
pdf.setStrokeColor(magenta)
pdf.setLineWidth(0.5)
# negate slope because it is defined as a rise/run in pixel
# coordinates and page coordinates have the y axis flipped
pdf.line(
line_box.x1,
baseline_y2,
line_box.x2,
self.polyval((-slope, baseline_y2), line_box.x2 - line_box.x1),
)
# light green for bounding box of word/line
pdf.setDash(6, 3)
pdf.setStrokeColor(red)
#text.setTextTransform(cos_a, -sin_a, sin_a, cos_a, line_box.x1, baseline_y2)
text.setTextOrigin(line_box.x1, baseline_y2)
##pdf.translate(line_box.x1, baseline_y2)
pdf.setFillColor(black) # text in black
elements = line.findall(self._child_xpath('span', elemclass))
for elem in elements:
elemtxt = self._get_element_text(elem).strip()
elemtxt = self.replace_unsupported_chars(elemtxt)
if elemtxt == '':
continue
pxl_coords = self.element_coordinates(elem)
box = self.pt_from_pixel(pxl_coords)
if False:
# if `--interword-spaces` is true, append a space
# to the end of each text element to allow simpler PDF viewers
# such as PDF.js to better recognize words in search and copy
# and paste. Do not remove space from last word in line, even
# though it would look better, because it will interfere with
# naive text extraction. \n does not work either.
elemtxt += ' '
box = Rect._make(
(
box.x1,
line_box.y1,
box.x2 + pdf.stringWidth(' ', fontname, line_height),
line_box.y2,
)
)
box_width = box.x2 - box.x1
font_width = pdf.stringWidth(elemtxt, fontname, fontsize)
# draw the bbox border
if False: # pragma: no cover
pdf.rect(
box.x1, self.height - line_box.y2, box_width, line_height, fill=0
)
# Adjust relative position of cursor
# This is equivalent to:
# text.setTextOrigin(pt.x1, self.height - line_box.y2)
# but the former generates a full text reposition matrix (Tm) in the
# content stream while this issues a "offset" (Td) command.
# .moveCursor() is relative to start of the text line, where the
# "text line" means whatever reportlab defines it as. Do not use
# use .getCursor(), since moveCursor() rather unintuitively plans
# its moves relative to .getStartOfLine().
# For skewed lines, in the text transform we set up a rotated
# coordinate system, so we don't have to account for the
# incremental offset. Surprisingly most PDF viewers can handle this.
cursor = text.getStartOfLine()
dx = box.x1 - cursor[0]
dy = baseline_y2 - cursor[1]
text.moveCursor(dx, dy)
# If reportlab tells us this word is 0 units wide, our best seems
# to be to suppress this text
if font_width > 0:
#text.setHorizScale(100 * box_width / font_width)
text.textOut(elemtxt)
pdf.drawText(text)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Convert hocr file to PDF')
parser.add_argument(
'-b',
'--boundingboxes',
action="store_true",
default=False,
help='Show bounding boxes borders',
)
parser.add_argument(
'-r',
'--resolution',
type=int,
default=300,
help='Resolution of the image that was OCRed',
)
parser.add_argument(
'-i',
'--image',
default=None,
help='Path to the image to be placed above the text',
)
parser.add_argument(
'--interword-spaces',
action='store_true',
default=False,
help='Add spaces between words',
)
parser.add_argument('hocrfile', help='Path to the hocr file to be parsed')
parser.add_argument('outputfile', help='Path to the PDF file to be generated')
args = parser.parse_args()
hocr = HocrTransform(hocr_filename=args.hocrfile, dpi=args.resolution)
hocr.to_pdf(
out_filename=args.outputfile,
image_filename=args.image,
show_bounding_boxes=args.boundingboxes,
interword_spaces=args.interword_spaces,
)

23
scripts/templates/base.html Normal file
View File

@ -0,0 +1,23 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1" />
<link
rel="stylesheet"
href="{{ url_for('static', filename='styles.css') }}"
/>
<title>{% block title %} {% endblock %} WordMord</title>
</head>
<body>
<!-- <nav class="navbar navbar-expand-md navbar-light">
<a href="/">Wordmord</a>
<a href="/about">About</a>
</nav> -->
<div class="content">
{% block content %} {% endblock %}
</div>
</body>
</html>

121
scripts/templates/results.html Normal file
View File

@ -0,0 +1,121 @@
{% extends 'base.html' %}
{% block content %}
<!-- <iframe src="https://pad.vvvvvvaria.org/wordmord.test" height="800" width="800"></iframe> -->
<table>
<tr>
<td colspan="2">
<button onclick="decode()">provoli / show PDF</button>
<button id="button1" onclick="transform()">metalucksi PDF / dostort PDF</button>
<button id="button2" onclick="copy()"> antigrafi keimenou ponikou kodika / copy penal code text</button>
</td>
<td>
<div>
<span id="title3"> perioxi epikollimenou keimenou / pasted text area</span>
</div>
</td>
</tr>
<tr>
<td colspan="2">
<!-- <input type="text" id="my_field" value="" wrap="soft"> -->
<textarea name="text" id="my_field" wrap="soft" placeholder="Paste the URL of the PDF here. Κάνε επικόλληση του λινκ εδώ. WordMord believes that the violence of language is not eradicated by merely deleting/erasing words, but rather by transversing their violent imposition through specific practices that trouble and disrupt grammatical consistency, semantic norms, ‘correct’ pronunciation, ‘proper’ bodily posture. The rupture of linguistic limits suggests the possibility of experiencing language in its materiality.
WordMord poses questions on the relationship between language, technology, trauma and violence. The collective artistic research will evolve through workshops, presentations and artworks. Through collaborations with artists, activists and groups working on feminist coding,WordMord seeks to shape an online rhizomatic space as an active feminist archive. At the same time, it project will provide tools and methods towards a poetically subversive meta/para/re-writing of derogatory narratives and consequently of trauma and violence.
#WordMord´s initial research group: Vassiliea Stylianidou aka Franck-Lee Alli-Tis, Angeliki Diakrousi, Christina Karagianni, Stylianos Benetos aka Oýto Arognos, Mounologies: Eleni Diamantouli and Anna Delimpasi. It started in collaboration with the #CNMFPP in 2019.
"></textarea>
</td>
<td rowspan="3">
<textarea name="empty" id="empty_frame" wrap="soft" placeholder="Paste the copied text here. Κάνε επικόλληση του κειμένου εδώ"></textarea>
</td>
</tr>
<tr>
<td>
<span id="title1"> epishmeiomenos poinikos kodikas / annotated penal code</span>
</td>
<td>
<div>
<span id="title2">metalucktriomeno PDF / distorted PDF</span>
</div>
</td>
</tr>
<tr>
<td>
<div id="initialpdf">
<iframe id="inputpdf" src="" height="800rem" width="800rem"></iframe>
</div>
</td>
<td>
<iframe id="showmonster" src="{{url_for('static', filename='pdf/output.pdf')}}" height="800rem" width="800rem"></iframe>
</td>
</tr>
</table>
<!-- WordMord believes that the violence of language is not eradicated by merely deleting/erasing words, but rather by transversing their violent imposition through specific practices that trouble and disrupt grammatical consistency, semantic norms, ‘correct’ pronunciation, ‘proper’ bodily posture. The rupture of linguistic limits suggests the possibility of experiencing language in its materiality.
WordMord poses questions on the relationship between language, technology, trauma and violence. The collective artistic research will evolve through workshops, presentations and artworks. Through collaborations with artists, activists and groups working on feminist coding,WordMord seeks to shape an online rhizomatic space as an active feminist archive. At the same time, it project will provide tools and methods towards a poetically subversive meta/para/re-writing of derogatory narratives and consequently of trauma and violence.
#WordMord´s initial research group: Vassiliea Stylianidou aka Franck-Lee Alli-Tis, Angeliki Diakrousi, Christina Karagianni, Stylianos Benetos aka Oýto Arognos, Mounologies: Eleni Diamantouli and Anna Delimpasi. It started in collaboration with the #CNMFPP in 2019.
Tags -->
<script>
function decode() {
var url = document.getElementById("my_field").value;
var urlDecode = decodeURIComponent(url);
// console.log(urlDecode);
document.getElementById("my_field").value = urlDecode;
const showpdf = document.getElementById("inputpdf");
var textarea = document.getElementById('my_field');
var result
function updateResult() {
result.textContent = textarea.value;
}
textarea.addEventListener('keyup', updateResult);
showpdf.src = textarea.value;
const pdf = document.getElementById("initialpdf");
pdf.style.visibility = "visible";
const button1 = document.getElementById("button1");
button1.style.visibility = "visible";
const button2 = document.getElementById("button2");
button2.style.visibility = "visible";
const title1 = document.getElementById("title1");
title1.style.visibility = "visible";
// textarea.style.color="#fd69B3";
}
function transform() {
const show = document.getElementById("showmonster");
show.style.visibility = "visible";
const title2 = document.getElementById("title2");
title2.style.visibility = "visible";
}
function copy() {
const show = document.getElementById("empty_frame");
show.style.visibility = "visible";
const title3 = document.getElementById("title3");
title3.style.visibility = "visible";
}
$(document).ready(function() {
$('form').find("input[type='text'],textarea").val("");
});
// extract the text of a pdf/ copy on clipboard
</script>
{% endblock %}