pipeline: Move weave* to its own file
This commit is contained in:
James R. Barlow
@@ -0,0 +1,332 @@
|
||||
# © 2018 James R. Barlow: github.com/jbarlow83
|
||||
#
|
||||
# This file is part of OCRmyPDF.
|
||||
#
|
||||
# OCRmyPDF is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
#
|
||||
# OCRmyPDF is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU General Public License for more details.
|
||||
#
|
||||
# You should have received a copy of the GNU General Public License
|
||||
# along with OCRmyPDF. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
from pathlib import Path
|
||||
from itertools import groupby
|
||||
|
||||
import pikepdf
|
||||
|
||||
from .helpers import flatten_groups, page_number
|
||||
|
||||
|
||||
def _update_page_resources(*, page, font, font_key, procset):
|
||||
"Update this page's fonts with a reference to the Glyphless font"
|
||||
|
||||
if '/Resources' not in page:
|
||||
page['/Resources'] = pikepdf.Dictionary({})
|
||||
resources = page['/Resources']
|
||||
try:
|
||||
fonts = resources['/Font']
|
||||
except KeyError:
|
||||
fonts = pikepdf.Dictionary({})
|
||||
if font_key not in fonts:
|
||||
fonts[font_key] = font
|
||||
resources['/Font'] = fonts
|
||||
|
||||
# Reassign /ProcSet to one that just lists everything - ProcSet is
|
||||
# obsolete and doesn't matter but recommended for old viewer support
|
||||
resources['/ProcSet'] = procset
|
||||
|
||||
|
||||
def _weave_layers_graft(
|
||||
*, pdf_base, page_num, text, font, font_key, procset, rotation, log):
|
||||
"""Insert the text layer from text page 0 on to pdf_base at page_num"""
|
||||
|
||||
log.debug("Grafting")
|
||||
if Path(text).stat().st_size == 0:
|
||||
return
|
||||
|
||||
# This is a pointer indicating a specific page in the base file
|
||||
pdf_text = pikepdf.open(text)
|
||||
pdf_text_contents = pdf_text.pages[0].Contents.read_bytes()
|
||||
|
||||
base_page = pdf_base.pages.p(page_num)
|
||||
|
||||
# The text page always will be oriented up by this stage but the original
|
||||
# content may have a rotation applied. Wrap the text stream with a rotation
|
||||
# so it will be oriented the same way as the rest of the page content.
|
||||
# (Previous versions OCRmyPDF rotated the content layer to match the text.)
|
||||
mediabox = [float(pdf_text.pages[0].MediaBox[v].decode())
|
||||
for v in range(4)]
|
||||
wt, ht = mediabox[2] - mediabox[0], mediabox[3] - mediabox[1]
|
||||
|
||||
mediabox = [float(base_page.MediaBox[v].decode())
|
||||
for v in range(4)]
|
||||
wp, hp = mediabox[2] - mediabox[0], mediabox[3] - mediabox[1]
|
||||
|
||||
translate = pikepdf.PdfMatrix((1, 0, 0, 1, -wt / 2, -ht / 2))
|
||||
untranslate = pikepdf.PdfMatrix((1, 0, 0, 1, wp / 2, hp / 2))
|
||||
# -rotation because the input is a clockwise angle and this formula
|
||||
# uses CCW
|
||||
rotation = -rotation % 360
|
||||
if rotation == 0:
|
||||
c, s = 1, 0
|
||||
elif rotation == 90:
|
||||
c, s = 0, 1
|
||||
elif rotation == 180:
|
||||
c, s = -1, 0
|
||||
elif rotation == 270:
|
||||
c, s = 0, -1
|
||||
else:
|
||||
raise NotImplementedError("rotation to arbitrary angle")
|
||||
|
||||
rotate = pikepdf.PdfMatrix((c, s, -s, c, 0, 0))
|
||||
|
||||
# Because of rounding of DPI, we might get a text layer that is not
|
||||
# identically sized to the target page. Scale to adjust. Normally this
|
||||
# is within 0.998.
|
||||
scale_x = wp / wt
|
||||
scale_y = hp / ht
|
||||
if rotation % 90 == 0:
|
||||
scale_x, scale_y = scale_y, scale_x
|
||||
|
||||
log.debug('%r', (scale_x, scale_y))
|
||||
scale = pikepdf.PdfMatrix((scale_x, 0, 0, scale_y, 0, 0))
|
||||
|
||||
# Translate the text so it is centered at (0, 0), rotate it there, adjust
|
||||
# for a size different between initial and text PDF, then untranslate
|
||||
ctm = translate @ rotate @ scale @ untranslate
|
||||
|
||||
pdf_text_contents = (
|
||||
b'q %s cm\n' % ctm.encode() +
|
||||
pdf_text_contents +
|
||||
b'\nQ\n'
|
||||
)
|
||||
|
||||
new_text_layer = pikepdf.Stream(pdf_base, pdf_text_contents)
|
||||
|
||||
base_page.page_contents_add(new_text_layer, prepend=True)
|
||||
|
||||
_update_page_resources(
|
||||
page=base_page, font=font, font_key=font_key, procset=procset
|
||||
)
|
||||
|
||||
|
||||
def _find_font(text, pdf_base):
|
||||
"Copy a font from the filename text into pdf_base"
|
||||
|
||||
font, font_key = None, None
|
||||
possible_font_names = ('/f-0-0', '/F1')
|
||||
try:
|
||||
pdf_text = pikepdf.open(text)
|
||||
pdf_text_fonts = pdf_text.pages[0].Resources.get('/Font', {})
|
||||
except Exception:
|
||||
return None, None
|
||||
|
||||
for f in possible_font_names:
|
||||
pdf_text_font = pdf_text_fonts.get(f, None)
|
||||
if pdf_text_font is not None:
|
||||
font_key = f
|
||||
break
|
||||
if pdf_text_font:
|
||||
font = pdf_base.copy_foreign(pdf_text_font)
|
||||
return font, font_key
|
||||
|
||||
|
||||
def _fix_toc(pdf_base, pageref_remap, log):
|
||||
"""Repair the table of contents
|
||||
|
||||
Whenever we replace a page wholesale, it gets assigned a new objgen number
|
||||
and other references to it within the PDF become invalid, most notably in
|
||||
the table of contents (/Outlines in PDF-speak). In weave_layers we collect
|
||||
pageref_remap, a mapping that describes the new objgen number given an old
|
||||
one. (objgen is a tuple, and the gen is almost always zero.)
|
||||
|
||||
The /Outlines data structure is a messy data structure, but rather than
|
||||
navigating hierarchically we just track unique nodes. Enqueue nodes when
|
||||
we find them, and never visit them again. set() is awesome. We look for
|
||||
the two types of object in the table of contents that can be page bookmarks
|
||||
and update the page entry.
|
||||
|
||||
It may ultimately be better to find a way to rebuild a page in place.
|
||||
|
||||
"""
|
||||
|
||||
visited = set()
|
||||
queue = set()
|
||||
link_keys = ('/Parent', '/First', '/Last', '/Prev', '/Next')
|
||||
|
||||
if not '/Outlines' in pdf_base.root:
|
||||
return
|
||||
if not pageref_remap:
|
||||
return
|
||||
|
||||
def remap_dest(dest_node):
|
||||
if not isinstance(dest_node, pikepdf.Array):
|
||||
return
|
||||
pageref = dest_node[0]
|
||||
if pageref['/Type'] == '/Page' and \
|
||||
pageref._objgen in pageref_remap:
|
||||
new_objgen = pageref_remap[pageref._objgen]
|
||||
dest_node[0] = pdf_base._get_object_id(*new_objgen)
|
||||
|
||||
queue.add(pdf_base.root.Outlines._objgen)
|
||||
while queue:
|
||||
objgen = queue.pop()
|
||||
visited.add(objgen)
|
||||
node = pdf_base._get_object_id(*objgen)
|
||||
log.debug('fix toc: visiting %r', objgen)
|
||||
|
||||
# Enumerate other nodes we could visit from here
|
||||
for key in link_keys:
|
||||
if key not in node:
|
||||
continue
|
||||
item = node[key]
|
||||
if not item.is_indirect:
|
||||
continue
|
||||
objgen = item._objgen
|
||||
if objgen not in visited:
|
||||
queue.add(objgen)
|
||||
|
||||
if '/Dest' in node:
|
||||
remap_dest(node['/Dest'])
|
||||
elif '/A' in node:
|
||||
if '/S' in node['/A'] and node['/A']['/S'] == '/GoTo':
|
||||
remap_dest(node['/A']['/D'])
|
||||
|
||||
|
||||
def weave_layers(
|
||||
infiles,
|
||||
output_file,
|
||||
log,
|
||||
context):
|
||||
"""Apply text layer and/or image layer changes to baseline file
|
||||
|
||||
This is where the magic happens. infiles will be the main PDF to modify,
|
||||
and optional .text.pdf and .image-layer.pdf files, organized however ruffus
|
||||
organizes them.
|
||||
|
||||
From .text.pdf, we copy the content stream (which contains the Tesseract
|
||||
OCR results), and rotate it into place. The first time we do this, we also
|
||||
copy the GlyphlessFont, and then reference that font again.
|
||||
|
||||
For .image-layer.pdf, we check if this is a "pointer" to the original file,
|
||||
or a new file. If a new file, we replace the page and remember that we
|
||||
replaced this page.
|
||||
|
||||
Every 100 open files, we save intermediate results, to avoid any resource
|
||||
limits, since pikepdf/qpdf need to keep a lot of open file handles in the
|
||||
background. When objects are copied from one file to another qpdf, qpdf
|
||||
doesn't actually copy the data until asked to write, so all the resources
|
||||
it may need to remain available.
|
||||
|
||||
For completeness, we set up a /ProcSet on every page, although it's
|
||||
unlikely any PDF viewer cares about this anymore.
|
||||
|
||||
"""
|
||||
|
||||
def input_sorter(key):
|
||||
try:
|
||||
return page_number(key)
|
||||
except ValueError:
|
||||
return -1
|
||||
flat_inputs = sorted(flatten_groups(infiles), key=input_sorter)
|
||||
groups = groupby(flat_inputs, key=input_sorter)
|
||||
|
||||
# Extract first item
|
||||
_, basegroup = next(groups)
|
||||
base = list(basegroup)[0]
|
||||
path_base = Path(base).resolve()
|
||||
pdf_base = pikepdf.open(path_base)
|
||||
keep_open = []
|
||||
font, font_key, procset = None, None, None
|
||||
pdfinfo = context.get_pdfinfo()
|
||||
pagerefs = {}
|
||||
|
||||
procset = pdf_base.make_indirect(
|
||||
pikepdf.Object.parse(b'[ /PDF /Text /ImageB /ImageC /ImageI ]'))
|
||||
|
||||
# Iterate rest
|
||||
for page_num, layers in groups:
|
||||
layers = list(layers)
|
||||
log.debug(page_num)
|
||||
log.debug(layers)
|
||||
|
||||
text = next(
|
||||
(ii for ii in layers if ii.endswith('.text.pdf')), None
|
||||
)
|
||||
image = next(
|
||||
(ii for ii in layers if ii.endswith('.image-layer.pdf')), None
|
||||
)
|
||||
|
||||
if text and not font:
|
||||
font, font_key = _find_font(text, pdf_base)
|
||||
|
||||
replacing = False
|
||||
content_rotation = pdfinfo[page_num - 1].rotation
|
||||
|
||||
path_image = Path(image).resolve() if image else None
|
||||
if path_image is not None and path_image != path_base:
|
||||
# We are replacing the old page with a rasterized PDF of the new
|
||||
# page
|
||||
log.debug("Replace")
|
||||
old_objgen = pdf_base.pages[page_num - 1]._objgen
|
||||
|
||||
pdf_image = pikepdf.open(image)
|
||||
keep_open.append(pdf_image)
|
||||
image_page = pdf_image.pages[0]
|
||||
pdf_base.pages[page_num - 1] = image_page
|
||||
|
||||
# We're adding a new page, which will get a new objgen number pair,
|
||||
# so we need to update any references to it. qpdf did not like
|
||||
# my attempt to update the old object in place, but that is an
|
||||
# option to consider
|
||||
pagerefs[old_objgen] = pdf_base.pages[page_num - 1]._objgen
|
||||
replacing = True
|
||||
|
||||
autorotate_correction = context.get_rotation(page_num - 1)
|
||||
if replacing:
|
||||
content_rotation = autorotate_correction
|
||||
text_rotation = autorotate_correction
|
||||
text_misaligned = (text_rotation - content_rotation) % 360
|
||||
log.debug('%r', [
|
||||
text_rotation, autorotate_correction, text_misaligned,
|
||||
content_rotation]
|
||||
)
|
||||
|
||||
if text and font:
|
||||
# Graft the text layer onto this page, whether new or old
|
||||
_weave_layers_graft(
|
||||
pdf_base=pdf_base, page_num=page_num, text=text, font=font,
|
||||
font_key=font_key, rotation=text_misaligned, procset=procset,
|
||||
log=log
|
||||
)
|
||||
|
||||
# Correct the rotation if applicable
|
||||
pdf_base.pages[page_num - 1].Rotate = \
|
||||
(content_rotation - autorotate_correction) % 360
|
||||
|
||||
if len(keep_open) > 100:
|
||||
# qpdf limitations require us to keep files open when we intend
|
||||
# to copy content from them before saving. However, we want to keep
|
||||
# a lid on file handles and memory usage, so for big files we're
|
||||
# going to stop and save periodically. Attach the font to page 1
|
||||
# even if page 1 doesn't use it, so we have a way to get it back.
|
||||
page0 = pdf_base.pages[0]
|
||||
_update_page_resources(
|
||||
page=page0, font=font, font_key=font_key, procset=procset)
|
||||
interim = output_file + '_working{}.pdf'.format(page_num)
|
||||
pdf_base.save(interim)
|
||||
del pdf_base
|
||||
keep_open = []
|
||||
|
||||
pdf_base = pikepdf.open(interim)
|
||||
procset = pdf_base.pages[0].Resources.ProcSet
|
||||
font = pdf_base.pages[0].Resources.Font.get(font_key)
|
||||
|
||||
_fix_toc(pdf_base, pagerefs, log)
|
||||
pdf_base.save(output_file)
|
||||
@@ -168,3 +168,11 @@ if sys.version_info[0:2] <= (3, 5):
|
||||
else:
|
||||
universal_open = open
|
||||
fspath = os.fspath
|
||||
|
||||
|
||||
def flatten_groups(groups):
|
||||
for obj in groups:
|
||||
if is_iterable_notstr(obj):
|
||||
yield from obj
|
||||
else:
|
||||
yield obj
|
||||
|
||||
+2
-271
@@ -19,7 +19,6 @@ from contextlib import suppress
|
||||
from shutil import copyfileobj
|
||||
from pathlib import Path
|
||||
from datetime import datetime, timezone
|
||||
from itertools import groupby
|
||||
|
||||
import sys
|
||||
import os
|
||||
@@ -36,7 +35,7 @@ from ruffus import formatter, regex, Pipeline, suffix
|
||||
from .hocrtransform import HocrTransform
|
||||
from .pdfinfo import PdfInfo, Encoding, Colorspace
|
||||
from .pdfa import generate_pdfa_ps, encode_pdf_date
|
||||
from .helpers import re_symlink, is_iterable_notstr, page_number
|
||||
from .helpers import re_symlink, is_iterable_notstr, page_number, flatten_groups
|
||||
from .exec import ghostscript, tesseract, qpdf
|
||||
from .lib import fitz
|
||||
from .exceptions import PdfMergeFailedError, UnsupportedImageFormatError, \
|
||||
@@ -44,6 +43,7 @@ from .exceptions import PdfMergeFailedError, UnsupportedImageFormatError, \
|
||||
from . import leptonica
|
||||
from . import PROGRAM_NAME, VERSION
|
||||
from ._optimize import optimize
|
||||
from ._weave import weave_layers
|
||||
|
||||
|
||||
VECTOR_PAGE_DPI = 400
|
||||
@@ -679,275 +679,6 @@ def render_hocr_page(
|
||||
interwordSpaces=True)
|
||||
|
||||
|
||||
def flatten_groups(groups):
|
||||
for obj in groups:
|
||||
if is_iterable_notstr(obj):
|
||||
yield from obj
|
||||
else:
|
||||
yield obj
|
||||
|
||||
|
||||
def _update_page_resources(*, page, font, font_key, procset):
|
||||
# Update page fonts with reference to Glyphless
|
||||
if '/Resources' not in page:
|
||||
page['/Resources'] = pikepdf.Dictionary({})
|
||||
resources = page['/Resources']
|
||||
try:
|
||||
fonts = resources['/Font']
|
||||
except KeyError:
|
||||
fonts = pikepdf.Dictionary({})
|
||||
if font_key not in fonts:
|
||||
fonts[font_key] = font
|
||||
resources['/Font'] = fonts
|
||||
|
||||
# Reassign /ProcSet to one that just lists everything - ProcSet is
|
||||
# obsolete and doesn't matter but recommended for old viewer support
|
||||
resources['/ProcSet'] = procset
|
||||
|
||||
|
||||
def _weave_layers_graft(
|
||||
*, pdf_base, page_num, text, font, font_key, procset, rotation, log):
|
||||
log.debug("Grafting")
|
||||
if Path(text).stat().st_size == 0:
|
||||
return
|
||||
|
||||
# This is a pointer indicating a specific page in the base file
|
||||
pdf_text = pikepdf.open(text)
|
||||
pdf_text_contents = pdf_text.pages[0].Contents.read_bytes()
|
||||
|
||||
base_page = pdf_base.pages.p(page_num)
|
||||
|
||||
# The text page always will be oriented up by this stage but the original
|
||||
# content may have a rotation applied. Wrap the text stream with a rotation
|
||||
# so it will be oriented the same way as the rest of the page content.
|
||||
# (Previous versions OCRmyPDF rotated the content layer to match the text.)
|
||||
mediabox = [float(pdf_text.pages[0].MediaBox[v].decode())
|
||||
for v in range(4)]
|
||||
wt, ht = mediabox[2] - mediabox[0], mediabox[3] - mediabox[1]
|
||||
|
||||
mediabox = [float(base_page.MediaBox[v].decode())
|
||||
for v in range(4)]
|
||||
wp, hp = mediabox[2] - mediabox[0], mediabox[3] - mediabox[1]
|
||||
|
||||
translate = pikepdf.PdfMatrix((1, 0, 0, 1, -wt / 2, -ht / 2))
|
||||
untranslate = pikepdf.PdfMatrix((1, 0, 0, 1, wp / 2, hp / 2))
|
||||
# -rotation because the input is a clockwise angle and this formula
|
||||
# uses CCW
|
||||
rotation = -rotation % 360
|
||||
if rotation == 0:
|
||||
c, s = 1, 0
|
||||
elif rotation == 90:
|
||||
c, s = 0, 1
|
||||
elif rotation == 180:
|
||||
c, s = -1, 0
|
||||
elif rotation == 270:
|
||||
c, s = 0, -1
|
||||
else:
|
||||
raise NotImplementedError("rotation to arbitrary angle")
|
||||
|
||||
rotate = pikepdf.PdfMatrix((c, s, -s, c, 0, 0))
|
||||
|
||||
# Because of rounding of DPI, we might get a text layer that is not
|
||||
# identically sized to the target page. Scale to adjust. Normally this
|
||||
# is within 0.998.
|
||||
scale_x = wp / wt
|
||||
scale_y = hp / ht
|
||||
if rotation % 90 == 0:
|
||||
scale_x, scale_y = scale_y, scale_x
|
||||
|
||||
log.debug('%r', (scale_x, scale_y))
|
||||
scale = pikepdf.PdfMatrix((scale_x, 0, 0, scale_y, 0, 0))
|
||||
|
||||
# Translate the text so it is centered at (0, 0), rotate it there, adjust
|
||||
# for a size different between initial and text PDF, then untranslate
|
||||
ctm = translate @ rotate @ scale @ untranslate
|
||||
|
||||
pdf_text_contents = (
|
||||
b'q %s cm\n' % ctm.encode() +
|
||||
pdf_text_contents +
|
||||
b'\nQ\n'
|
||||
)
|
||||
|
||||
new_text_layer = pikepdf.Stream(pdf_base, pdf_text_contents)
|
||||
|
||||
base_page.page_contents_add(new_text_layer, prepend=True)
|
||||
|
||||
_update_page_resources(
|
||||
page=base_page, font=font, font_key=font_key, procset=procset
|
||||
)
|
||||
|
||||
|
||||
def _find_font(text, pdf_base):
|
||||
"Copy a font from the filename text into pdf_base"
|
||||
|
||||
font, font_key = None, None
|
||||
possible_font_names = ('/f-0-0', '/F1')
|
||||
try:
|
||||
pdf_text = pikepdf.open(text)
|
||||
pdf_text_fonts = pdf_text.pages[0].Resources.get('/Font', {})
|
||||
except Exception:
|
||||
return None, None
|
||||
|
||||
for f in possible_font_names:
|
||||
pdf_text_font = pdf_text_fonts.get(f, None)
|
||||
if pdf_text_font is not None:
|
||||
font_key = f
|
||||
break
|
||||
if pdf_text_font:
|
||||
font = pdf_base.copy_foreign(pdf_text_font)
|
||||
return font, font_key
|
||||
|
||||
|
||||
def _fix_toc(pdf_base, pageref_remap, log):
|
||||
visited = set()
|
||||
queue = set()
|
||||
link_keys = ('/Parent', '/First', '/Last', '/Prev', '/Next')
|
||||
|
||||
if not '/Outlines' in pdf_base.root:
|
||||
return
|
||||
if not pageref_remap:
|
||||
return
|
||||
|
||||
def remap_dest(dest_node):
|
||||
if not isinstance(dest_node, pikepdf.Array):
|
||||
return
|
||||
pageref = dest_node[0]
|
||||
if pageref['/Type'] == '/Page' and \
|
||||
pageref._objgen in pageref_remap:
|
||||
new_objgen = pageref_remap[pageref._objgen]
|
||||
dest_node[0] = pdf_base._get_object_id(*new_objgen)
|
||||
|
||||
queue.add(pdf_base.root.Outlines._objgen)
|
||||
while queue:
|
||||
objgen = queue.pop()
|
||||
visited.add(objgen)
|
||||
node = pdf_base._get_object_id(*objgen)
|
||||
log.debug('fix toc: visiting %r', objgen)
|
||||
|
||||
# Enumerate other nodes we could visit from here
|
||||
for key in link_keys:
|
||||
if key not in node:
|
||||
continue
|
||||
item = node[key]
|
||||
if not item.is_indirect:
|
||||
continue
|
||||
objgen = item._objgen
|
||||
if objgen not in visited:
|
||||
queue.add(objgen)
|
||||
|
||||
log.debug(repr(node))
|
||||
if '/Dest' in node:
|
||||
remap_dest(node['/Dest'])
|
||||
elif '/A' in node:
|
||||
if '/S' in node['/A'] and node['/A']['/S'] == '/GoTo':
|
||||
remap_dest(node['/A']['/D'])
|
||||
|
||||
|
||||
def weave_layers(
|
||||
infiles,
|
||||
output_file,
|
||||
log,
|
||||
context):
|
||||
"Apply text layer and/or image layer changes to baseline file"
|
||||
|
||||
def input_sorter(key):
|
||||
try:
|
||||
return page_number(key)
|
||||
except ValueError:
|
||||
return -1
|
||||
flat_inputs = sorted(flatten_groups(infiles), key=input_sorter)
|
||||
groups = groupby(flat_inputs, key=input_sorter)
|
||||
|
||||
# Extract first item
|
||||
_, basegroup = next(groups)
|
||||
base = list(basegroup)[0]
|
||||
path_base = Path(base).resolve()
|
||||
pdf_base = pikepdf.open(path_base)
|
||||
keep_open = []
|
||||
font, font_key, procset = None, None, None
|
||||
pdfinfo = context.get_pdfinfo()
|
||||
pagerefs = {}
|
||||
|
||||
procset = pdf_base.make_indirect(
|
||||
pikepdf.Object.parse(b'[ /PDF /Text /ImageB /ImageC /ImageI ]'))
|
||||
|
||||
# Iterate rest
|
||||
for page_num, layers in groups:
|
||||
layers = list(layers)
|
||||
log.debug(page_num)
|
||||
log.debug(layers)
|
||||
|
||||
text = next(
|
||||
(ii for ii in layers if ii.endswith('.text.pdf')), None
|
||||
)
|
||||
image = next(
|
||||
(ii for ii in layers if ii.endswith('.image-layer.pdf')), None
|
||||
)
|
||||
|
||||
if text and not font:
|
||||
font, font_key = _find_font(text, pdf_base)
|
||||
|
||||
replacing = False
|
||||
content_rotation = pdfinfo[page_num - 1].rotation
|
||||
|
||||
path_image = Path(image).resolve() if image else None
|
||||
if path_image is not None and path_image != path_base:
|
||||
# We are replacing the old page
|
||||
log.debug("Replace")
|
||||
old_objgen = pdf_base.pages[page_num - 1]._objgen
|
||||
|
||||
pdf_image = pikepdf.open(image)
|
||||
keep_open.append(pdf_image)
|
||||
image_page = pdf_image.pages[0]
|
||||
pdf_base.pages[page_num - 1] = image_page
|
||||
|
||||
pagerefs[old_objgen] = pdf_base.pages[page_num - 1]._objgen
|
||||
replacing = True
|
||||
|
||||
autorotate_correction = context.get_rotation(page_num - 1)
|
||||
if replacing:
|
||||
content_rotation = autorotate_correction
|
||||
text_rotation = autorotate_correction
|
||||
text_misaligned = (text_rotation - content_rotation) % 360
|
||||
log.debug('%r', [
|
||||
text_rotation, autorotate_correction, text_misaligned,
|
||||
content_rotation]
|
||||
)
|
||||
|
||||
if text and font:
|
||||
# Graft the text layer onto this page, whether new or old
|
||||
_weave_layers_graft(
|
||||
pdf_base=pdf_base, page_num=page_num, text=text, font=font,
|
||||
font_key=font_key, rotation=text_misaligned, procset=procset,
|
||||
log=log
|
||||
)
|
||||
|
||||
# Correct the rotation if applicable
|
||||
pdf_base.pages[page_num - 1].Rotate = \
|
||||
(content_rotation - autorotate_correction) % 360
|
||||
|
||||
if len(keep_open) > 100:
|
||||
# qpdf limitations require us to keep files open when we intend
|
||||
# to copy content from them before saving. However, we want to keep
|
||||
# a lid on file handles and memory usage, so for big files we're
|
||||
# going to stop and save periodically. Attach the font to page 1
|
||||
# even if page 1 doesn't use it, so we have a way to get it back.
|
||||
page0 = pdf_base.pages[0]
|
||||
_update_page_resources(
|
||||
page=page0, font=font, font_key=font_key, procset=procset)
|
||||
interim = output_file + '_working{}.pdf'.format(page_num)
|
||||
pdf_base.save(interim)
|
||||
del pdf_base
|
||||
keep_open = []
|
||||
|
||||
pdf_base = pikepdf.open(interim)
|
||||
procset = pdf_base.pages[0].Resources.ProcSet
|
||||
font = pdf_base.pages[0].Resources.Font.get(font_key)
|
||||
|
||||
_fix_toc(pdf_base, pagerefs, log)
|
||||
pdf_base.save(output_file)
|
||||
|
||||
|
||||
def ocr_tesseract_textonly_pdf(
|
||||
infiles,
|
||||
outfiles,
|
||||
|
||||
Reference in New Issue
Block a user