def init(self, page:fitz.Page) -> Layout:
'''Initialize layout object.'''
# Layout object based on raw dict
# NOTE: all these coordinates are relative to un-rotated page
# https://pymupdf.readthedocs.io/en/latest/page.html#modifying-pages
raw_layout = page.getText('rawdict')
# though 'width', 'height' are contained in `raw_dict`, they are based on un-rotated page.
# so, update page width/height to right direction in case page is rotated
*_, w, h = page.rect # always reflecting page rotation
raw_layout.update({
'width' : w,
'height': h
})
self._layout = Layout(raw_layout, page.rotationMatrix)
# get rectangle shapes from page source
self._layout.rects.from_stream(self.doc_pdf, page)
# get annotations(comment shapes) from PDF page, e.g.
# highlight, underline and strike-through-line
self._layout.rects.from_annotations(page)
# plot raw layout
if self.debug_mode:
# new section for current pdf page
new_page_section(self._doc_debug, self._layout.width, self._layout.height, f'Page {page.number}')
# initial layout
self._layout.plot(self._doc_debug, 'Original Text Blocks', key=PlotControl.LAYOUT)
self._layout.plot(self._doc_debug, 'Original Rectangle Shapes', key=PlotControl.SHAPE)
return self._layout
def initialize(self, page: fitz.Page):
'''Initialize layout object.'''
# -----------------------------------------
# Layout object based on raw dict
# -----------------------------------------
# NOTE: all these coordinates are relative to un-rotated page
# https://pymupdf.readthedocs.io/en/latest/page.html#modifying-pages
raw_layout = page.getText('rawdict')
# -----------------------------------------
# page size
# -----------------------------------------
# though 'width', 'height' are contained in `raw_dict`, they are based on un-rotated page.
# so, update page width/height to right direction in case page is rotated
*_, w, h = page.rect # always reflecting page rotation
raw_layout.update({'width': w, 'height': h})
# -----------------------------------------
# page images
# -----------------------------------------
# image bytes from page.getText('rawdict') can't reproduce transparent images,
# so we re-extract page images
for block in raw_layout['blocks']:
# disable image in raw dict
if block['type'] == 1: block['type'] = -1
# extract and recover images
images = ImagesExtractor.extract_images(page)
raw_layout['blocks'].extend(images)
# -----------------------------------------
# page paths
# -----------------------------------------
# convert vector graphic paths to pixmap
self._paths_extractor = PathsExtractor()
images, paths = self._paths_extractor.extract_paths(page)
raw_layout['blocks'].extend(images)
raw_layout['paths'] = paths
# init layout
self._layout = Layout(raw_layout, page.rotationMatrix)
return self._layout
def _page_blocks(page:fitz.Page):
'''Get page blocks and adjust image blocks.'''
# Layout object based on raw dict:
# NOTE: all these coordinates are relative to un-rotated page
# https://pymupdf.readthedocs.io/en/latest/page.html#modifying-pages
raw_layout = page.getText('rawdict')
# Adjust image blocks:
# Image blocks are generated for every image location – whether or not there are any duplicates.
# This is in contrast to Page.getImageList(), which will contain each image only once.
# https://pymupdf.readthedocs.io/en/latest/textpage.html#dictionary-structure-of-extractdict-and-extractrawdict
#
# So, a compromise:
# - get image contents with `page.getImageList`
# - get image location with `page.getText('rawdict')`
#
# extract and recover images
recovered_images = ImagesExtractor.extract_images(page)
# group original image blocks by image contents
image_blocks_group = defaultdict(list)
for block in raw_layout['blocks']:
if block['type'] != 1: continue
image_blocks_group[hash(block['image'])].append(block)
# update raw layout blocks
def same_images(img, img_list):
bbox = list(map(round, img['bbox']))
for _img in img_list:
if list(map(round, _img['bbox']))==bbox: return True
return False
for image in recovered_images:
for k, image_blocks in image_blocks_group.items():
if not same_images(image, image_blocks): continue
for image_block in image_blocks: image_block['image'] = image['image']
break
# an image outside the page is not counted in page.getText(), so let's add it here
else:
raw_layout['blocks'].append(image)
return raw_layout
def initialize(self, page:fitz.Page):
'''Initialize layout object.'''
# Layout object based on raw dict
# NOTE: all these coordinates are relative to un-rotated page
# https://pymupdf.readthedocs.io/en/latest/page.html#modifying-pages
raw_layout = page.getText('rawdict')
# though 'width', 'height' are contained in `raw_dict`, they are based on un-rotated page.
# so, update page width/height to right direction in case page is rotated
*_, w, h = page.rect # always reflecting page rotation
raw_layout.update({ 'width' : w, 'height': h })
self._layout = Layout(raw_layout, page.rotationMatrix)
# get rectangle shapes from page source
self._layout.rects.from_stream(self.doc_pdf, page)
# get annotations(comment shapes) from PDF page, e.g.
# highlight, underline and strike-through-line
self._layout.rects.from_annotations(page)
return self._layout
def init(self, page: fitz.Page) -> Layout:
'''Initialize layout object.'''
# Layout object based on raw dict
raw_layout = page.getText('rawdict')
self._layout = Layout(raw_layout)
# get rectangle shapes from page source:
# these shapes are generally converted from docx, e.g. highlight, underline,
# which are different from PDF comments like highlight, rectangle.
if not page._isWrapped:
page._wrapContents()
# transformation matrix from PDF to PyMuPDF
M = page.transformationMatrix # PyMuPDF>=1.17.0
for xref in page.getContents():
page_content = self._doc_pdf._getXrefStream(xref).decode(
encoding="ISO-8859-1")
self._layout.rects.from_stream(page_content, M)
# get annotations(comment shapes) from PDF page: consider highlight, underline,
# strike-through-line only.
annots = page.annots()
self._layout.rects.from_annotations(annots)
# plot raw layout
if self.debug_mode:
# new section for current pdf page
new_page_section(self._doc_debug, self._layout.width,
self._layout.height, f'Page {page.number}')
# initial layout
self._layout.plot(self._doc_debug,
'Original Text Blocks',
key=PlotControl.LAYOUT)
self._layout.plot(self._doc_debug,
'Original Rectangle Shapes',
key=PlotControl.SHAPE)
return self._layout
def initialize(self, page:fitz.Page):
'''Initialize layout object.'''
# Layout object based on raw dict
# NOTE: all these coordinates are relative to un-rotated page
# https://pymupdf.readthedocs.io/en/latest/page.html#modifying-pages
raw_layout = page.getText('rawdict')
# though 'width', 'height' are contained in `raw_dict`, they are based on un-rotated page.
# so, update page width/height to right direction in case page is rotated
*_, w, h = page.rect # always reflecting page rotation
raw_layout.update({ 'width' : w, 'height': h })
# pdf paths and converted images
self._paths = PathsExtractor()
images, paths = self._paths.parse(page).filter_pixmaps(page)
raw_layout['blocks'].extend(images)
raw_layout['paths'] = paths
# init layout
self._layout = Layout(raw_layout, page.rotationMatrix)
return self._layout
def init(self, page:fitz.Page) -> Layout:
'''Initialize layout object.'''
# Layout object based on raw dict
raw_layout = page.getText('rawdict')
new_layout = split_blocks(raw_layout)
self._layout = Layout(new_layout)
# get rectangle shapes from page source
self._layout.rects.from_stream(self.doc_pdf, page)
# get annotations(comment shapes) from PDF page, e.g.
# highlight, underline and strike-through-line
self._layout.rects.from_annotations(page)
# plot raw layout
if self.debug_mode:
# new section for current pdf page
new_page_section(self._doc_debug, self._layout.width, self._layout.height, f'Page {page.number}')
# initial layout
self._layout.plot(self._doc_debug, 'Original Text Blocks', key=PlotControl.LAYOUT)
self._layout.plot(self._doc_debug, 'Original Rectangle Shapes', key=PlotControl.SHAPE)
return self._layout