def pdfminer_parse(file_path, file_hash):
output_string = StringIO()
try:
with open(file_path, 'rb') as fp:
parser = PDFParser(fp)
doc = PDFDocument(parser)
laparams = LAParams()
rsrcmgr = PDFResourceManager(caching=True)
txt_device = TextConverter(rsrcmgr, output_string, codec='utf-8', laparams=laparams)
aggr_device = PDFPageAggregator(rsrcmgr, laparams=laparams)
txt_interpreter = PDFPageInterpreter(rsrcmgr, txt_device)
aggr_interpreter = PDFPageInterpreter(rsrcmgr, aggr_device)
from pdfminer.high_level import extract_pages
extract_pages("test.pdf")
for page in PDFPage.get_pages(fp, caching=True):
txt_interpreter.process_page(page)
aggr_interpreter.process_page(page)
layout = aggr_device.get_result()
# for element in layout:
# if isinstance(element, LTImage):
# print()
# elif isinstance(element, LTFigure):
# print()
# TODO Complete
except pdfparser.PDFSyntaxError:
print(f"{file_hash} is not a PDF")
return None
return {
'text': output_string.getvalue(),
'meta': flatten_and_unicode(doc.info)
}
def test_get_common_font_from_pages(get_test_decision_standard_pdf_path,
get_test_decision_v1_pdf_path):
pages_1 = extract_pages(get_test_decision_standard_pdf_path.open('rb'))
assert common.get_common_font_from_pages(
list(pages_1)) == 'TimesNewRomanPSMT'
pages_2 = extract_pages(get_test_decision_v1_pdf_path.open('rb'))
assert common.get_common_font_from_pages(list(pages_2)) == 'Arial'
def test_line_margin(self):
# The lines have margin 0.2 relative to the height.
# Extract with line_margin 0.19 should break into 3 separate textboxes.
pages = list(
extract_pages(self._get_test_file_path(),
laparams=LAParams(line_margin=0.19)))
self.assertEqual(len(pages), 1)
page = pages[0]
elements = [
element for element in page if isinstance(element, LTTextContainer)
]
self.assertEqual(len(elements), 3)
self.assertEqual(elements[0].get_text(), "Text1\n")
self.assertEqual(elements[1].get_text(), "Text2\n")
self.assertEqual(elements[2].get_text(), "Text3\n")
# Extract with line_margin 0.21 should merge into one textbox.
pages = list(
extract_pages(self._get_test_file_path(),
laparams=LAParams(line_margin=0.21)))
self.assertEqual(len(pages), 1)
page = pages[0]
elements = [
element for element in page if isinstance(element, LTTextContainer)
]
self.assertEqual(len(elements), 1)
self.assertEqual(elements[0].get_text(), "Text1\nText2\nText3\n")
def test_extract_text_containers(get_test_decision_standard_pdf_path):
page = extract_pages(get_test_decision_standard_pdf_path.open('rb'),
page_numbers=[1])
text_containers = common.extract_text_containers(page)
assert len(text_containers) == 31
page = extract_pages(get_test_decision_standard_pdf_path.open('rb'),
page_numbers=[2])
text_containers = common.extract_text_containers(page)
assert len(text_containers) == 33
def load(
cls, fn, page_numbers=None, preload=False, reader_kwargs=None,
**kwargs
):
"""Load PDF file.
Parameters
----------
fn : str
Filename of PDF.
page_numbers : list
Pages to be loaded. (0-indexed)
preload : bool
Preload content from all pages.
reader_kwargs : dict
Other kwargs for `pdfminer.high_level.extract_pages()`.
"""
if reader_kwargs:
reader_kwargs.pop('page_numbers', None)
else:
reader_kwargs = {}
pages = pmhl.extract_pages(
fn, page_numbers=page_numbers, **reader_kwargs
)
if preload:
pages = list(pages)
return cls(pages)
def element_generator(
file_path: str,
page_numbers=None) -> Generator[LTTextContainer, None, None]:
"""
yields flat list of paragraphs within a document.
:param file_path:
:return:
"""
pNumber = 0
# disable boxes_flow, style based hierarchy detection is based on purely flat list of paragraphs
params = LAParams(boxes_flow=None,
detect_vertical=False) # setting for easy doc
# params = LAParams(boxes_flow=0.5, detect_vertical=True) # setting for column doc
# todo, do pre-analysis in count_sizes --> are there many boxes within same line
# todo, understand LAParams, for columns, NONE works better, for vertical only layout LAParams(boxes_flow=None, detect_vertical=False) works better!! :O
# do some sort of layout analyis, if there are many boxes vertically next to each other, use layout analysis
# - column type
# - straight forward document
for page_layout in extract_pages(file_path,
laparams=params,
page_numbers=page_numbers):
for element in page_layout:
if isinstance(element, LTTextContainer):
element.meta = {"page": pNumber}
yield element
pNumber += 1
def parse():
files = os.listdir(bp.config['UPLOAD_FOLDER'])
filepaths = [bp.config['UPLOAD_FOLDER'] + '/' + x for x in files]
d_files = {k: v for k, v in enumerate(filepaths)}
print(d_files)
VocabularyIndex = InvertedIndex()
for k, v in d_files.items():
# iterate through document apges
for page_layout in extract_pages(v):
# compile all text on page
page_text = ""
for (count, element) in enumerate(page_layout, 1):
if isinstance(element, LTTextContainer):
page_text += element.get_text()
#print(page_text)
# add page to inverted index
page_no = int(page_layout.pageid)
VocabularyIndex.index_document(page_no, page_text)
print(f"processed {page_no}")
data = VocabularyIndex.get_index()
print(data)
'''
def findMaxFontSize(pdf):
""" 求出目标 pdf 第一页中的最大字体大小并返回
:param pdf: 待处理的 pdf 文件url
:return: 最大字体大小
"""
maxFontSize = 0
count = 0
LIMIT = 4 # 最多处理到第 LIMIT 个元素
for page_layout in extract_pages(pdf, page_numbers=0):
for element in page_layout:
count += 1
if not isinstance(element, LTTextContainer): continue
first_line = next(element.__iter__())
first_ch = next(first_line.__iter__())
if not isinstance(first_ch, LTChar): continue
# 只要求该元素第一行的第一个字符的大小即可
size = first_ch.size
if (size > maxFontSize):
maxFontSize = size
if count == LIMIT: break # end for element
break # end for page_layout
return maxFontSize
def get_layout(file_path):
for page_layout in extract_pages(file_path):
for element in page_layout:
#if isinstance(element, LTTextBoxHorizontal):
#print(element.get_text())
with open('../data/xml_files/adho_conferences/testlayout.txt', 'a',encoding="utf-8") as fd:
fd.write(str(element))
def pdf_to_text(fp, parse_page=[]):
'''
Input parameters
------
fp: file path of pdf
parse_page: page numbers (starting at 0), to conver to text. All pages if empty []
Output returns
------
string_output: String output of contents in pdf
'''
output_string = StringIO()
with open(fp, 'rb') as in_file:
num_pages = len(list(extract_pages(in_file)))
#print('Number of pages {}'.format(num_pages))
parser = PDFParser(in_file)
doc = PDFDocument(parser)
rsrcmgr = PDFResourceManager()
device = TextConverter(rsrcmgr, output_string, laparams=LAParams())
interpreter = PDFPageInterpreter(rsrcmgr, device)
for i, page in enumerate(PDFPage.create_pages(doc)):
if len(parse_page) > 0 and i not in parse_page:
continue
#print('Processing page number {}'.format(i))
interpreter.process_page(page)
string_output = output_string.getvalue()
# Close open handles
device.close()
output_string.close()
return string_output
def get_aadhar_details(path, password):
i = 0
try:
for page_layout in extract_pages(path, password):
for element in page_layout:
if isinstance(element, LTTextBoxHorizontal):
i += 1
if i == 2:
addr = element.get_text()
elif i == 24:
aadhar_no = element.get_text()
addr_list = addr.splitlines()
num_list = aadhar_no.splitlines()
#name
# print(addr_list[2])
# #mobile number
# print(addr_list[-1])
# #aadhar number
aadhar = num_list[0]
splitted_aadhar_with_space = aadhar.split(' ')
aadhar = "".join(splitted_aadhar_with_space)
return (addr_list[2], addr_list[-1], aadhar)
except:
return False
def identify_features(self,pdf_path,doc2vec):
#we download text examples representing features
features=pickle.load(open("features.p","rb"))
#we transform all the features to doc2vec embedings
f_vecs=doc2vec.transform([v for k,v in features.items()])
paragraphs=[]
#we go through the pages of the pdf_document
for page_layout in extract_pages(pdf_path):
#through all the elements in the page_layout
for i, element in enumerate(page_layout):
#we Only append if it is a textCOntainer and it has a certain length
if isinstance(element, LTTextContainer) and len(element.get_text()) > 20 and re.match('[a-zA-Z]+',element.get_text()[0]) != None:
paragraphs.append(element.get_text())
#we transform all paragraphs in doc2vec embeddings
paragraphs_d2v=doc2vec.transform(paragraphs)
#we create in which we will keep which paragraphs are most representative of certain feature. Important it is an ordered dict
#because the order will used to access the calculated similarities
features_similarities=OrderedDict({k:[] for k in features.keys()})
for i,p in enumerate(paragraphs_d2v):
#we calculate of a paragraph and all the feature examples
sims=doc2vec.model.wv.cosine_similarities(p.toarray().transpose(),f_vecs.toarray())
#go through all the features and append their corresponding similarity
for ix,k in enumerate(features_similarities.keys()):
features_similarities[k].append([i,sims[ix][ix]])
#sorting the similarities
for k in features_similarities.keys():
features_similarities[k]=sorted(features_similarities[k],key=lambda x: x[1])
return paragraphs,features_similarities
def read(
self,
override_la_params=None,
override_page_numbers=None
) -> Generator[LTTextContainer, Any, None]:
pNumber = 0
# disable boxes_flow, style based hierarchy detection is based on purely flat list of paragraphs
# params = LAParams(boxes_flow=None, detect_vertical=False) # setting for easy doc
# params = LAParams(boxes_flow=0.5, detect_vertical=True) # setting for column doc
# todo, do pre-analysis in count_sizes --> are there many boxes within same line
# todo, understand LAParams, for columns, NONE works better, for vertical only layout LAParams(boxes_flow=None, detect_vertical=False) works better!! :O
# do some sort of layout analyis, if there are many boxes vertically next to each other, use layout analysis
# - column type
# - straight forward document
for page_layout in extract_pages(
self.uri,
laparams=self.la_params
if not override_la_params else override_la_params,
page_numbers=self.page_numbers
if not override_page_numbers else override_page_numbers):
for element in page_layout:
element.page = pNumber
if isinstance(element, LTTextContainer):
yield from self.split_boxes_by_style(element)
#yield element
elif isinstance(element, LTFigure):
yield from self.__handle_lt_figure(element)
pNumber += 1
def layout():
writeText()
la_params = LAParams(boxes_flow=-0.5)
pages = extract_pages(FILE_PATH, laparams=la_params)
for page_layout in pages:
for element in page_layout:
if isinstance(element, LTTextContainer):
print(element.get_text().encode('utf8'))
def mock_download(self):
pages = list(
extract_pages(
join(DOWNLOADING_PATH, f'{product}_sections_{language}.pdf'),
laparams=LAParams(boxes_flow=BOXES_FLOW,
char_margin=CHAR_MARGIN),
))
return pages
def _extract_pages_and_text_containers(self, pdf):
self._pages = list(
extract_pages(pdf, laparams=self.data.get('laparams', None)))[1:]
for page in self._pages:
self._text_containers.append([
element for element in page
if isinstance(element, LTTextContainer) and element.get_text()
!= '' and not re.search(r'^\s+$', element.get_text())
])
def extractAuthor(pdf, elem, line):
""" 根据作者所在元素等位置信息求出目标 pdf 中作者信息并处理后返回
:param pdf: 待处理的 pdf 文件url
:param elem: 含作者信息的元素位置
:param line: 作者信息所在行的偏移量
:return: 作者信息列表
"""
# print('[Authors]')
author_str = ''
for page_layout in extract_pages(pdf, page_numbers=0):
elem_count = 0
for element in page_layout:
if not isinstance(element, LTTextContainer): continue
if elem_count != elem:
elem_count += 1
continue
if line == 0: # 此时读入目标元素的全部内容
for text_line in element:
size = next(text_line.__iter__()).size
for character in text_line:
ch = character.get_text()
# 去掉特殊的角标
if ch == ' ' or ch == '†' or ch == '*':
author_str += ' '
elif not isinstance(character, LTChar):
continue
elif str(character.size) == str(size):
author_str += ch
else: # 此时从当前元素的剩余行读入作者信息
line_count = 0
for text_line in element:
if line_count < line:
line_count += 1
continue
size = next(text_line.__iter__()).size
for character in text_line:
ch = character.get_text()
# 去掉特殊的角标
if ch == ' ' or ch == '†' or ch == '*':
author_str += ' '
elif not isinstance(character, LTChar):
continue
elif str(character.size) == str(size):
author_str += ch
line_count += 1
break # end for element
break # end for page_layout
author_str = author_str.replace(u'\xa0',
u'').replace('&',
',').replace('and ', ',')
return author_str.encode('gbk', 'ignore')
def _extract_pages_and_text_containers(self, pdf):
self._pages = list(
extract_pages(pdf, laparams=self.data.get('laparams', None)))
for page in self._pages:
self._text_containers.append([
element for element in page
if isinstance(element, LTTextContainer) and all([
element.get_text().strip() != '',
])
])
def extractpages(i):
for layout in extract_pages(i):
for element in layout:
if isinstance(element, LTTextContainer):
text = element.get_text()
c.append(text)
#elif isinstance(element, LTImage):
#text = element.get_any()
#c.append(text)
return c
def _extract_pages_and_text_containers(self, pdf):
"""Method called to prepare the pages and containers from pdf for processing during pre-process."""
self._pages = list(
extract_pages(pdf, laparams=self.data.get('laparams', None)))
for page in self._pages:
self._text_containers.append([
element for element in page
if isinstance(element, LTTextContainer) and element.get_text()
!= '' and not re.search(r'^\s+$', element.get_text())
])
def download(self):
"""Download EPAR pdf."""
path = urlretrieve(self.download_url_)[0]
pages = list(
extract_pages(
path,
laparams=LAParams(boxes_flow=self.BOXES_FLOW_,
char_margin=self.CHAR_MARGIN_),
))
return pages
def page():
layout = extract_pages(f)
for i in layout:
if isinstance(i, LTTextContainer):
print(i.get_text())
for text_line in i:
for character in text_line:
if isinstance(character, LTChar):
print(character.fontname)
print(character.size)
def check_decision(cls,
item: Optional[PDPCDecisionItem] = None,
options: Optional[Options] = None) -> bool:
with pdpc_decisions.classes.PDFFile(item, options) as pdf:
first_page = extract_pages(pdf, page_numbers=[0])
containers = common.extract_text_containers(first_page)
for container in containers:
if container.get_text().strip() == 'SUMMARY OF THE DECISION':
return True
return False
def load(
pdf_file: IO,
pdf_file_path: Optional[str] = None,
la_params: Optional[Dict] = None,
**kwargs,
) -> PDFDocument:
"""
Loads the pdf file into a PDFDocument.
Args:
pdf_file (io): The PDF file.
la_params (dict): The layout parameters passed to PDF Miner for analysis. See
the PDFMiner documentation here:
https://pdfminersix.readthedocs.io/en/latest/reference/composable.html#laparams.
Note that py_pdf_parser will re-order the elements it receives from PDFMiner
so options relating to element ordering will have no effect.
pdf_file_path (str, optional): Passed to `PDFDocument`. See the documentation
for `PDFDocument`.
kwargs: Passed to `PDFDocument`. See the documentation for `PDFDocument`.
Returns:
PDFDocument: A PDFDocument with the file loaded.
"""
if la_params is None:
la_params = {}
la_params = {**DEFAULT_LA_PARAMS, **la_params}
pages: Dict[int, Page] = {}
for page in extract_pages(pdf_file, laparams=LAParams(**la_params)):
elements = [
element for element in page if isinstance(element, LTTextBox)
]
# If all_texts=True then we may get some text from inside figures
if la_params.get("all_texts"):
figures = (element for element in page
if isinstance(element, LTFigure))
for figure in figures:
elements += [
element for element in figure
if isinstance(element, LTTextBox)
]
if not elements:
logger.warning(
f"No elements detected on page {page.pageid}, skipping this page."
)
continue
pages[page.pageid] = Page(width=page.width,
height=page.height,
elements=elements)
return PDFDocument(pages=pages, pdf_file_path=pdf_file_path, **kwargs)
def from_pdf_to_txt(read_file):
results = ''
for page_layout in extract_pages(read_file):
for element in page_layout:
# print(element)
if isinstance(element, LTTextBoxHorizontal):
string = element.get_text()
string = string.replace(":", '')
string = string.replace("：", '')
results = results + string.replace(' ', '')
# print(results)
return results
def test_no_boxes_flow(self):
pages = list(
extract_pages(self._get_test_file_path(),
laparams=LAParams(boxes_flow=None)))
self.assertEqual(len(pages), 1)
page = pages[0]
elements = [
element for element in page if isinstance(element, LTTextContainer)
]
self.assertEqual(len(elements), 1)
self.assertEqual(elements[0].get_text(), "Text1\nText2\nText3\n")
def extract_PDF_textbox(pdf_name=PDF_NAME):
text_group_index = 0
for page_layout in extract_pages(pdf_name, laparams=LAParams(line_margin=LINE_MARGIN)):
for element in page_layout:
if isinstance(element, LTTextBox):
text_group_index += 1
RESULT[text_group_index] = []
for text_line in element:
text = text_line.get_text()
RESULT[text_group_index].append(text)
return f"Converted {text_group_index} group of texts from PDF"
def _extract_pages_and_text_containers(self, pdf):
self._pages = list(extract_pages(pdf, laparams=self.data.get('laparams', None)))
if common.check_first_page_is_cover(pdf):
self._pages = self._pages[1:]
for page in self._pages:
containers = [element for element in page if isinstance(element, LTTextContainer) and
element.get_text() != '' and not
re.search(r'^\s+$', element.get_text())]
containers = common.split_joined_text_containers(containers)
containers = sorted(containers, key=lambda item: item.x0)
containers = sorted(containers, key=lambda item: item.y0, reverse=True)
self._text_containers.append(containers)
def process_args(inputfile: str, outputfile: str):
FINAL_OUTPUT = "Created with Ardio by Bell Eapen at nuchange.com. "
print('{0} is the input and {1} is output'.format(inputfile, outputfile))
common_font_size = get_common_font_size(inputfile)
for page_layout in extract_pages(inputfile):
for element in page_layout:
if isinstance(element, LTTextContainer):
if(get_common_font_size_of_element(element) == common_font_size):
FINAL_OUTPUT = FINAL_OUTPUT + \
remove_all_but_alpabets(element.get_text())
print(FINAL_OUTPUT)
write_audio_file(FINAL_OUTPUT, outputfile)
def get_common_font_size(inputfile: str):
SIZE_COUNT = {}
for page_layout in extract_pages(inputfile):
for element in page_layout:
if isinstance(element, LTTextContainer):
for text_line in element:
for character in text_line:
if isinstance(character, LTChar):
if character.size in SIZE_COUNT:
SIZE_COUNT[character.size] = SIZE_COUNT[character.size] + 1
else:
SIZE_COUNT[character.size] = 1
return max(SIZE_COUNT.items(), key=operator.itemgetter(1))[0]
