I working on a pdf file dedupe project and analyzed many libraries in python, which read files, then generate hash value of it and then compare it with the next file for duplication - similar to logic below or using python filecomp lib. But the issue I found these logic is like, if a pdf is generated from a source DOCX(Save to PDF) , those outputs are not considered duplicates - even content is exactly the same. Why this happens? Is there any other logic to read the content, then create a unique hash value based on the actual content.
def calculate_hash_val(path, blocks=65536):
file = open(path, 'rb')
hasher = hashlib.md5()
data = file.read()
while len(data) > 0:
hasher.update(data)
data = file.read()
file.close()
return hasher.hexdigest()
One of the things that happens is that you save metadata to the file including the time of creation. It is invisible in the PDF, but that will make the hash different.
Here is an explanation of how to find and strip out that data with at least one tool. I am sure that there are many others.
I have scenario where I have PDFs with a letterhead and table-like body of text. I have tried using pdfminer but I'm struggling to figure out how to approach my problem
An example of the format for one my PDFs
In specific, pdf miner reads the data starting from the letterhead up until the table header. It then reads the table header in a row like fashion from left to right. From there it's just beyond messy.
Here is python to convert pdf to text:
import pdfminer
import sys
from pdfminer.high_level import extract_text
text = extract_text('./quote2.pdf')
print((text))
f = open("results2.txt", "w")
f.write(text)
And here is a snippet of what the output looks like:
... letter head info
ITEM�#
DESCRIPTION
561347
55�PCs-792.00�LB
6061-T651�PLATE�AMS�4027
4�S/C�6"�SQUARE
CUTTING�PLATE�SAW�ALUM
PACKAGING�SKIDDING
SHIP�VIA�:�OUR�TRUCK
Quotation
DATE:
CUSTOMER NUMBER:
QUOTE NUMBER:
FOB:
4/1/2022
319486
957242
Destination
SHIP TO:
The idea was to use regex to extract relevant numbers. As you can see it read the first 2 records for columns ITEM and DESCRIPTION, but from there it starts back up from the letterhead, and it's even more messy below
Is there perhaps a way to seperate the letterhead from the rest of the body as a starting step? Very new to python, not sure how to get what I want, help much appreciated!
I have a pdf file over 100 pages. There are boxes and columns of text. When I extract the text using PyPdf2 and tika parser, I get a string of of data which is out of order. It is ordered by columns in many cases and skips around the document in other cases. Is it possible to read the pdf file starting from the top, moving left to right until the bottom? I want to read the text in the columns and boxes, but I want the line of text displayed as it would be read left to right.
I've tried:
PyPDF2 - the only tool is extracttext(). Fast but does not give gaps in the elements. Results are jumbled.
Pdfminer - PDFPageInterpeter() method with LAParams. This works well but is slow. At least 2 seconds per page and I've got 200 pages.
pdfrw - this only tells me the number of pages.
tabula_py - only gives me the first page. Maybe I'm not looping it correctly.
tika - what I'm currently working with. Fast and more readable, but the content is still jumbled.
from tkinter import filedialog
import os
from tika import parser
import re
# select the file you want
file_path = filedialog.askopenfilename(initialdir=os.getcwd(),filetypes=[("PDF files", "*.pdf")])
print(file_path) # print that path
file_data = parser.from_file(file_path) # Parse data from file
text = file_data['content'] # Get files text content
by_page = text.split('... Information') # split up the document into pages by string that always appears on the
# top of each page
for i in range(1,len(by_page)): # loop page by page
info = by_page[i] # get one page worth of data from the pdf
reformated = info.replace("\n", "&") # I replace the new lines with "&" to make it more readable
print("Page: ",i) # print page number
print(reformated,"\n\n") # print the text string from the pdf
This provides output of a sort, but it is not ordered in the way I would like. I want the pdf to be read left to right. Also, if I could get a pure python solution, that would be a bonus. I don't want my end users to be forced to install java (I think the tika and tabula-py methods are dependent on java).
I did this for .docx with this code. Where txt is the .docx. Hope this help link
import re
pttrn = re.compile(r'(\.|\?|\!)(\'|\")?\s')
new = re.sub(pttrn, r'\1\2\n\n', txt)
print(new)
I have a PDF with a big table splitted in pages, so I need to join the per-page tables into a big table in a large page.
Is this possible with PyPDF2 or another library?
Cheers
Just working on something similar, it takes an input pdf and via a config file you can set the final pattern of single pages.
Implementation with PyPDF2 but it still has issues with some pdf-files (have to dig deeper).
https://github.com/Lageos/pdf-stitcher
In principle adding a page right to another one works like:
import PyPDF2
with open('input.pdf', 'rb') as input_file:
# load input pdf
input_pdf = PyPDF2.PdfFileReader(input_file)
# start new PyPDF2 PageObject
output_pdf = input_pdf.getPage(page_number)
# get second page PyPDF2 PageObject
second_pdf = input_pdf.getPage(second_page_number)
# dimensions for offset from loaded page (adding it to the right)
offset_x = output_pdf.mediaBox[2]
offset_y = 0
# add second page to first one
output_pdf.mergeTranslatedPage(second_pdf, offset_x, offset_y, expand=True)
# write finished pdf
with open('output.pdf', 'wb') as out_file:
write_pdf = PyPDF2.PdfFileWriter()
write_pdf.addPage(output_pdf)
write_pdf.write(out_file)
Adding a page below needs an offset_y. You can get the amount from offset_y = first_pdf.mediaBox[3].
My understanding is that this is quite hard. See here and here.
The problem seems to be that tables aren't very well represented in pdfs but are simply made from absolutely positioned lines (see first link above).
Here are two possible workarounds (not sure if they will do it for you):
you could print multiple pages on one page and scale the page to make it readable....
open the pdf with inkscape or something similar. Once ungrouped, you should have access to the individual elements that make up the tables and be able to combine them the way that suit you
EDIT
Have a look at libre office draw, another vector package. I just opened a pdf in it and it seems to preserve some of the pdf structure and editing individual elements.
EDIT 2
Have a look at pdftables which might help.
PDFTables helps with extracting tables from PDF files.
I haven't tried it though... might have some time a bit later to see if I can get it to work.
for working with MS word files in python, there is python win32 extensions, which can be used in windows. How do I do the same in linux?
Is there any library?
Use the native Python docx module. Here's how to extract all the text from a doc:
document = docx.Document(filename)
docText = '\n\n'.join(
paragraph.text for paragraph in document.paragraphs
)
print(docText)
See Python DocX site
Also check out Textract which pulls out tables etc.
Parsing XML with regexs invokes cthulu. Don't do it!
You could make a subprocess call to antiword. Antiword is a linux commandline utility for dumping text out of a word doc. Works pretty well for simple documents (obviously it loses formatting). It's available through apt, and probably as RPM, or you could compile it yourself.
benjamin's answer is a pretty good one. I have just consolidated...
import zipfile, re
docx = zipfile.ZipFile('/path/to/file/mydocument.docx')
content = docx.read('word/document.xml').decode('utf-8')
cleaned = re.sub('<(.|\n)*?>','',content)
print(cleaned)
OpenOffice.org can be scripted with Python: see here.
Since OOo can load most MS Word files flawlessly, I'd say that's your best bet.
I know this is an old question, but I was recently trying to find a way to extract text from MS word files, and the best solution by far I found was with wvLib:
http://wvware.sourceforge.net/
After installing the library, using it in Python is pretty easy:
import commands
exe = 'wvText ' + word_file + ' ' + output_txt_file
out = commands.getoutput(exe)
exe = 'cat ' + output_txt_file
out = commands.getoutput(exe)
And that's it. Pretty much, what we're doing is using the commands.getouput function to run a couple of shell scripts, namely wvText (which extracts text from a Word document, and cat to read the file output). After that, the entire text from the Word document will be in the out variable, ready to use.
Hopefully this will help anyone having similar issues in the future.
Take a look at how the doc format works and create word document using PHP in linux. The former is especially useful. Abiword is my recommended tool. There are limitations though:
However, if the document has complicated tables, text boxes, embedded spreadsheets, and so forth, then it might not work as expected. Developing good MS Word filters is a very difficult process, so please bear with us as we work on getting Word documents to open correctly. If you have a Word document which fails to load, please open a Bug and include the document so we can improve the importer.
(Note: I posted this on this question as well, but it seems relevant here, so please excuse the repost.)
Now, this is pretty ugly and pretty hacky, but it seems to work for me for basic text extraction. Obviously to use this in a Qt program you'd have to spawn a process for it etc, but the command line I've hacked together is:
unzip -p file.docx | grep '<w:t' | sed 's/<[^<]*>//g' | grep -v '^[[:space:]]*$'
So that's:
unzip -p file.docx: -p == "unzip to stdout"
grep '<w:t': Grab just the lines containing '<w:t' (<w:t> is the Word 2007 XML element for "text", as far as I can tell)
sed 's/<[^<]>//g'*: Remove everything inside tags
grep -v '^[[:space:]]$'*: Remove blank lines
There is likely a more efficient way to do this, but it seems to work for me on the few docs I've tested it with.
As far as I'm aware, unzip, grep and sed all have ports for Windows and any of the Unixes, so it should be reasonably cross-platform. Despit being a bit of an ugly hack ;)
If your intention is to use purely python modules without calling a subprocess, you can use the zipfile python modude.
content = ""
# Load DocX into zipfile
docx = zipfile.ZipFile('/home/whateverdocument.docx')
# Unpack zipfile
unpacked = docx.infolist()
# Find the /word/document.xml file in the package and assign it to variable
for item in unpacked:
if item.orig_filename == 'word/document.xml':
content = docx.read(item.orig_filename)
else:
pass
Your content string however needs to be cleaned up, one way of doing this is:
# Clean the content string from xml tags for better search
fullyclean = []
halfclean = content.split('<')
for item in halfclean:
if '>' in item:
bad_good = item.split('>')
if bad_good[-1] != '':
fullyclean.append(bad_good[-1])
else:
pass
else:
pass
# Assemble a new string with all pure content
content = " ".join(fullyclean)
But there is surely a more elegant way to clean up the string, probably using the re module.
Hope this helps.
Unoconv might also be a good alternative: http://linux.die.net/man/1/unoconv
To read Word 2007 and later files, including .docx files, you can use the python-docx package:
from docx import Document
document = Document('existing-document-file.docx')
document.save('new-file-name.docx')
To read .doc files from Word 2003 and earlier, make a subprocess call to antiword. You need to install antiword first:
sudo apt-get install antiword
Then just call it from your python script:
import os
input_word_file = "input_file.doc"
output_text_file = "output_file.txt"
os.system('antiword %s > %s' % (input_word_file, output_text_file))
If you have LibreOffice installed, you can simply call it from the command line to convert the file to text, then load the text into Python.
Is this an old question?
I believe that such thing does not exist.
There are only answered and unanswered ones.
This one is pretty unanswered, or half answered if you wish.
Well, methods for reading *.docx (MS Word 2007 and later) documents without using COM interop are all covered.
But methods for extracting text from *.doc (MS Word 97-2000), using Python only, lacks.
Is this complicated?
To do: not really, to understand: well, that's another thing.
When I didn't find any finished code, I read some format specifications and dug out some proposed algorithms in other languages.
MS Word (*.doc) file is an OLE2 compound file.
Not to bother you with a lot of unnecessary details, think of it as a file-system stored in a file. It actually uses FAT structure, so the definition holds. (Hm, maybe you can loop-mount it in Linux???)
In this way, you can store more files within a file, like pictures etc.
The same is done in *.docx by using ZIP archive instead.
There are packages available on PyPI that can read OLE files. Like (olefile, compoundfiles, ...)
I used compoundfiles package to open *.doc file.
However, in MS Word 97-2000, internal subfiles are not XML or HTML, but binary files.
And as this is not enough, each contains an information about other one, so you have to read at least two of them and unravel stored info accordingly.
To understand fully, read the PDF document from which I took the algorithm.
Code below is very hastily composed and tested on small number of files.
As far as I can see, it works as intended.
Sometimes some gibberish appears at the start, and almost always at the end of text.
And there can be some odd characters in-between as well.
Those of you who just wish to search for text will be happy.
Still, I urge anyone who can help to improve this code to do so.
doc2text module:
"""
This is Python implementation of C# algorithm proposed in:
http://b2xtranslator.sourceforge.net/howtos/How_to_retrieve_text_from_a_binary_doc_file.pdf
Python implementation author is Dalen Bernaca.
Code needs refining and probably bug fixing!
As I am not a C# expert I would like some code rechecks by one.
Parts of which I am uncertain are:
* Did the author of original algorithm used uint32 and int32 when unpacking correctly?
I copied each occurence as in original algo.
* Is the FIB length for MS Word 97 1472 bytes as in MS Word 2000, and would it make any difference if it is not?
* Did I interpret each C# command correctly?
I think I did!
"""
from compoundfiles import CompoundFileReader, CompoundFileError
from struct import unpack
__all__ = ["doc2text"]
def doc2text (path):
text = u""
cr = CompoundFileReader(path)
# Load WordDocument stream:
try:
f = cr.open("WordDocument")
doc = f.read()
f.close()
except: cr.close(); raise CompoundFileError, "The file is corrupted or it is not a Word document at all."
# Extract file information block and piece table stream informations from it:
fib = doc[:1472]
fcClx = unpack("L", fib[0x01a2l:0x01a6l])[0]
lcbClx = unpack("L", fib[0x01a6l:0x01a6+4l])[0]
tableFlag = unpack("L", fib[0x000al:0x000al+4l])[0] & 0x0200l == 0x0200l
tableName = ("0Table", "1Table")[tableFlag]
# Load piece table stream:
try:
f = cr.open(tableName)
table = f.read()
f.close()
except: cr.close(); raise CompoundFileError, "The file is corrupt. '%s' piece table stream is missing." % tableName
cr.close()
# Find piece table inside a table stream:
clx = table[fcClx:fcClx+lcbClx]
pos = 0
pieceTable = ""
lcbPieceTable = 0
while True:
if clx[pos]=="\x02":
# This is piece table, we store it:
lcbPieceTable = unpack("l", clx[pos+1:pos+5])[0]
pieceTable = clx[pos+5:pos+5+lcbPieceTable]
break
elif clx[pos]=="\x01":
# This is beggining of some other substructure, we skip it:
pos = pos+1+1+ord(clx[pos+1])
else: break
if not pieceTable: raise CompoundFileError, "The file is corrupt. Cannot locate a piece table."
# Read info from pieceTable, about each piece and extract it from WordDocument stream:
pieceCount = (lcbPieceTable-4)/12
for x in xrange(pieceCount):
cpStart = unpack("l", pieceTable[x*4:x*4+4])[0]
cpEnd = unpack("l", pieceTable[(x+1)*4:(x+1)*4+4])[0]
ofsetDescriptor = ((pieceCount+1)*4)+(x*8)
pieceDescriptor = pieceTable[ofsetDescriptor:ofsetDescriptor+8]
fcValue = unpack("L", pieceDescriptor[2:6])[0]
isANSII = (fcValue & 0x40000000) == 0x40000000
fc = fcValue & 0xbfffffff
cb = cpEnd-cpStart
enc = ("utf-16", "cp1252")[isANSII]
cb = (cb*2, cb)[isANSII]
text += doc[fc:fc+cb].decode(enc, "ignore")
return "\n".join(text.splitlines())
I'm not sure if you're going to have much luck without using COM. The .doc format is ridiculously complex, and is often called a "memory dump" of Word at the time of saving!
At Swati, that's in HTML, which is fine and dandy, but most word documents aren't so nice!
Just an option for reading 'doc' files without using COM: miette. Should work on any platform.
Aspose.Words Cloud SDK for Python is a platform independent solution to convert MS Word/Open Office files to text. It is a commercial product but free trial plan provides 150 monthly API calls.
P.S: I am a developer evangelist at Aspose.
# For complete examples and data files, please go to https://github.com/aspose-words-cloud/aspose-words-cloud-python
# Import module
import asposewordscloud
import asposewordscloud.models.requests
from shutil import copyfile
# Please get your Client ID and Secret from https://dashboard.aspose.cloud.
client_id='xxxxxxx-xxxx-xxxx-xxxxx-xxxxxxxxxx'
client_secret='xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'
words_api = asposewordscloud.WordsApi(client_id,client_secret)
words_api.api_client.configuration.host='https://api.aspose.cloud'
filename = 'C:/Temp/02_pages.docx'
dest_name = 'C:/Temp/02_pages.txt'
#Convert RTF to text
request = asposewordscloud.models.requests.ConvertDocumentRequest(document=open(filename, 'rb'), format='txt')
result = words_api.convert_document(request)
copyfile(result, dest_name)