Tesseract error in preprocessing
one2many opened this issue · 3 comments
Attempting to OCR a table and I keep getting an error.
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/table_ocr/pdf_to_images/init.py", line 69, in preprocess_img
rotate = get_rotate(filepath, tess_params)
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/table_ocr/pdf_to_images/init.py", line 79, in get_rotate
subprocess.check_output(tess_command)
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/subprocess.py", line 336, in check_output
**kwargs).stdout
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/subprocess.py", line 418, in run
output=stdout, stderr=stderr)
subprocess.CalledProcessError: Command '['tesseract', '--psm', '0', '--oem', '0', '/Users/andrewmcfadden/Documents/GitHub/one2many.github.io/image-table-ocr/dance/ga-20190131-001.png', '-']' returned non-zero exit status 1.
The image is the logo at the top of the page (every page).
That last line of the error message is informative.
'tesseract', '--psm', '0', '--oem', '0', '/Users/andrewmcfadden/Documents/GitHub/one2many.github.io/image-table-ocr/dance/ga-20190131-001.png', '-'
Try running that command and see if it gives you any more useful info.
tesseract --psm 0 --oem 0 /Users/andrewmcfaddn/Documents/GitHub/one2many.github.io/image-table-ocr/dance/ga20190131-001.png -
Maybe tesseract takes a --verbose flag that will print additional useful information about the error?
Looking at that image of the FirstBank logo... it doesn't look like something this library was written for. I don't think it will parse anything from that as-is. It will require a huge amount of customization for that.
I will try to add the verbose flag and see what happens. The logo is where it gets stuck, but I really don't need it. My end goal is to run this on 60 pdfs of bank statements that are 98% tables. The logo is on the top left of the first page of every document though. I ran this on one pdf and it got stuck on the logo, but if I could ignore it that would be fine.
Attached is sample of one of the tables.
Also, have you seen Tabula?
This library will need a lot of customization to work with a table like that. The code looks for vertical and horizontal lines to detect a table. And then when it finds a cell, it expects the cell to contain a single line of text. Your example is incompatible with both of those expectations.
The best I can suggest if you want to use anything from this library is to use it as a reference while writing a lot of custom code.