Extract / Identify Tables from PDF python [closed]

Question

Are there any open source libraries that support table identification & extraction?

By this I mean:

1. Identify a table structure exists
2. Classify the table from its contents
3. Extract data from the table in a useful output format e.g. JSON / CSV etc.

I have looked through similar questions on this topic and found the following:

• PDFMiner which addresses problem 3, but it seems the user is required to specify to PDFMiner where a table structure exists for each table (correct me if I'm wrong)
• pdf-table-extract which attempts to address problem 1 but according to the To-Do list, cannot currently identify tables that are separated by whitespace. This is a problem as all tables in my PDFs are separated by whitespace!

Currently, I am thinking that I would have to spend a lot of time developing a Machine Learning solution to identify table structures from PDFs. Therefore, any alternative approaches would be more than welcome!

Answer 1

After many fruitful hours of exploring OCR libraries, bounding boxes and clustering algorithms - I found a solution so simple it makes you want to cry!

I hope you are using Linux;

pdftotext -layout NAME_OF_PDF.pdf

AMAZING!!

Now you have a nice text file with all the information lined up in nice columns, now it is trivial to format into a csv etc..

It is for times like this that I love Linux, these guys came up with AMAZING solutions to everything, and put it there for FREE!

Answer 2

You should definitely have a look at this answer of mine:

• Extracting table contents from a collection of PDF files

and also have a look at all the links included therein.

Tabula/TabulaPDF is currently the best table extraction tool that is available for PDF scraping.

Answer 3

I'd just like to add to the very helpful answer from Kurt Pfeifle - there is now a Python wrapper for Tabula, and this seems to work very well so far: https://github.com/chezou/tabula-py

This will convert your PDF table to a Pandas data frame. You can also set the area in x,y co-ordinates which is obviously very handy for irregular data.