Ocr_pdf_table_csv

This is a project that extracts data from pdf snapshot and enters data into csv This project is using 2 methods to detect boxes from table. First method depends on the border of table. Check the result of first method in "focus_border_Images" Second method depends on the text. Check the result of first method in "focus_text_Images" It was tested on Window and is using multi-threading on 2 stages to speed up.

• Extract JPGs from pdf
• Extract data from JPG Pretrained Tesseract model is used in this project. You can find the project that is used custom model by CNN on https://github.com/Supernova1024/OCR-PDF-to-CSV-Table-by-CNN

Dependencies

• poppler-0.68.0
• Anaconda3-2020.11-Windows-x86_64.exe
• tesseract-ocr-w64-setup-v5.1.0.20220510.exe

Installing

• Extract poppler-0.68.0 and copy it into C:\Program Files\
• Install ananconda and add path to environment variables
• Install tesseract-ocr-w64-setup-v5.1.0.20220510.exe
• Download this repository
• Install requirements.txt in project root directory

How to Run

• Convert pdf to images Open the pdf_image.py and define the parameters. You can check the parameters here. https://pdf2image.readthedocs.io/en/latest/reference.html Then, run this command in project root dirctory

  python pdf_img.py
  

  Extracted JPG files are stored in "pdf_img" folder

• Extract boxes by border

  python focus_border.py
  

  The result is stored in "focus_border_Images" folder

• Extract boxes by text

  python focus_text.py
  

  The result is stored in "focus_text_Images" folder

  

• You can get the final result by running only below command after running pdf_img.py

  python ocr_border.py
  

  This script extracts the boxes by border and get the OCR result by pytesseract The results are stored in "output_img" folder and "table_1.csv" file.

Result Description

In this project, I used the table that has 7 columns 1, 4, and 5 columns can't be recognized by pytesseract. The boxes of these columns are stored in "output_img" folder as JPG and added their file name to csv file. You can check example of "output_img" folder here. https://drive.google.com/drive/folders/1nrns5zZkzfVP9o8aCyjkj9O4O-TwJAvP?usp=sharing Other columns can be recognized by pytesseract and the results are stored in csv directly. The csv keeps table structure of original pdf I attached example "output_img" folder and "table_1.csv"

Futher development

• Development of ocr_text.py based on text of table.
• Improve accuracy
• Extract data from any table.

Version

1.0.2

Please give me star if this project was helpful to your startup project. :)