Optical Character Recognition - An OCR invoice reader

Description

The process is powered by pdf2image and pytesseract. We load pdf file as an image data and feed in to pytesseract. pytesseract would recognize the text and then we can use regular expression to extract the key info from the text. The regular expression template is stored in a YAML file.

Simple invoice template

Here is a sample of a valid template

issuer: ABC Company
keywords:
  - ABC Company
exclude_keywords:
  - Paid
fields:
  amount: (?<=TOTAL\s\$)\d+?\.\d+\b
  date: (?<=DUE\sDATE\s)20[0-9][0-9]\/[0-1][0-9]\/[0-3][0-9]\b
  invoice_number: (?<=INVOICE\s\#\s)[A-Z]{3}\-[0-9]{4}

issuer, keywords, exclude_keywords and fields are required attributes of a valid template. issuer is defined at the time where the template is specifically written for a company. keywords and exclude_keywords are also defined by users to help the computer to read the right invoices. fields is the data that we want to extract. This should be written in a regular expression.

Simple Demo

Import the modules

from DataExtraction.Regex import fieldExtractor
from DataExtraction.Scanner import pdfScanner

Set the directory where invoices and templates are located

pdf_dir = 'Demo Invoices/'
template_dir = 'DataExtraction/Templates/'

Convert pdf to text

ocr = pdfScanner(pdf_dir)
texts = ocr.extract_text()

Extract data from the text and save it as a json file

worker = fieldExtractor(template_dir, texts)
data = worker.fetch()
worker.save(data)

Result data.json

[
  {
    "invoice_number": [
      "TWA-1190"
    ],
    "amount": [
      "204.75"
    ],
    "date": [
      "2021/07/23"
    ]
  }
]