ullah2017ocr.md

OCR Engine to extract Food-items and Prices from Receipt Images via Pattern matching and heuristics approach

Rafi Ullah, Ali Sohani, Faraz Ali, Athaul Rai

Browse

@inproceedings{ullah2017ocr,
author = {Ullah, Rafi and Sohani, Ali and Ali, Faraz and Rai, Athaul},
year = {2017},
month = {12},
pages = {},
title = {OCR Engine to extract Food-items and Prices from Receipt Images via Pattern matching and heuristics approach}
}

Pipeline

Receipt detection	Receipt localization	Receipt normalization	Text line segmentation	Optical character recognition	Semantic analysis
❌	✔️	✔️	❌	❗	✔️

Receipt localization

• Image Background Removal

Receipt normalization

• Otsu’s Binarization

• Image de-skewing with https://www.pyimagesearch.com/2017/02/20/text-skew-correction-opencv-python/

• Image resizing

  Image having DPI (Dots per Inch) greater than 300 has been observed good results.

  bicubic interpolation

Optical character recognition

• Tesseract OCR Library

Semantic analysis

• Fields extracted:
  • item names,
  • item prices,
  • item quantities.
• Regular expressions for the pattern matching I-e item names, item prices and items quantities.
• Change due, walmart, total etc. words are removed.
• Words and line containing “constant words” such as total, discount, sub total etc. will be removed.
• Heuristics that if line contain item name and number (price and quantity), we consider those lines and discard rest of the lines.
• But some of the receipts we observed, having different structure. Our heuristic is applied, but very weak over these kind of receipts. Items and prices on separate lines.

Notes

• Before localization, Image stitching is performed if more than one photo of receipt is available.

• Generic receipt parser.

•