---

Faded, Torn, Rotated Receipt OCR with Image Preprocessing

As part of KaggleX BIPOC Mentorship Program-Cohort2 (Dec 2022 - Mar 2023), I created a Receipt OCR Web App called WeSplit. This app utilizes Tesseract OCR engine, an open source library, and custom preprocessing steps that I designed to improve the quality of the input image. The subset of the code is shared here to demonstrate the preprocessing steps and its positive impact to the OCR output.

• Author: Jaelin Lee
• Date: Mar 18, 2023
• LinkedIn

---

Installation

1. Download this repository to your local machine
2. Open the project directory
3. In Terminal, run automated script to install packages

• chmod +x install_packages.sh
• sh install_packages.sh

4. Create folders under the project directory

• raw
• preprocessed
• output

Input

• Add a scanned receipt (i.e. .JPG, .PNG) to raw folder

Run

• python3 run.py

Output

• Enhanced receipt (enhanced.jpg) is saved in preprocessedfolder
• OCR text ourput (enhanced.txt) is saved in output folder

Enjoy!

Faded Receipt:

Rotated / Crumbled Receipt:

Torn Receipt:

Preprocessing Components

Reference:

• For a detailed explanation of the preprocessing steps used in this project (https://medium.com/@jaelin_75015/faded-torn-rotated-receipt-ocr-with-image-preprocessing-1fb03c036504)
• For a quick sample run of this project in Kaggle notebook (https://www.kaggle.com/jaelin215/faded-receipt-ocr-tesseract-preprocessing)

License:

• MIT License
• I would appreciate if you quote my name and the link to this GitHub repo if you reference this code. Thanks!

---