Handwritten Persian Digit Recognition - Custom Classifier

Recognizes handwritten Persian (Farsi) digits with a from-scratch Minimum Distance Classifier - no black-box ML model.

Overview

Implements a custom Minimum Distance Classifier and applies it to the Hoda handwritten-digit dataset (the standard Farsi digit benchmark), demonstrating the classifier mechanics end-to-end.

Approach

1. Load & preprocess the Hoda digit images (normalize, flatten/feature-prep).
2. Build class prototypes - compute a representative (mean) vector per digit class.
3. Classify each test image by assigning it to the nearest class prototype (minimum distance).
4. Evaluate classification accuracy across the 10 digit classes.

Tech stack

Python, NumPy, Pandas. Jupyter Notebook. (Classifier implemented from scratch - no scikit-learn model.)

Repo contents

• HodaDigitRecognition.ipynb - data loading, custom classifier, evaluation.

Run it

pip install numpy pandas
jupyter notebook