/arabic-calligraphy-classification

A MATLAB code to classify an image of Arabic text into different calligraphy styles

Primary LanguageMATLAB

Arabic Calligraphy Style Classification

This MATLAB code was developed as a project for the course Statistical Learning - Classification (STAT 841) at University of Waterloo.

Steps to run:

  1. Download Husni Al-Muhtaseb's PATS-A01 dataset from here. This code has been tested for classfication of Akhbar, Thuluth, Naskh, and Andalus. About 200 images of each style for the training set is sufficient. Put these in the ./*/train/ folders
    Put rest of the images in the ./*/test/ folders

  2. Run extract.m. This extracts the feature set for each image into a .dat file.

  3. Run train_nn.m. This fits a single layer neural network to the training data.

  4. Run test_nn.m. This displays the test error, and summary of misclassifications.

Also, run testing.m to understand how features are being are extracted.

Thanks to Alessandro Mannini for his Freeman Chain Code MATLAB function