Handwritten Arabic Similarity

Overview

A computer vision project that takes a query image of a handwritten Arabic character, extracts features with a pretrained EfficientNet-B3 model, reduces dimensionality with PCA, and returns the top 5 most similar characters using cosine similarity.

Problem

Finding visually similar handwritten Arabic characters requires more than simple pixel comparison — it needs learned visual features.

Solution

Build a PyTorch pipeline with feature extraction, PCA, and cosine similarity to rank the closest matches from a handwritten Arabic character dataset.

My role

Dataset handling, feature extraction pipeline, similarity ranking, and evaluation visualization.

Key features

• Handwritten Arabic character search
• EfficientNet-B3 feature extraction
• PCA dimensionality reduction
• Cosine similarity ranking
• Top-5 visual match results
• KAUST Academy Exam — Question 2