Enhancing Arabic character recognition via feature engineering and PSO

Abstract

Accurate recognition of handwritten Arabic characters poses significant challenges, especially for non-native learners. With the increasing adoption of digital teaching and distance learning, there is a pressing need for efficient and robust automatic recognition systems for Arabic characters. This work proposes a novel approach to address this challenge. First, handwritten Arabic character images are processed to extract features using two pre-trained deep learning models: EfficientNet B2 and DenseNet 201. The extracted features from these models are then concatenated to form a comprehensive feature set. Subsequently, the Particle Swarm Optimization (PSO) algorithm is employed to identify the most relevant features from this concatenated set through an optimized feature selection process. Finally, the selected features are fed into a classical classifier for character recognition. The proposed approach achieves an accuracy exceeding 90%, demonstrating its effectiveness in recognizing handwritten Arabic characters.