Hybrid framework for Arabic handwriting recognition using deep learning and NLP

Abstract

The scientific community continues to face challenges in developing effective solutions for recognizing handwritten Arabic words, primarily due to limited research addressing the unique complexities of the Arabic language, such as its cursive nature and the variability in handwriting styles across individuals and over time. This study proposes a hybrid framework that integrates deep learning and natural language processing to enhance handwritten Arabic word recognition. The model combines Convolutional Neural Networks (CNN) for extracting visual features with Long Short-Term Memory (LSTM) networks for processing the features extracted from CNN, further improved by an N-gram-based correction mechanism. The framework was trained and evaluated using the Arabic Handwritten Database (AHDB), comprising 6,615 images of 63 dis- tinct words written by 100 different writers. While not fully overcoming the inherent challenges of handwritten Arabic recognition, the proposed approach yielded promising results, achieving an accuracy of 84%, with additional improvements from the correction mechanism, demon- strating its potential to advance the field and laying the groundwork for future enhancements.