Enhancing Segmentation of Handwritten Arabic Texts Using Associative Auto-Encoders and Super-Resolution Techniques

Abstract

One of the biggest challenges in document image analysis is still the segmentation of thin and cursive handwritten text, particularly in complex scripts like Arabic. This study presents a novel method for improving segmentation performance by integrating an associative auto-encoder framework with super-resolution techniques. While super-resolution methods concentrate on converting low-resolution images to high-resolution formats, auto-encoders are great at de-noising, feature extraction, and dimensionality reduction. By combining the best features of the two approaches, the suggested system improves the accuracy of handwriting segmentation by expanding to the pixel level and reconstructing every detail. The experimental findings show that this method works, significantly improving the accuracy of segmenting thinner handwritten Arabic text. This development highlights the promise of merging auto-encoder designs with super-resolution methods to enhance document image analysis and handwriting detection.