RECOGNITION OF HANDWRITTEN URDU TEXT USING THE BERT MODEL WITH HIGH ACCURACY

Abstract: Urdu handwritten text recognition presents unique challenges due to the inherent complexity of its script, including cursive connectivity and context-sensitive character forms. These challenges are amplified in handwritten text, where variations in individual writing styles, age, gender, and regional influences introduce significant inconsistencies. A small work has been made in Urdu handwriting recognition, however, existing methods struggle with unconstrained handwritten Urdu text. Also the state-of-the-art approaches offers high computational complexity while using conventional deep learning models. Hence in this research paper, we present a new approach on the recognition of unconstrained handwritten Urdu text using BERT model high recognition capability. The proposed approach outperforms the existing deep learning CNN models with respect to accuracy and error rate.