Automating the Evaluation of Urdu Handwriting for Novice Writers With Localized Feedback

Handwriting is one of the basic skills crucial in communication and early childhood development. It plays a major role in the psychomotor and cognitive growth of children. Therefore, it is required to devise interesting and modern ways to teach this skill. A handwriting tutoring system designed to teach Urdu transcription engagingly and efficiently can help to solve this problem. Essential to these teaching systems is the quality evaluation module that analyzes and evaluates the quality of transcription based on legibility and writing process and provides feedback to the writer. This feedback can not only help the young students to improve and identify their mistakes but can also help to repair the handwriting skills of any individual wanting to improve their transcription abilities. The existing handwriting evaluation methods are either designed for Arabic writing style, or those designed for Urdu need more comprehensive feedback on the quality of handwriting. Therefore, it is the need of time to develop a technology-assisted educational system for teaching Urdu handwriting that evaluates and provides intelligent feedback to the learners. In this work, we have proposed an innovative model for evaluating different aspects of Urdu handwriting and providing qualitative and quantitative feedback. The system takes on real-time writing samples, and its transcription profiler module then individually evaluates six aspects of handwriting, including legibility, shape formation, the direction of writing, proportion with the baseline, count, and order of strokes. Based on the evaluation, the feedback generator module provides overall feedback and detects mistakes made while writing. A machine learning-based hybrid architecture is constructed that combines spatio-temporal and structural features to do a quality analysis of each aspect of the transcript. The contribution of this work is twofold: first, to develop a standard for evaluating the quality of Urdu handwriting or any cursive writing script, and second, to create an architecture that relies on machine learning-based models to provide detailed and intelligent feedback while learning handwriting. The system’s performance is verified by comparing it with the test data evaluated by real teachers. It achieved an average of 97% accuracy among all character classes in all handwriting aspects. The results show that the system is effective in identifying and correcting mistakes, thus helping individuals seeking to improve their handwriting skills.