Efficient Gesture-Based Recognition of Tifinagh Characters in Air Handwriting with a TinyDL Model
Ismail Lamaakal, Naima El Mourabit, Khalid El Makkaoui, Ibrahim Ouahbi, Yassine Maleh
Abstract
Gesture-based recognition of Tifinagh characters in air handwriting presents a novel approach to intuitive input methods, particularly beneficial for users in regions where Tifinagh scripts are prevalent. In this paper, we propose an efficient solution leveraging TinyDL (Tiny Deep Learning) models to achieve remarkable accuracy in character recognition. Our TinyDL model, developed through rigorous experimentation and refinement, achieves an impressive accuracy of 97.8%. Furthermore, we address the challenge of model size by implementing advanced pruning techniques, resulting in a drastic reduction from the original 648KB model to a compact 53KB TinyDL model. This significant compression ensures efficient deployment and utilization, particularly in resource-constrained environments such as mobile devices and embedded systems. Our findings underscore the potential of TinyDL models in enabling high-performance recognition systems while mitigating the overhead associated with model size, paving the way for practical deployment of gesture-based Tifinagh character recognition solutions.