TamilLite-Gan: A lightweight graph attention network for handwritten Tamil character recognition using enhanced single shot optimization
TL;DR
A novel deep learning–based handwritten Tamil character recognition framework that achieves superior recognition accuracy with reduced computational complexity compared to existing methods, highlighting its effectiveness and suitability for practical OCR applications.
Abstract
Handwritten character recognition plays a crucial role in optical character recognition systems, particularly for low-resource and structurally complex scripts such as Tamil. Despite significant advances in deep learning, accurate recognition of handwritten Tamil characters remains challenging due to large variations in writing styles, complex character structures, and high computational requirements of existing models. Conventional approaches often suffer from limited detection accuracy, inefficient parameter tuning, and poor generalization in real-world scenarios. To address these challenges, this paper proposes a novel deep learning–based handwritten Tamil character recognition framework. The proposed model employs an Adaptive Single Shot Detector (ASSD) for efficient and accurate character localization, with its performance further enhanced through hyperparameter optimization using the Enhanced Running City Game Optimizer (ERCGO). For classification, a Multi-scale Dilated Graph Attention Network (MDGANet) is introduced to effectively capture spatial and structural relationships within handwritten Tamil characters. Extensive experimental evaluations conducted on benchmark datasets demonstrate that the proposed framework achieves superior recognition accuracy with reduced computational complexity compared to existing methods, highlighting its effectiveness and suitability for practical OCR applications.