Deep Learning Approaches for Sinhala Handwritten Character Recognition

Sinhala is the primary national language of Sri Lanka and comprises a script of 60 characters, making it considerably more complex than alphabet-based writing systems such as English. Despite approximately 25–30 studies conducted since 1990, Sinhala handwritten character recognition remains a developing area within pattern recognition. Accurately recognizing Sinhala handwritten characters continues to be challenging due to the high visual similarity among many characters and the curved, variable nature of handwritten forms. As a result, existing recognition systems often fail to achieve the reliability required for practical deployment. This study focuses on deep learning–based approaches to improve Sinhala handwritten character recognition. Two Convolutional Neural Networks (CNNs) models and Gabor-initialized Convolutional Neural Networks are implemented and evaluated. The research investigates the effect of dropout on recognition accuracy and analyzes the influence of different Gabor filter parameters on the performance of the Gabor-initialized CNN model when applied to Sinhala character datasets. The CNN model I achieved a training accuracy of 96.33% and a testing accuracy of 90.14%, representing the highest reported performance for all 60 Sinhala characters compared with previously published methods. The optimal results were obtained with a dropout rate of 0.5. The Gabor-initialized CNN model achieved 95.15% training accuracy and 80% testing accuracy; although the training accuracy was slightly lower, this architecture exhibited faster convergence, reducing computational cost and training time. The findings indicate that the Gabor-initialized CNN model offers the best overall performance for Sinhala handwritten character recognition. The analyses of dropout effects and Gabor filter parameters provide useful insights for future model refinement and contribute to the development of more accurate and computationally efficient recognition systems for Sinhala script.