<![CDATA[This paper presents SiPuTiH, a Convolutional Neural Network (CNN)-based approach for handwritten Hijaiyah character recognition that addresses performance degradation caused by morphological variations in handwriting. The study employs a dataset of 1,680 handwritten images representing 30 Hijaiyah characters, where domain-specific data augmentation is applied solely during the training phase. The augmentation strategy incorporates controlled geometric and stroke-based transformations, including rotation, scaling, shear, slant variation, and stroke thickness adjustment, to model realistic handwriting diversity. The proposed CNN architecture consists of multiple convolutional layers with ReLU activation, max-pooling operations, and a softmax classifier. Experimental results show that the proposed method achieves an accuracy of 99.70%, with weighted precision and F1-score of 99.85% and 99.77%, respectively. Furthermore, the use of domain-specific data augmentation effectively reduces misclassification among visually similar characters, such as ta and tsa, demonstrating improved robustness and generalization capability.]]>
Copyrights © 2025
