Efficient Tamil Character Recognition: A Comparative Study of Learning-Based Algorithms with ResNet-50, DenseNet121, and EfficientNet Architectures

Computer vision finds practical application in handwriting recognition, a technology widely employed to interpret handwritten characters on documents, bank cheques, and postal addresses. Several studies have explored the identification of Tamil characters. Deep learning techniques have been used recently to implement Tamil character recognition. This research presents a comparative analysis of Tamil language character recognition using ResNet, DenseNet, and EfficientNet. Architectural enhancements like skip connections, designed to mitigate the problem of small gradient value and facilitate training of deeper networks, are commonly integrated into these models. The experiment encompassed diverse datasets, and model Performance was evaluated with the use of criteria including recall, accuracy, and precision. The EfficientNet architecture achieves the highest precision (99.1%), recall (97.2%) and accuracy (98.7%) value when compared to other models.