Comparative Analysis of Deep Learning Models for Kannada Handwritten Character Recognition

This study presents a comprehensive comparison of multiple innovative techniques of deep learning for recognizing handwritten characters in Kannada, an Indian language. To facilitate this evaluation, a standardized benchmark dataset for Kannada handwritten character recognition (KHCR) was considered. Handwritten character recognition has wide-ranging applications, including intelligent character recognition, language translation, and number plate recognition. A large-scale classification of Kannada handwriting at the character level would enable users to convert their handwritten text into printable text, tailored to their needs. Initially, a dataset comprising 699 classes was constructed to train the deep learning models. Also, these models were tested and validated using a separate test dataset. The proposed work investigates many deep learning frameworks, such as CNN architectures including VGG19, ConvNeXT, ResNet50, DenseNet121, DenseNet201, to address the multi-class problem involving 699 classes.