Handwritten Character Recognition for South Indian Languages Using Deep Learning

Character recognition is a crucial component of many applications, including document processing, handwriting recognition, and optical character recognition (OCR). We present an OCR model-based character recognition workflow in this research that involves feature extraction from the input images, model comparison, and performance assessment. We compared the performance of four OCR models, including FAST R-CNN, Efficient Net B7, VGG16, and VGG19, on Kannada and Malayalam handwritten character datasets. The results showed that Efficient Net B7 outperformed the other models with the highest accuracy. We also analysed the performance of each model in terms of precision, recall, and F1-score. Our results showed that FAST R-CNN was second with accuracy of 94.55%, Efficient Net B7 was first with accuracy of 96.66%, VGG19 was third with accuracy of 92.23%, and VGG16 was last with accuracy of 91.38%. The workflow provides a structured approach for developing and evaluating OCR models for character recognition tasks in Indian scripts. The findings can be useful for various applications that require accurate and efficient character recognition. Future work could focus on exploring other deep learning architectures and datasets for character recognition in Indian scripts.