Classification of Modi Lipi Characters Using Deep Learning Approach

Deep learning is a recently developed multi-layer learning computation for brain networks. India is a multicultural nation with a wide range of traditions, religions, and languages. One of the earliest types of written media is Modi script. The majority of the first recorded information on topics like astrology, Buddhism, diet, and medicine were written in the Modi script/lipi. One language that presents particular challenges for OCR is Modi. The majority of the Modi script are written in cursive, and many of the characters have a similar appearance. Using the Residual and InceptionV3 frameworks, we used deep learning techniques to offline handwritten character recognition. We test the effectiveness of both algorithms using a database of real handwritten characters that was created by various individuals. Currently, the experiment is done on ten characters with around 70 samples of each character. Total 660 images are used for training the model. The average training accuracy of ResNet (Residual network) and IneptionV3 is 96.94% and 98.23%, respectively. It is observed that InceptionV3 helps in reducing training time. InceptionV3 performs better during training for ten characters.