Offline MODI script character recognition using deep learning techniques

Abstract

 The MODI script, historically used for writing Marathi and other languages in Western India, holds immense historical importance. It is essential for maintaining and comprehending this cultural legacy to recognize characters from ancient MODI texts. A major obstacle, which frequently affects the accuracy of character recognition, is the restricted availability of MODI script datasets. With the goal of achieving the maximum accuracy while minimizing noise interference, this paper presents an extensive study of MODI script character recognition using modern Machine Learning (ML) and Deep Learning (DL) techniques. To identify characters in the MODI script, we use a variety of ML and DL models, such as Convolutional Neural Networks (CNNs), VGG 16, ResNet, KNN, and SVM.The dataset contains a lot of similar characters so distinguishing them and increasing the accuracy of the algorithm used in order to get the best result.  These models are trained on the dataset, enabling them to learn and distinguish between similar-looking characters effectively. The models are evaluated based on their accuracy.