Offline Handwritten MODI Character Recognition Using GoogLeNet and AlexNet

Abstract

“MODI lipi” is one of the scripts used to write religious scriptures of Maharashtra in Western India and it was also the official script for the Maratha administration from the 17th century to the middle of the 20th century. This cultural treasure, “MODI-manuscript,” speaks about the history of its time. Although it has immense importance as a source of inspiration and information to the present generation, very few people know this “lipi”. The field of Handwritten Character Recognition offers a scope to develop a recognition system for MODI to make it easy to learn. However its structural characteristics demand a special approach. Deep Convolutional Neural Networks (DCNN) has shown their remarkable potential in distinct feature extraction and classification of characters. So, in this paper we are focusing primarily on the performance evaluation of DCNN and their comparative study for MODI handwritten character recognition. Networks are evaluated based on trainable parameters, training time and memory consumption. Later, the tuned networks are also tested for transformed MODI dataset. The result shows the effectiveness of deep learning approach on Handwritten MODI character recognition.