Isolated Bangla Handwritten Character Classification using Transfer Learning.

Bangla language consists of fifty distinct characters and many compound characters to be named. Several notable studies have been performed to recognize the Bangla characters both as handwritten and optical characters. Our approach is to use transfer learning to classify the basic distinct as well as compound Bangla Handwritten characters avoiding the vanishing gradient problem. Deep Neural Network techniques such as 3D Convolutional Neural Network (3DCNN), Residual Neural Network (ResNet), and MobileNet have been applied to generate an end-to-end classification of all the possible standard formation of the handwritten characters in Bangla language. Bangla Lekha Isolated dataset is used to apply this classification model, which has a total of 1,66,105 Bangla Character images sample data categorized in 84 distinct classes. This classification model achieved 99.82% accuracy on training data and 99.46% accuracy on test data. Comparison has been made among the various state-of-the-art benchmarks of Bangla Handwritten Characters classification, which shows that this proposed model got better accuracy in classifying the data.