Enhanced Deep Learning Techniques to Classify Tamil Handwritten Characters

Tamil handwritten document digitization is becoming increasingly dependent on exploit of handwritten character identification. Tamil is solitary of the mainly used traditional languages within India, especially in south. It can be exceedingly difficult to detect a non-digitized character because tamil has such a great and multipart character set. The Inception model and other convolution neural networks aid in classification of various type of objects on images. Here, symmetric and asymmetric structure blocks to facilitate the model itself include convolutions, average, max pooling, concatenations, dropouts, and completely linked layers. The batch normalization, which is used broadly in the model and is also functional to the inputs for activation of the function. SoftMax is used to compute loss. To differentiate handwritten characters for better classification, we employed Inception. In this article, we evaluate the handwritten character identification for tamil by means of ResNet, Inception v3, VGG16, and VGG19. There are 1609 test images and 3901 train images in the 156-class dataset.