A Deep Learning Approach to Recognize Handwritten Telugu Character Using Convolution Neural Networks
Anupama Angadi, V. Valli Kumari, Satya Keerthi Gorripati · SSRN Electronic Journal · 2018
Automated character recognition is one of the most vital components which enable a data processor to distinguish letters and digits possibly using contextual data. Various attempts at resolving this problem using different selections of classifiers and features have been established and still the problem remains challenging. In the proposed work, we suggest and evaluate a classic Convolutional Neural Network (CNN) for the identification of online Telugu characters. The network comprises of 4 layers, one with 5 X 5 and remaining with 3 X 3 kernels and ReLU, softmax activation functions, followed by max pooling and two dense layers. The final layer has 168 outputs, matching to the classes considered: vowels, consonants. To train and estimate the CNN, we preowned 45,133 images written by many telugu writers. A qualified analysis proved the efficiency of the proposed CNN against previous methods in an interesting dataset. On test dataset, the classification technique provided 92.4% accuracy. The conclusion is improved than some recently proposed literature used for the identification of online handwritten Telugu characters.