Recognition of Kannada Characters in Scene Images using Neural Networks
Shahzia Siddiqua, Naveena Chikkaguddaiah, Sunilkumar S. Manvi · 2019
In recent years, there has been significant progress in recognition of text from scene images. However, very less work is done for Kannada language. The focus of this paper is to identify Kannada character set. Researchers have purported a number of deep learning procedures for natural scene text detection and recognition. In this work, basic transfer learning neural network classification models are trained with 1700 scene Kannada character images extracted from Chars74K standard dataset, and labelled individual folders as per character names. Stochastic gradient descent solver with different epoch, learning rate and momentum is applied. Also, the performance of Alex net model is improved by inducing a batch normalization layer in the hidden layers. The results show that for characteristic features of Kannada text, the combination of Vgg19 network model used with higher split ratio and stated training options and values gave 100% accuracy. Also, the modified Alex net model performed better with accuracy up to 96 percent and lower error rate of 0.0134.