Comparative Study and Implementation of Supervised and Unsupervised Models for Recognizing Handwritten Kannada Characters

The objective of this paper is to classify characters written in Kannada, a south Indian language. The characters are extracted from written documents, segmented and processed using various image processing techniques such as contrast normalization, denoising, thinning etc. using numpy and OpenCV. The dataset used is a combination of the Chars74k dataset and a custom-made dataset. Classifiers have been implemented based on K Nearest Neighbors, Support Vector Machines, Inception V3 and Convolutional Neural Networks using OpenCV and Keras and a comparison of their accuracies has been provided. The CNN classifier has been able to achieve an accuracy of 99.84% on the Chars74k dataset.