Online handwritten Gujarati character recognition using SVM, MLP, and K-NN
Vishal A. Naik, Apurva A. Desai - · 2017
In this paper, we present a system to recognize online handwritten character for the Gujarati language. Support Vector Machine (SVM) with linear, polynomial & RBF kernel, k-Nearest Neighbor (k-NN) with different values of k and multi-layer perceptron (MLP) are used to classify strokes using hybrid feature set. This system is trained using a dataset of 3000 samples and tested by 100 different writers. We have achieved highest accuracy of 91.63% with SVM-RBF kernel and lowest accuracy of 86.72% with MLP. We have achieved minimum average processing time of 0.056 seconds per stroke with SVM linear kernel and maximum average processing time of 1.062 seconds per stroke with MLP.