This paper presents a Manipuri Handwritten Character Recognition System for Meetei Mayek script. The work comprises of development of standard Database for isolated and connected handwritten characters for Meetei Mayek. Experiments are conducted using the database, where State-of-Art feature extraction techniques viz. Feature Extraction using a pretrained Convolutional Neural Network (CNN) VGG16 and Histogram of Oriented Gradients (HOG) are used. Three different types of Support Vector Machines (SVM) using linear, polynomial and radial basis function kernels have been compared achieving Recognition Accuracies 93.12%, 94.27% and 95.03%. Further the individual Support Vector Machine Classifiers have been combined using Ensemble Voting technique which has led to an increase in accuracy of 96.21%.
Volume 11 | 03-Special Issue
Pages: 1267-1273