Online recognition of Malayalam handwritten scripts — A comparison using KNN, MLP and SVM
K B Baiju, K Sabeerath · 2016
This paper experiments writer independent OHCR for Malayalam script using KNN, MLP and SVM classifiers. The key aspect of the work is to identify the best classifier based on recognition accuracy for a set of features. The system is trained using a database of 44 character classes with 100 samples per class. Accurate Dominant Points, Aspect Ratio, Start-End Octants and Intersections are used as features. The feature vector has been reduced to minimize training and testing time. The system reported an average recognition rate of 95.12 for SVM with RBF kernel, 93.17 for ANN with MLP and 90.39 for KNN. And the results interprets that, SVM with RBF is the best classifier for the features selected in the work.