A New Decision Tree for Recognition of Persian Handwritten Characters

Abstract

In this paper a binary decision tree, based on Neural Networks, Support Vector Machine and K-Nearest Neighbor is employed and presented for recognition of Persian handwritten isolated digits and characters. In the proposed method, a part of the training data is divided into two clustersusing a clustering algorithm, and this process continues until each subtree reaches clusters with optimum clustering, where the tree leaves are the final obtained clusters. According to the clustering results, classifiers such as ANN and SVM can perform correctly, therefore the decision tree can be built. A part of the test data is selected as validation data and in each node of the tree, a classifier with the highest recognition accuracy on validation data is selected. Recognition accuracy at 8, 20, and 33 clusters have been evaluated and compared with other existing methods. Recognition accuracy of 98.72% and 97.3% on IFHCDB database is obtained respectivelywhen 8-class and 20-class problems is assumed. Again 98.9% accuracy on HODA database is achieved.