Off-line Myanmar Character Recognition based on Competitive Neural Trees

Abstract

Neural network classifier methods and decision trees are widely used in various pattern recognition research areas. Among them, printed character recognition still faces some issues in all languages. Myanmar sentences character recognition based on Competitive Neural Trees (CNeT) is proposed in this paper. CNeT performs hierarchical classification and apply competitive unsupervised learning at node label. The goals of Myanmar character recognition are to obtain better recognition accuracy rate and robust in geometric character shapes of different styles. Three main steps such as preprocessing, shape feature descriptors extraction and recognition are implemented in our experiment. Shape feature descriptors are extracted from preprocessed images which are used in Competitive Neural Trees (CNeT) recognition. This paper discusses a global search method for the CNeT, which is utilized for training.