Handwritten Character Recognition based on Competitive Neural Trees

Abstract

Competitive Neural Trees (CNeT) are widely used for classification in pattern recognition. This paper applies this technique for recognizing of Myanmar handwritten characters. This paper involves three important steps, typically preprocessing, feature extraction and classification. The aim of preprocessing is to improve the quality of the images for further processing. For the extraction of features, four of the 3×3 masks are applied to word images to extract horizontal, vertical, right and left-diagonal lines. Afterwards, decomposed images should be partitioned to eight sectors around the center of image and the number of black pixels in each sector calculated and normalized by dividing them upon the total number of black pixels in word images for feature vector. These feature vector from word images which are used in Competitive Neural Trees (CNeT) for recognition purpose. This paper introduces a global search method for the CNeT, which is utilized for training.