Feature Extraction Based on Moment Invariants for Handwriting Recognition

Abstract

This paper presents an experimental evaluation of the effectiveness of various techniques based upon moment invariants (Hu, 1961). The features that have been extracted are based on moments, image partition, principal component axes (PCA), correlation coefficient and perturbed moments. In image partition method, the image is divided into four parts with three different ways. The principal component axes (PCA) method has been used to balance the distribution of pixels in different regions of the image. Correlation coefficient provides dependencies of different moments on each other. In perturbed moment method, moment invariants are computed by small perturbation in image and information is extracted from the perturbation. All techniques have been applied on 2000 handwritten Devanagari numerals. The Gaussian distribution function has been adopted for classification. The success rate can be enhanced to 92% using image partition method