Feature image generation using energy distribution for face recognition in transform domain

Abstract

In this paper, we propose a feature image generation method for face recognition. Feature extraction is done using three transforms viz. Discrete Cosine Transform, Slant Transform and Walsh Transform. Energy distribution defined as magnitude of effective information is used to create a feature image in transform domain by retaining high energy distribution coefficients. The proposed method consists of three steps. First, the face images are transformed into the frequency domain. Second, transformed coefficient matrix and energy distribution matrix is divided into three equal regions. Thresholds are selected in each region to retain the most significant features. Finally feature image is generated from these coefficients. Recognition is performed on generated feature images using Mahalanobis distance. Experimental results shows that the proposed method improve the face recognition rate as compared to previously proposed methods. © Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2012.