Human face recognition with different statistical features

Abstract

This paper examines application of various feature domains for recognition of human face images to introduce an efficient feature extraction method. The proposed feature extraction method comprised of two steps. In the first step, a human face localization technique with defining a new parameter to eliminate the effect of irrelevant data is applied to the facial images. In the next step three different feature domains are applied to localized faces to generate the feature vector. These include Pseudo Zernike Moments (PZM), Principle Component Analysis (PCA) and Discrete Cosine Transform (DCT). We have compared the effectiveness of each of the above feature domains through the proposed feature extraction for human face recognition. The Radial Basis Function (RBF) neural network has been utilized as classifier. Simulation results on the ORL database indicate the effectiveness of the proposed feature extraction with the PZM for human face recognition.