Multiresolution eigenspace and fisherspace face recognition

Abstract

The paper introduces a multiresolution face recognition system using features extracted from eigen and fisher spaces. Discrete wavelet transform has been used to generate images at varying resolutions. Two methods are proposed to combine the features extracted from a set of face images with varying resolution. The first method is called the multiresolution feature concatenation (MFC), where we use principal component analysis (PCA) and linear discriminant analysis (LDA) as a dimensionality reduction process on each subband. Then the resulting projection coefficients of each subband are concatenated to perform classification. The second method is called the multiresolution majority voting (MMV), where the classification are done separately on each subband and then the majority voting is applied for making decision. The results obtained from both of the methods show promising results and MMV approach outperforms the MFC approach. Moreover, the two methods outperform the conventional PCA and LDA approaches respectively approach.