Discriminant Tchebichef based moment features for face recognition

Face representation using small number of features with highest discriminatory measure is vital in the development of a face recognition system. In the holistic based approach, features are extracted from the global appearance of a face. Fisher´s Linear Discriminant Analysis (FLD) is a popular holistic based feature extractor. However, it is subject to several limitations such as small sample size problem and heavy computation due to large scatter matrix. In this paper, a new approach is introduced to remedy this problem. Firstly, discrete orthogonal Tchebichef moments are computed to summarise the information lying in a large sized facial image. Secondly, the scatter matrices are calculated on the Tchebichef moments to obtain an optimised discriminant feature using FLD. The proposed method is tested on ORL database, which consist of 40 subjects with 400 images. Highest recognition rate of 96.5% were obtained using only 29 features.