ABSTRACT

Recent years, as a research topic valued in both theory and application, face recognition has been concerned and emphasized by more and more researchers. Various methods for face recognition emerge one after another, and PCA (Principal Component Analysis) is one of these methods (Matthew et al., 1991). Basic principle of conventional PCAs is to construct an eigenface space by using K-L transform (Ming-Hsuan et al., 2002) which can extract the main features from a face, and to obtain a set of projection coefficients by projecting the tested image onto this space, and then the recognition can be achieved by making comparisons among face images. Such a method lets the MSE (mean square error) before and after the compression minimized, and the transformed lowdimensional space has good performance in resolution. However, when processing a face image, this method extracts only global features which will be largely affected by light conditions and facial expression changes. Consequently, its recognition effect is not so good.