Automatic Decision Method of Effective Transform Coefficients for Face Recognition

In this paper, we propose a novel face recognition method using the effective transform coefficients of face images. The method is based on extraction of effective vector in Discrete Cosine Transform (DCT) domain and Linear Discriminant Analysis (LDA) of effective vector. In general, face images have characteristics that they show larger energy congestion in horizontal frequency coefficients than in vertical or diagonal frequency coefficients. However, many previous methods have shortcomings that they don't utilize the facial energy characteristics. To overcome shortcomings above, the proposed method selects the effective coefficients of the face in DCT domain and then extracts feature vector through LDA analysis on DCT coefficients. Experimental results show that our method has improvements of recognition performance over the previous methods.