Kernel Holistic Orthogonal Analysis of Discriminant Transforms

Kernel method is an effective technique in extracting nonlinear discriminative features. In this paper, we propose a new color face image recognition approach based on kernel holistic orthogonal analysis (KHOA) of discriminant transforms. Original color face images are mapped to high dimensional feature space by kernel function, then extract discriminant transforms of red, green, blue color image in turn by using Fisher criterion and then reduce the correlation of red, green, blue color image in the pixel level. Experimental results on AR public color face image databases demonstrate that the proposed approach acquires higher recognition rates than linear color face image holistic orthogonal analysis of discriminant transforms method.