What kind of color spaces is suitable for color face recognition?

Human faces display different color and recent research efforts show that color is useful for face recognition. This paper presents a discriminant color space method and demonstrates its effectiveness using the FRGC Experiment 4 database and the AR database. We find that the discriminant color space is an approximate double-zero-sum (DZS) color space, and further show that a color space with DZS characteristic is more powerful than other color spaces without this characteristic. We finally provide the justification for why the DZS color spaces is more effective than non-DZS color spaces for face verification and recognition from the mutual correlation point of view.