Towards race-related face identification: research on skin color transfer

We developed a novel skin-color transfer algorithm in order to explore the potential information from human racial groups to improve the face identification. We attempt to find a general technique for transferring human skin colors of different races. The idea is to match the intensity values of two images and transfer the entire mood of the source image to the target image. Further enhancement can be achieved in two ways. One is facial multi-region subdivision to limit the color transfer between corresponding regions. The other is to use "region growing" method based on hue values for color transfer in sub-areas, such as lip, where H channel is dominant in determining pixel colors. The technique appears to be quite simple and results turn out to be encouraging. Our new algorithm is analyzed and compared to other existing algorithm, and evaluated through the psychological study on race related identification. The experiment shows that skin color is amongst one of the factors for race identification, but not the dominant one. This is the first preliminary step as an initial screening process towards the race-related face identification.