A face recognition algorithm robust against illumination variations - estimation of surface reflectance from face image

摘要

The authors propose a face recognition algorithm robust against the illumination variations, using an average face shape. This algorithm uses the surface reflectance as a matching feature, and it is possible to recognize faces from minimal registration data: single face image for each person. By using the average face shape for each local area, the deviation from the individual face shapes is suppressed and the estimated error of surface reflectance is decreased. The correct recognition rate for images of 41 persons with 5 illumination conditions was 97.0%, which was much higher than that by the edge-based matching method 82.9%.