Combine image quality fusion and illumination compensation for video-based face recognition

Illumination variation is a critical factor affecting the performance of face recognition especially in video-based face recognition. This paper addresses this problem in two aspects: (i) a novel approach is proposed to classify the illumination orientation so that the uneven illumination on face can be compensated with pertinence. Illumination direction map and plane-fit method are put forward to determine illumination orientations. Experimental results on several public available face databases show that our approach achieved considerable performance gain in contrast to other state-of-the-art methods. (ii) Image quality fusion rule is designed to reduce the influence which is caused by the degradation of facial image quality due to illumination compensation. The degradation can lower the recognition performance. Motivated by human cognitive process and combined with video features, the rule fuses the recognition result of every face video frame to opt best result. This quality fusion rule exhibits effectiveness for illumination compensation, and experimental results on the face video databases with varied illuminations demonstrated that the proposed approach achieved satisfactory recognition rate.