Contourlet Transform Based Algorithm of Shadow Compensation for Face Recognition

摘要

This paper researches of the new multi-scale geometric analysis tool—Contourlet and proposes a new Contourlet multi-threshold method of shadow compensation for uneven illumination face images. The proposed algorithm combines hard threshold with 2d shadow compensation method and selects proper thresholds depending on the sub-band layers of Contourlet transform. It takes full advantage of the shadow elimination with Contourlet multi-threshold method and the 2D shadow compensation method, so that it could obtain the information of the shadow field and non-shadow field. Experiments are carried out using the Yale B database and the results demonstrate that the face images dealt with the proposed method have good subjective vision and impersonal identify ratio. For images under different illumination angles, compared with 2D shadow compensation algorithm, the proposed method has an average recognition ratio increase of 21.20% to 55.84% in extreme condition.