Face Recognition under Varying Lighting Based on the Harmonic Images
Qing Lai · Chinese Journal of Computers · 2006
The performances of the current face recognition systems suffer heavily from the variations in lighting. To deal with this problem, this paper presents an illumination normalization approach based on the harmonic images model. There are two steps in the algorithm: illumination estimation and illumination normalization. Benefiting from the observations that human faces share similar shape, and the albedos of the face surfaces are quasi-constant, the authors first estimate the nine coefficients of the low-frequency components of the illumination. Then the authors compensate the effect of the illumination by defining two canonical images, the texture image and the difference image. The texture image is defined as the ratio between the input image and its irradiance and it is illumination invariant. The difference image is defined as the difference between the input image and the image of the average face model under the same illumination. Therefore the effect of the illumination is weakened and the valuable information for recognition is still encoded in the difference image. The experiments on the CMU-PIE face database and the Yale B face database have shown that the proposed method improve the performance of a face recognition system significantly when the probes are collected under varying lighting conditions.