Abstract: In order to reduce the light on the face recognition effect, a method of obtaining reflection coefficient through the multi-level self-quotient image is proposed. Firstly, source image as input image is enhanced through the method of non-linearity unsharp masking. It can get reflection coefficient and light component through self-quotient image method. Then, the paper takes the light component as a new input image and repeats the above operations. Finally, it can obtain the final reflection coefficient after all reflection coefficients got in each level is fusion. The results on YaleB face database show that the method can remove the influence of light, make the face recognition rate has improved to a certain extent.

摘要： 为了降低光照对人脸识别效果的影响，通过多层次自商图像的方法获得反射系数。将原图作为输入图像采用反锐化掩模滤波的方式进行增强，再通过自商图像方法获得反射系数和光照成分；把光照成分作为输入图像，重复以上操作，对每次得到的反射系数按比例融合得到最终的反射系数。YaleB人脸库上的实验结果表明，该方法能在一定程度上去除光照的影响，使人脸识别率有一定的提高。

关键词: 反射光照模型, 自商图像, 反锐化掩模滤波