关键词: 人脸识别, 光照分类, 垂直积分投影

Abstract: A novel approach based on illumination categorization is proposed to solve the problem of illumination variation in face recognition，learning from the thoughts of “divide and conquer”.According to the gray level distribution of faces，illumination is grouped into three categories：Non-polarized，left-polarized and right-polarized illumination subsets.Then image processing and face recognition are implemented respectively in each subset.The experimental results on YALEB face database show that the proposed approach which is no need for illumination normalization weakens the impact of uneven illumination on face recognition effectively，and the recognition accuracy is improved from 86.7% of pre-categorization to 99.6%，furthermore，the computational complexity is reduced.This method can be applied in face recognition under variable lighting.

Key words: face recognition, illumination categorization, vertical integral projection