Abstract: The aim of this paper is to solve the problem of large variation in illumination in face recognition.The Difference Of Gaussian（DOG） filters are adopted in the preprocessing step to correct face images in the original dataset under extremes of illumination conditions.The resulting image dataset is combined with the original dataset to provide more sufficient information to maintain system performance for normally illuminated images.Sparse Representation Classification（SRC） instead of traditional classification is introduced to improve the recognition rate at a low false acceptance rate and solve the problem that faces can not be recognized because of the variations in lighting.Experimental results on the Yale-B face database，CMU-PIE face database and ORL face database show that the proposed method improves the recognition rate under different illumination.

Key words: face recognition, varying lighting, difference of Gaussian, Sparse Representation Classification, low false acceptance

摘要： 主要解决人脸识别中因光照变化导致误识或者拒识的问题。使用DOG（高斯差分变换）对原始人脸图像样本集（A）进行处理，将滤波后的人脸图像样本集（B）加入到原始样本集（A）中，采用了新的方法将样本集A和B进行融合，则既对极端光照条件下人脸图像进行了矫正，又不影响正常光照条件下的人脸识别。在分类阶段，引入了SRC（Sparse Representation Classification）分类器代替传统分类器，提升了在低错误接收率下的识别率，改善因光照剧烈变换而导致的无法识别或者误识的情况。在公开人脸库Yale-B、CMU-PIE以及ORL上的实验结果表明，该方法在不同光照条件下可以提高识别率，改善拒识和误识情况。

关键词: 人脸识别, 光照变化, 高斯差分变换, SRC分类器, 低错误接受率