Abstract: In order to improve the accuracy of face recognition under complex illumination, a face recognition algorithm based on improved single scale Retinex and Local Binary Pattern（LBP） is proposed. Firstly, bilateral filtering is used to replace Gaussian filtering in the Retinex to process face images. At the same time, Laplacian of Gaussian（LoG） and normalization are applied to extract the edge details features of face images. The two processed images are fused by the weighting method of standard deviation. Then, LBP is used to extract features from fused images. Finally, Sparse Representation based Classification algorithm（SRC） is used to classify. Experiments on AR and Yale B+ face database show that the illumination robustness of face recognition is improved under complex illumination. It can achieve good recognition results under less training samples and complex illumination.

Key words: feature fusion, complex illumination, feature extraction, face recognition

摘要： 为了提高复杂光照条件下人脸识别准确率，提出一种基于改进单尺度Retinex并结合局部二值模式（LBP）的人脸识别算法。首先，利用双边滤波代替Retinex的高斯滤波处理人脸图像，同时使用高斯-拉普拉斯（LoG）及归一化处理提取人脸图像的边缘细节特征，采用标准差的加权方法将两幅处理后的图像进行特征融合，然后使用LBP对融合后的图像进行特征提取，最后通过稀疏表示（SRC）算法对数据样本进行判别归类。在AR和Yale B+人脸库上的实验测试表明，提高了复杂光照下人脸识别的光照鲁棒性，在训练样本较少、光照复杂环境下能取得较好的识别效果。

关键词: 特征融合, 复杂光照, 特征提取, 人脸识别