Abstract: LBP（Local Binary Pattern） operator is effective method used in texture analysis and face recognition，but not considering the contrast value between pixels，it doesn’t represent the important texture feature.In order to overcome this disadvantage，the paper proposes an improved LBP method—LMCP（Local Multi-layer Contrast Pattern）.This method uses preprocessing to limit the illumination variation.Afterward，the contrast values between center pixel in local area and its neighbor pixels are calculated，and the value range between the max contrast value and the min contrast value is divided into several layers.And then，every contrast value can be mapped into a certain layer.So，using the rule which is similar to LBP，can get a LMCP feature composed of several decimal values.Additionally，a statistics mapping method is adopted to reduce the expanded feature dimension.Plentiful experiment data improve the priority of LMCP than LBP.

Key words: face recognition, illumination normalization, Local Binary Pattern（LBP）

摘要： LBP算子是在人脸识别和纹理分析领域比较成功的一种方法，但是由于没有考虑像素值之间的对比度，因而丢弃掉了重要的纹理特征。提出了一种LMCP方法，解决了LBP方法的这个缺点。该方法先通过预处理，将光照变化控制在一定范围内，然后求得局部区域中心像素点和邻居像素点之间的对比度值，并将其最大值和最小值之间的值域划分为若干个层次，将每个对比度值映射到某个层次上，再使用LBP类似方法获得若干个数值组合而成的LMCP特征值。此外，还使用了统计映射的方法进行降维。实验结果证明了LMCP方法比LBP方法更加有效。

关键词: 人脸识别, 光照正常化, LBP算子