Abstract: A hybrid approach combining Multi-scale Block Local Binary Patterns （MB-LBP） algorithm and Fisherfaces method is proposed for robust face recognition in the uncontrolled conditions. MB-LBP algorithm is utilized to describe and extract texture information of face images, and corresponding histogram vectors is obtained. Fisherfaces algorithm is employed to deal with the dimension and classifier problems. Experiments are conducted with the K-Nearest Neighbor algorithm （KNN） on the ORL and Yale face databases respectively, and the proposed approach yields impressive results compared to other LBP based and MB-LBP based approaches.

Key words: Multi-scale Block Local Binary Patterns（MB-LBP）, Fisherfaces, face recognition, K-Nearest Neighbor（KNN）

摘要： 针对不可控条件对人脸识别的影响，提出一种基于多尺度分块局部二值模式（Multi-scale Block Local Binary Patterns，MB-LBP）和Fisherfaces融合的人脸识别算法。采用适当模块大小的MB-LBP算子提取图像的纹理结构信息，得到相应的特征直方图;通过Fisherfaces方法对MB-LBP提取的特征进行降维和分类;经由最近邻方法进行匹配识别。在ORL和Yale人脸库上进行实验，分别与其他基于LBP和MB-LBP算法的识别效果进行比对。实验结果表明，识别效率显著提高，鲁棒性更好。

关键词: 多尺度分块二值模式, Fisherfaces, 人脸识别, 最近邻法