中文摘要：

针对大尺度LBP算子模式种类过多,导致数据量过大,直方图过于稀疏的问题,提出一种基于局部二值模式与K-均值的人脸识别算法。利用16像素的大邻域LBP算子描述人脸图像的纹理特征,通过K-Means聚类算法将所有LBP编码映射到最近的聚类中心,并建立查找表,快速统计出人脸图像的LBP直方图特征,将其作为人脸的鉴别特征,用于分类识别。实验结果表明,该算法具有较强的人脸图像描述能力和可鉴别性,在AR人脸数据库中取得了很高的人脸识别率,对时间、表情及光照的变化具有较高的鲁棒性。

英文摘要：

Because large-scale LBP operator contains too many patterns, the amount of data is usually large, especially when the number of neighbors is more than 8. A face recognition algorithm was proposed which fused the local binary pattern （LBP） and K-Means to solve this problem. Firstly, the LBP operator with 16 neighbors was used to describe the texture feature of the face image. Then K-Means clustering mapped all the LBP codes to the nearest cluster center, and a lookup table was established. Finally, fast statistics on the LBP histogram was used as the face discriminant feature. The experimental results show that the proposed method is possessed of strongly descriptive and discriminable abilities. It is robust to face expressions and illumination variations, and can achieve excellent performance.