一种基于改进LBP特征的人脸识别  

Face recognition method based on improved LBP algorithm

作　　者：郭贺飞 陆建峰[1] 董忠汶 

机构地区：[1]南京理工大学计算机科学与工程学院,江苏南京210094 [2]江苏省有色金属华东地质勘查局地质信息中心,江苏南京210007

出　　处：《现代电子技术》2015年第4期98-101,105,共5页Modern Electronics Technique

基　　金：项目国家自然基金(30971698);2011年永州市科技计划项目

摘　　要：传统的LBP算子只利用了局部的信息,而忽略了全局信息。MB_LBP算子虽然充分考虑了全局信息,但对局部信息的表示不足。在此提出一种改进后的LBP特征的人脸识别方法,改进后的LBP算子不仅能够利用局部特征,同时也兼顾了全局信息。该方法首先将人脸图像分块,对于每个分块,计算LBP特征,对于得到的LBP特征,根据其中心像素和分块灰度均值关系重新进行计算得到改进后的LBP特征,最后采用最近邻分类器进行识别。在ORL和YALE标准人脸数据库上的实验表明,改进后的识别效果优于使用传统LBP算子和MB_LBP算子。采用改进后的LBP算子,能够明显提高识别率,在ORL和YALE的实验显示能提高3%~8%左右的识别率。The traditional LBP（local binary pattern）operator can capture the local information only,but ignores global in-formation. MB_LBP operator expresses local information insufficiently,though it fully considers global information. A new facerecognition algorithm based on an improved LBP operator is proposed in this paper. The improved LBP operator can capture notonly the local features,but also the global features. In this method,the original face image is divided into small blocks,andthen the LBP features of each block are calculated. According to the relation of center pixel and gray mean value of each block,the obtained LBP features are re-calculated to get the improved LBP features. Finally,the classification is performed by a nearestneighbor classifier. The result of the experiment with ORL and YALU standard face database shows that the recognition efficien-cy of the improved LBP operator is better than traditional LBP operator and MB-LBP operator,the recognition rate is increasedby 3%~8%. The application of the improved LBP operator can improve the recognition efficiency obviously.

关 键 词：LBP 人脸识别 特征提取 人脸数据库 

分 类 号：TN919-34[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]