基于LMSRBP和软直方图交的人脸表示与识别算法  

Face representation and recognition algorithm based on LMSRBP and soft histogram intersection

作　　者：王超[1] 冯晓毅[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710129

出　　处：《电子设计工程》2013年第11期181-185,共5页Electronic Design Engineering

基　　金：国家自然科学基金(60875016)

摘　　要：针对传统局部二进制模式(LBP)存在的固有缺陷,即小尺度LBP算子无法反应人脸图像的宏观特征,大尺度LBP算子特征维数通常很高,提出一种局部多尺度多分辨率二进制模式(LMSRBP)的人脸表示算法。通过使用高斯金字塔得到一系列不同尺度不同分辨率的人脸图像,然后对这些图像使用同种LBP算子得到LMSRBP特征谱。该方法在不改变特征维数的情况下,能同时提取出图像的微观和宏观特征。在分类器设计方面,考虑不同面部区域对识别贡献不同的问题,提出软直方图交方法构建多弱分类器,最后集成所有弱分类器得到识别结果。在YaleB,ORL标准人脸库上的实验结果表明,该改进算法能显著提高人脸识别率。The traditional local binary pattern （LBP） has the inherent defects ：small scale LBP operator can not reflect the macro feature in face images, the dimensionality of the feature with a large scale LBP operator is usually very high. To solve this problem, a face representation method based on multi-scale and multiresolution local binary pattern （LMSRBP） is proprosed. Without changing the dimensionality of the feature, the new LMSRBP operator can extract the microscopic and macroscopic features in face images, through the use of Gauss Pyramid for a series of face images with different multiscale and resolution,. On the design of classifier, considering different facial regions contributed to identifying different problems, a soft histogram intersection method is presented to construct many weak classifiers, then integrates all the weak classifiers. The results show that, the improved algorithm improves both the face recognition rate and recognition rate.

关 键 词：人脸识别 LBP 尺度空间 LMSRBP 软直方图交 

分 类 号：TP931.4[自动化与计算机技术]