摘要
传统LBP模式在提取图像的纹理特征时,没有对图像中的不同子块加以区分。一般情况下图像的不同子块包含的纹理信息不尽相同,不能真实地反映图像纹理的变化情况。为了解决传统LBP算法在人脸识别过程中产生的直方图维数过长、鉴别力不高、对噪声反应敏感等问题,提出一种基于对数能量熵与LBP特征提取的人脸识别方法。首先将一副人脸图像分成互不重叠的大小相等的子块,然后计算每个子块的LBP直方图,同时对每个子块计算对数能量熵值;其次把每个子块的LBP直方图特征与对数能量熵值组合成一个新的特征向量;最后,将每个图像块的特征向量连接成一个全局的特征向量,将该特征向量用作分类识别。基于YALE人脸库,ORL人脸库和FERET人脸库的实验结果与数据分析表明,文中提出的算法能够更加准确地提取图像的特征信息,有效地提高了人脸识别率。
Traditional LBP does not distinction different sub-block in the image when extracting texture features of the image.General speaking,different sub-block of the image contains different texture information,can not truly reflect the change of image texture.In order to solve the problems in the traditional LBP algorithm,such as histogram dimension is too long,the resolution is not high and sensitive response to noise,this paper puts forward a kind of feature extraction logarithmic based on log energy entropy and LBP for face recognition.In the proposed algorithm,it first divides a face image equal-sized non-overlapping sub-block,and then calculate the LBP histogram for each sub-block and computes log energy entropy;and then,each sub-block LBP histogram feature in combination with the log energy entropy become a new feature vector;finally,each feature vector of sub-block connects into a global feature vector,the feature vectors are used for classification.The experiments are done on YALE face database,ORL face database and FERET face experimental results and database analysis show that the proposed algorithm can extract more accurate feature information of image with higher recognition rate.
作者
付余伟
胡方媛
晋杰
陈熙
FU Yu-wei;HU Fang-yuan;JIN Jie;CHEN Xi(School of Information Engineering and Automation,Kunming Science and Technology University,Kunming 650500,China)
出处
《信息技术》
2017年第7期1-4,共4页
Information Technology
基金
国家自然科学基金(61262040)
云南省应用基础研究计划项目(KKSY201203062
KKS0201503018)
关键词
人脸识别
特征提取
局部二元模式
对数能量熵
face recognition
feature extraction
local binary pattern
log energy entropy
