一种基于LTP自适应阈值的人脸识别方法  

Research on the Local Adaptive Threshold Ternary Pattern for Face Recognition

作　　者：李伟生[1] 王立逗[1] 周丽芳[1] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《小型微型计算机系统》2014年第9期2099-2103,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61272195)资助;教育部新世纪优秀人才资助计划项目(NCET-11-1085)资助

摘　　要：局部三值模式(LTP)采用自定义阈值编码对局部二值模式(LBP)进行扩展,然而其阈值通常通过先验知识获取,因此,LTP阈值存在优化问题及普适性问题.针对这一问题提出一种基于LTP自适应阈值的人脸识别方法:首先在提取图像特征时计算各邻域内像素间对比值的离散度,通过该离散度定义自适应于各局部区域的LTP阈值;然后将提取到的特征分层进行编码,并引入一种信息熵权重的方法来融合各层的特征信息,最后使用卡方距离对样本的直方图进行匹配.在Extended Yale B和CMU PIE标准人脸库上实验,结果表明此方法对光照变化和噪声更加鲁棒,有效地提高了人脸识别的精度.The Local Ternary Pattern extends the Local Binary Pattern using the custom threshold coding, but it usually determine the threshold by the prior knowledge. This paper proposes a novel approach for the optimization of the threshold. First, when extract image feature information,it computes the discrete degree of the contrast between each neighborhood pixels and defines the adaptive thresholds of each local area by the discrete degree. Second, it encodes the extracted features by means of layer and fuses the feature information of each layer by the information entropy weight. Finally,it matches the histogram of samples using the chi-square distance. This proposed approach is applied to the Extended Yale B and CMU PIE standard face database. The experimental result shows that the method effectively improve the accuracy of the face recognition, and rarely influenced by the illumination changes and noise.

关 键 词：人脸识别 局部二值模式 局部三值模式 自适应阈值 信息熵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]