面向时延数据的分块PCA红外人脸识别  

Time-lapse Data Oriented Infrared Face Recognition Using Block-PCA

作　　者：谢志华[1,2] 伍世虔[2] 方志军[2] 卢宇[2] 

机构地区：[1]江西科技师范学院光电子与通信重点实验室,南昌330013 [2]江西财经大学信息管理学院,南昌330013

出　　处：《小型微型计算机系统》2012年第6期1292-1295,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60767001;F050103)资助;江西省教育厅科技项目(GJJ11225)资助

摘　　要：时延数据下的高识别性能是红外人脸识别系统应用的基础,提出一种基于分块PCA的红外人脸识别方法,提高红外人脸识别系统对时延数据识别率.为了充分利用图像的局部特征,对图像进行大小分块,并对各分块图像进行主成分分析;然后,为了缓解外部环境温度对识别的影响,通过训练学习不同温度下的红外人脸数据,去除每个图像块受环境温度影响较大的特征值.最后,为了结合整体特征,并将余下的特征值组合成特征向量用于识别.实验结果表明,本文提出的方法可以提高红外人脸识别系统对时延数据的识别率.The good performance of face recognition system under time-lapse data is the foundation of practical applications.This paper presents a time-lapse data oriented infrared face recognition method using block-PCA（Principal Component Analysis）.Firstly,to take full advantage of the local characteristics of images,the infrared face images are partitioned into several blocks and PCA are performed on those blocks.Then,in order to lessen the influence of ambient temperature on the performance of infrared face recognition,those components of PCA that are affected greatly by ambient temperature are discarded to get the robust local features based on training of infrared face images in different ambient temperatures.Finally,the remained components from all blocks are combined to represent global facial features for final recognition.Experimental results on time-lapse data illustrate that the new block-PCA can get robust local features in variable ambient temperatures,and the proposed infrared face recognition method outperforms the traditional method based on PCA.

关 键 词：时延数据 分块PCA 标准差 红外人脸识别 归一化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]