复用特征组合的单幅人脸图像识别  

Face recognition with one training sample by repeatedly used face features

作　　者：廖红文[1] 冯国灿[1] 

机构地区：[1]中山大学数学与计算科学学院

出　　处：《计算机应用》2005年第8期1777-1779,共3页journal of Computer Applications

基　　金：教育部重点基金资助项目(104145);广东省自然科学基金资助项目(031609)

摘　　要：单幅图像的人脸识别问题目前研究较少,而许多识别算法一旦应用到单幅训练图像的人脸库时,识别率会急剧下降。通过研究人脸的各个局部特征对识别人脸的影响,筛选出几个最能表达人脸信息的局部特征,然后利用Boosting思想,为从单个图像样本中挖掘更多的信息,重复使用人脸特征,将人脸的整体特征和局部特征结合起来构造了两个人脸识别系统———多特征投票法和复用特征法。Face recognition for one training sample is one of the most challenge tasks. Many developed systems, that work very well when there are sufficient representative training samples, often become less effective or low accuracy for one training sample. After analysing the local facial features, several significant features to represent the human face were selected. Then a novel algorithm of feature extraction for single sample by exploiting repeatedly the selected face features was proposed following the Boosting methods. Finally, two face recognition systems were built by combining whole face features with local face features—— voting-based method with multi-features and mothod with repeatedly used features. The experiments show that the developed systems have good performance in comparison with the existing system.

关 键 词：Boosting思想 复用特征 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]