摘要
针对人脸识别中经常遇到的"小样本"和"过学习"等问题,同时为了进一步改善人脸图像的奇异值特征在人脸识别中的识别性能,提出了一种基于奇异值分解和支持向量机的人脸识别新方法.在特征提取阶段,首先对训练样本集中的每一个人脸图像矩阵进行奇异值分解,得到训练样本的奇异值特征,然后对每个样本的奇异值特征向量进行降维、归一化、奇异值向量的分量重新排列等处理.在识别阶段,运用支持向量机作为分类工具,为了提高分类能力,选取径向基函数作为支持向量机的核函数.最后在ORL人脸数据库上验证了该方法.实验结果表明,通过对奇异值特征的相关处理,提高了识别速度和正确识别率.从而证明了所提出方法的有效性,具有一定的应用价值.
A new approach for face recognition based on singular value feature and support vector machine is presented to improve the recognition performance of singular value feature vector.At the same time,this method can be applied to solve both small sample problem and overfitting problem.Firstly,singular value decomposing is performed on every facial image of training set to get singular value features of training samples.Subsequently,several steps including dimension reduction,normalizing,and rearranging the elements order of every feature vector and so on are conducted over all the singular value feature vectors.Finally,support vector machine is used as classifier,and the RBF（radial basis of functions） function is adopted to be the kernel function to increase the classifying ability.Experiment results on ORL（olivetti research laboratory）database demonstrate that the approach proposed in this paper is efficient,and has some application values.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第6期981-985,共5页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金资助项目(60574006).
关键词
奇异值特征
支持向量机
人脸识别
singular value feature
support vector machines
face recognition
