基于奇异值分解的几种人脸识别方法研究

Face Recognition Approaches Based on Singular Value Decomposition

讨论了基于奇异值分解的人脸识别问题。基于传统奇异值向量进行人脸识别时,对于不同图像而言,由于提取奇异值向量所在的基图像空间不一致,导致人脸识别率低。奇异值向量包含的人脸图像的有效信息少也不足以进行有效的人脸识别。针对传统奇异值向量算法的缺点,提出了基于奇异值分解的基图像空间下的几种人脸识别方法。这些方法基于统一的基图像空间以及直接利用部分基图像进行人脸识别,从而克服了传统算法的缺点。利用ORL人脸数据库进行实验,采用最近邻决策规则来进行分类识别。实验结果显示,提出的方法比基于奇异值人脸识别方法具有优越性。

The problem of face recognition based on singular value decomposition is discussed.When doing face recognition based on the traditional singular value vector,the reason for the low rate of the recognition is found if the face imagesare is different,the spaces of basic image for singular value vector are inconsistent.The feature of singular value vector can not contain enough face information,so the method of singular value vector is not effective enough.As for the drawbacks of the traditional singular value vector,some approaches are proposed based on the space of basic image and singular value decomposition.Some approaches for face recognition are based on the united space of basic image,the other directly use some space of basic image.The disadvantage of the traditional algorithm.Experiments on the ORL face database is done,the nearest neighbor decision rule to recognize human face is used.The results of experiments show that these approaches are superior to the method of singular value vector.