基于多尺度自适应LDA的人脸识别方法

在人脸提取特征时,线性判别分析(LDA)方法受到光照、姿态等因素引起的高频部分影响较大,忽视了可能含有重要鉴别能力的低频信息.同时,人脸识别属于小样本问题,会使类内散布矩阵发生严重退化.针对以上两个问题,提出了一种基于多尺度自适应线性判别分析(MA-LDA)的人脸识别方法,并在ORL和Yale人脸库中进行了验证.MATLAB编程实验结果表明,该方法比传统方法有更好的性能.

Face recognition method based on multi-scale adaptive LDA

The facial feature extraction based on linear discriminant analysis(LDA) method is deeply influenced by high frequency information,such as lighting condition,gesture.So that some low frequency characters of an image are lost,but these characters may contain important discriminant information.Also,human face recognition is small sample size problem,it will degrade within-class scatter matrix greatly.To solve the two problems,a method based on multi-scale adaptive linear discriminant analysis(MA-LDA) is proposed,and some experiments are done based on ORL and Yale human face image database.The experimental results performed on Matlab show this method achieves better performance than traditional methods.