SVD与LDA相结合的人脸特征提取方法

提出一种新的SVD与LDA相结合的人脸特征提取方法.首先选用练训样本的均值图像作为标准图像,把训练样本投影到标准图像经奇异值分解产生的基空间中,其次提取投影系数矩阵左上角信息作为初步特征,最后再采用LDA分析方法降维提取最终的特征.该方法解决了奇异值分解用于人脸识别基空间不一致的固有缺陷,同时又增加的特征的类别信息,也避免了LDA的小样本问题.在ORL与CAS-PEAL人脸库的实验结果表明了该方法的有效性,同时对光照有一定的鲁棒性.

Method to extract features of face image combined SVD and LDA

A new method to extract features of face image based on singular value decomposition (SVD) and linear discriminant analysis (LDA) is proposed. First, the mean image of all Wain samples is selected as a standard face image, and all the train samples are projected into the two orthogonal matrixes which come form the SVD of the standard face image. Then the left-top information of projecting coefficient matrix is extracted as the primary feature. Finally, LDA is used to extracted the recognition feature. In this method, the problem of the equivalent basis space with SVD used into face recognition is resolved, at the same time, class information of samples is added and problem of small sample with LDA is abolished. ORL and CAS-PEAL database are used to test, the experimental results show the method is effective and its insensitivity to the illumination change.