一种广义的K_L变换与人脸识别.pdfVIP

23 3 Vol. 24 No. 3 2003 6 Journal of Qingdao Institute of Chemical T echnology Jun. 2003 : 00 - 4764(2003) 03- 0270- 03 K-L , 2 , 3 , , ( . , 2 0094; 2. , 22300 ; 3. , 030006) : 提出了一种直接基于图像矩阵的包含在类平均图像中判别信息的最优压缩方 法 在ORL 标准人脸库上的试验结果表明, 提出的方法不仅识别性能优于传统的主分量 分析和Fishr 线性鉴别分析, 而且特征抽取的速度得到了很大的提高 : 广义K-L 变换; 图像特征抽取; 人脸识别 : T P39 . 4　　　: A A Generalized K-L Transformation and Face Recognition , 2 , 3 - , - , - GAO Xiu mei LI Shu qin YANG ing yu ( . Department of Com puter Science, Nanjing University of Science and Technology, Nanjing 2 0094, China; 2. , , 22300 , ; Department of M athematics Huaiyin T eachers College Huaiyin China 3. , , 030006, ) Department of Com puter Science Shanxi Financial and Economical U niversity Taiyuan China Abstract: T he classical principal component analysis ( PCA ) for image feature extrac- tion is usually based on vectors, w hich make it very time-consuming, and the class in- formation in the training sample has not been utilized fully also T o overcome these tw o. , , - draw backs of PCA in this paper a generalized K L transformation is developed for im- age feature extraction. In this method, based on original image matrices directly, the discriminant information included in the class mean images is optimally compressed. Ex- perimental results on ORL face database sho