一种具有统计不相关性的最佳鉴别矢量集  

AN OPTIMAL SET OF UNCORRELATED DISCRIMINANT FEATURES

作　　者：金忠[1] 杨静宇[1] 陆建峰[1] 

机构地区：[1]南京理工大学计算机科学系,南京210094

出　　处：《计算机学报》1999年第10期1105-1108,共4页Chinese Journal of Computers

基　　金：国家自然科学基金

摘　　要：在模式识别领域，基于Ｆｉｓｈｅｒ鉴别准则函数的Ｆｏｌｅｙ－Ｓａｍ ｍ ｏｎ 最佳鉴别矢量集技术有着重大的影响．特征抽取的一般原则是最好抽取模式的不相关的特征，而Ｆｏｌｅｙ－Ｓａｍ ｍ ｏｎ 最佳鉴别矢量集的诸鉴别特征是统计相关的．文中提出了一种具有统计不相关性的最佳鉴别矢量集，并给出了计算公式．对ＯＲＬ人脸图像数据库作了实验，实验结果表明，具有统计不相关性的最佳鉴别矢量集有较强的特征抽取能力，优于Ｆｏｌｅｙ－Ｓａｍ ｍ ｏｎBased on Fisher's discriminant function, Foley Sammon's optimal set of discriminant vectors has great influence in the area of pattern recognition. A general rule for feature extraction is to extract features as uncorrelated as possible, but the discriminant vectors in Foley Sammon's optimal set are correlated. In the paper here, an optimal set of uncorrelated discriminant features is presented, and some formula to compute it is given. Experiments with ORL face image database have been performed. Experimental results show that new optimal set of uncorrelated discriminant features has powerful ability of feature extraction, and is superior to Foley Sammon's optimal set of discriminant vectors.

关 键 词：模式识别 特征抽取 鉴别分析 人脸图像 图像识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]