基于主分量分析的最优鉴别特征抽取

THE EXTRACTION OF OPTIMAL DISCRIMINATING FEATURES BASED ON PRINCIPAL COMPONENT ANALYSIS

传统的主分量分析方法(PCA)是最为经典的图像特征抽取方法之一,由于其本质上是在最小均方差意义下给出了模式样本的最优表示,所以它通常被作为对高维图像模式进行降维的一种常用方法。但就模式分类而言,这种表示并非是最有效的。首先从统计相关性的角度揭示了PCA抽取的特征本身就具有统计不相关的良好特性。然后通过引入一种新的最大散度差类别可分性判据,从而为在PCA抽取的特征中最优鉴别特征的选取提供了一种有效策略。最后,在AR标准人脸库上的实验结果验证了算法的有效性。

The classical principal component analysis （PCA） is one of the most effective feature extraction methods. PCA essentially gives the optimal representation in a minimum of mean Square sense, so it is usually used to minify the dimension of high dimensional image pattern modes. However, the representation is not the most effective for pattern classification. Firstly, the quality property that the featUres in PCA is originally statistical uncorrelated is analyzed. Then, with a new maximum scatter-difference criterion, a new method is devoted to extract the optimal discriminating features based on principal component analysis. Finally, the experiments performed on AR face database verify the effectiveness of the proposed method.