一种新颖的分组决策方法及其应用

A Novel Method of Grouped Decision and Its Application

该文分析了广义最佳鉴别向量集，给出了基于用于线性特征抽取的广义最佳鉴别向量的分组决策方法。将所有的样本分成若干组，从理论上说明每一组的Ｆｉｓｈｅｒ 鉴别函数值大于整体的Ｆｉｓｈｅｒ 鉴别函数值，因此，每一组的识别正确率远高于整体的识别正确率。为了验证所述方法的有效性，将其用于人脸识别。实验结果显示：当采用同样个数的广义最佳鉴别向量时，此方法比不分组的方法能得到更高的识别正确率；如果采用分类决策，可用较少的广义最佳鉴别向量得到良好的识别正确率，而用其它方法要达到同样的正确率，需要许多广义最佳鉴别向量。

This paper analyses the generalized optimal set of discriminant vectors,presents a grouped decision method based on generalized optimal set of discriminant vectors for linear feature extraction. We divide the whole sample to several groups, and illustrate that the every group's value of the Fisher discriminant function is higher than that of whole sample theoretically, so, every group's recognition correct rate is much higher than whole correct rate. In order to test the efficiency of our method, we apply it to the human facial recognition. Experimental results have shown:by using the same number of the generalized optimal discriminant vectors, our method can get higher recognition correct rate than that not grouping.If we use the grouped decision method , we can reach good recognition correct rate with very few generalized optimal discriminant vectors, yet the existing methods must use many generalized optimal discriminant vectors to reach.