一种新的基于Fisher准则的线性特征提取方法

A New Linear Feature Extraction Method Based on Fisher Criterion

针对现有的基于Fisher准则的线性特征提取方法存在的不足,提出了一种新的加权Fisher特征提取方法。该方法通过引入一个加权函数来削弱边缘类别的影响,减少投影空间中相邻类别间的重叠,提高了识别正确率。针对小样本问题,也给出了该算法的一个可行的最优判别矢量集的求解方法。分别对COIL图像数据库以及ORL人脸数据库进行实验,结果表明,就识别率而言,该方法得到的最优判别矢量具有更好的特征提取能力。

A novel weighted Fisher discriminant analysis is developed in this paper. The approach aims at overcoming the drawback of the previous FDA- based methods by introducing a weighting function to weaken the influence of outlier classes and reduce the large overlapping of neighboring classes in projection space. And a feasible solution to this new approach is suggested. Experimental results on the ODIL and ORL database show that new optimal discriminant vectors have more powerful ability of feature extraction in terms of rates of classification.