基于离散度准则的人脸局部特征权值学习算法

提出了一种利用离散度准则计算人脸局部特征权值的方法.在利用局部特征进行人脸识别的算法中,可以依据各个局部特征对人脸识别贡献的大小,对每一个局部特征分配不同的权值,以此来提高识别效果.从模式识别的角度来看,易于分类的局部特征对人脸识别有较大的贡献,反之亦然.在特征空间中,当类内模式较密集,不同类模式相距较远时,模式特征易于分类.离散度矩阵的迹描述了模式特征的分散程度,通过运用统计学习的方法,以类内离散度矩阵之逆和类间离散度矩阵的乘积的迹作为衡量局部特征分类性能的依据,根据局部特征的分类性能来确定其权值.试验结果表明该算法可行,采用离散度准则计算的局部特征权值能够明显提高人脸识别率.与同类算法相比,该方法具有计算简便、易于实现等优点.

Weights learning algorithm of facial local feature based on scatter criterion

This paper proposes a method for computing local facial feature weights based on scatter criterion.In the face recognition algorithm by using local features,different weights can be assigned to different local features according to their contributions to face recognition in order to increase the recognition performance.From the point of pattern recognition,those local features which can be easily classified contribute more to face recognition and vice versa.In the feature space,when the distributions of with-in classes are close and the distances between different classes are far,pattern features can be classified easily.The trace of scatter matrix describes the scatter degree of pattern features.By the means of statistical learning,the trace of the product of the inversion matrix of the within-class scatter matrix and between-class scatter matrix is used as a criterion for weighing the discriminative capability of a local feature,and the weight of local features are determined based on their discriminative capability.Experimental results show that the weights which are computed by scatter criterion can increase the recognition performance significantly,which confirm the feasibility of our algorithm.Compared with similar algorithms,this algorithm has the property of computing simply and being realized easily.