对角矩阵加入指数参数的AHP算法

基于AHP线性流形学习方法是通过适当的约束条件最小化目标函数来实现的,约束条件中对角矩阵的求解限制,使得公式不够灵活,于是考虑在对角矩阵求解时加入了指数参数,对公式进行泛化.通过人脸图像聚类实验,发现指数的改变对聚类结果能够产生较大影响,针对特定的人脸聚类,可以通过调整参数达到较好的聚类效果；另外,文中还对加入高斯白噪声的人脸数据库进行了实验,考察了参数对噪声的敏感度.

Approximately Harmonic Projection with the Exponential Parameter on the Diagonal Matrix

AHP(Approximately harmonic projection) is an effective linear manifold method.This method minimizes the objective function with the appropriate constraint.As the constraint is not flexible enough,this paper adds exponential parameters to regulate the result.Through experiments on face clustering,it finds that the change of the exponent can have a significant impact.For certain face clustering,it is a good choice to adjust the parameter to obtain a better cluster result.In addition,the experiments on face database with Gaussian white noise are conducted so as to examine the sensitivity of the parameter to noise.