Structured
Optimal Graph Feature Selection
对图像样本进行特征选择，使用自学习算法，然后得出相似度矩阵S(% Input
% X: dim*num data matrix
% gamma: coefficient of L21
% d: projection dim of W(dim*d)
% c: number of clusters
% k: nearest neighobrs

%Output
%id: sorted features by ||w_i||_2)