基于低秩表示的判别特征提取算法

Low-rank representation based discriminant feature extraction

为了更好地获取高维数据的特征,提出一种特征提取算法——低秩判别映射.首先基于低秩表示构造代表样本关联性的关联矩阵,然后利用关联矩阵应用判别准则.低秩表示以样本作为基函数,利用所有样本构建关联矩阵,其构造特点决定了关联矩阵能够很好地体现样本集的全局结构和样本之间的判别关系.人脸数据集的实验表明,低秩判别映射优于其他广泛应用的特征提取方法.

A novel feature extraction approach,namely low-rank representation discriminant projections,is proposed to better capture the feature information of high-dimensional data.In low-rank representation,the affinity matrix whose entries measure similarities among data points is constructed by first.Then the discriminate criterion is employed based on the input affinity matrix.Through the affinity matrix derived by low-rank representation,it can well capture the global structure of data and the discriminative property.Experimental results based on face data sets show that LRDP is superior to other widely-used feature extraction methods.