基于全局-局部保持投影的稀疏降维方法

该文提出了一种基于全局-局部结构保持的稀疏投影模型(GLSPP).通过对投影数据进行线性重构来保持数据的全局结构,从而保留投影数据的全局信息.通过约束重构系数矩阵与相似性矩阵的相似性来保持全局保持数据和局部保持投影数据的一致性.同时,对重构系数矩阵和相似性矩阵进行稀疏约束,保留主要信息,以减少冗余信息的干扰.在公开的4个人脸与物体数据集上的实验结果显示:该方法具有较高的分类准确率.

The Sparse Dimensional Reduction Based on Globality-Locality Preserving Projection

The global-local structure preserving sparse projection model(GLSPP)is proposed in this paper.The global structure of the projection data is preserved by linear reconstruction of the projection data,thus preserving the global information of the projection data.By constraining the similarity between reconstruction coefficient matrix and similarity matrix,the consistency of global preserving data and local preserving projection data is maintained.At the same time,sparse constraints to the reconstruction coefficient matrix and similarity matrix are applied to retain the main information in order to reduce the interference of redundant information.Experimental results on four face and object datasets show that the proposed algorithm has good classification accuracy.