Global structure-guided neighborhood preserving embedding for dimensionality reduction

Abstract

Graph embedding is one of the most efficient dimensionality reduction methods in machine learning and pattern recognition. Many local or global graph embedding methods have been proposed and impressive results have been achieved. However, little attention has been paid to the methods that integrate both local and global structural information without constructing complex graphs. In this paper, we propose a simple and effective global structure guided neighborhood preserving embedding method for dimensionality reduction called GSGNPE. Specifically, instead of constructing global graph, principal component analysis (PCA) projection matrix is first introduced to extract the global structural information of the original data, and then the induced global information is integrated with local neighborhood preserving structure to generate a discriminant projection. Moreover, the L2 , 1-norm regularization is employed in our method to enhance the robustness to occlusion. Finally, we propose an iterative optimization algorithm to solve the proposed problem, and its convergence is also theoretically analyzed. Extensive experiments on four face and six non-face benchmark data sets demonstrate the competitive performance of our proposed method in comparison with the state-of-the-art methods.