Multi-view dimensionality reduction via similarity or reconstruction
Xiaoyu Zhang, Xiumei Wang, Peitao Cheng · 2016
Dimensionality reduction methods for multiple view data have been widely used in the field of computer vision and multimedia research. Since different views could have very different statistical properties, how to learn a low-dimensional representation from multiple views is a challenging problem. In this paper, two unsupervised dimensionality reduction methods for multiple view data are proposed. They take advantage of the properties of Markov chain to effectively handle potential noise in transition probability matrix by low rank and sparse decomposition. This can significantly improve the performance of the dimensionality reduction methods. In particular, the proposed methods have only one parameter needed to be adjusted and the global optimal solutions can be obtained directly by generalized eigenvalue decomposition, which illustrate the simplicity of the algorithms. The experimental results show that the proposed methods can achieve high precision in classification tasks.