Multi-View Collaborative Representation Classification

Abstract

With the increase popularity of multi-view data, multi-view learning has attracted vital attentions in pattern recognition as well as machine learning. Most of existing methods apply in traditional single view learning. However, these methods neglect the complementary information among the views. The aim of multi-view is to discover complementary information and enhance the single view learning result. Multi-view is capable of capture incomplete and different types of information from multiple sources. However, multi-views may contain redundant information. Many multi-view methods assume that multi-views are generated from various view-specific generation matrices. This paper proposes the multi-view collaborative representation classification (MVCRC) algorithm which contains the information of different views and the connection of view-to-view. Experimental results conducted on five practical databases are used to confirm the effectiveness of the proposed approach.