Multi-View Kernel Dsiscriminant Correlation And Orthogonal Analysis For Image Classification

Recently, with the constant development of data-acquisition techniques. Multi-view learning has received extensive attention in the field of computer vision. On the one hand, it enables people to reveal the inherent property of pattern better by using multi-view date. On the other hand, it also brings new challenges to those traditional methods or algorithms which designed for single view data. How to fully exploit associated information and independent information of multi-view date has become one of the hot spots in the field of machine learning.Firstly, motivated by feature fusion, this paper adds the orthogonal constraints of transformation matrix on the foundation of Multi-view Discriminant Analysis(MvDA), and thus Multi-view Discriminant Orthogonal Analysis(MvDOA) is proposed. On the one hand, it preserves the reconstruction relationship in data. On the other hand, it can achieve the goal of reducing redundant information. The classification ability of the algorithm is further enhanced.Secondly, this paper proposes a novel approach named Multi-view Discriminant Correlation and Orthogonal Analysis(MvDCOA) on the foundation of MvDOA. This algorithm uses Mv DOA to obtain the independent information of each view, and then uses the Multi-view Canonical Correlation Analysis(MDCCA) to get the associated information of the multi-views. Finally, the associated features and independent features of multiple views are connected in series to make full use of the information contained in multiple views. The classification ability of the algorithm is further enhanced.Finally, in order to address the inseparable problem in samples, we extend it to the nonlinear space, and propose the Multi-view Kernel Discriminant Correlation and Orthogonal Analysis(MvKDCOA). MvKDCOA can explore the nonlinear structure of multi-view data to further enhance the performance of MvDCOA. The classification ability of the algorithm is further enhanced.The experimental results on the MFD, AR, Multi-PIE and PolyU databases demonstrate that the proposed approaches can effectively solve the multi-view classification problem. Compared with the popular multi-view classification methods, it improves the recognition performance.