Affiliations: [a] School of Digital Media, Jiangnan University, Wuxi, Jiangsu, P.R. China
| [b] School of Medical Informatics, Nantong University, Nantong, Jiangsu, P.R. China
Correspondence:
[*]
Corresponding author. Zhibin Jiang, School of Digital Media, Jiangnan University, Wuxi, Jiangsu, P.R. China.
Tel.: +86 133 45158572; Fax: +86 510 85913570; E-mail: [email protected].
Abstract: Since the density distribution of each view, which can be often built only from the corresponding partial data observed along each view from the whole face data, ignores the coherent information between all the views, multi-view face recognition sometimes is seriously troubled by an unavoidable phenomenon that the dissimilarity between the samples from the same class but different views is greater than that between the samples from the different classes of same view. In this study, by considering a common hidden space cross all the views, consistent hidden density distribution between views in the common hidden space is delved so as to address this issue. Accordingly, a novel multi-view support vector machine based on consistent hidden density distributions between views in common hidden space (2V-SVM-CHDD) is proposed for an efficient multi-view face recognition, and its theoretical convergence is also analyzed. Extensive experimental results on real face image datasets indicate the effectiveness of the proposed multi-view method.
Keywords: Multi-view face recognition, unavoidable phenomenon, common hidden space, multi-view support vector machine
