Multi-view Sparse Embedding Analysis Based Image Feature Extraction and Classification
Multi-view feature extraction is an attractive research topic in computer vision domain, since it can well reveal the inherent property of images. Most existing multi-view feature extraction methods focus on investigating the intra-view or inter-view corr
- PDF / 283,115 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 77 Downloads / 170 Views
College of Automation, Nanjing University of Posts and Telecommunications, Nanjing 210023, China zhuyangping [email protected], jingxy [email protected] 2 State Key Laboratory of Software Engineering, Wuhan University, Wuhan 430079, China 3 College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210003, China [email protected] 4 College of Communication and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China wufei [email protected] Abstract. Multi-view feature extraction is an attractive research topic in computer vision domain, since it can well reveal the inherent property of images. Most existing multi-view feature extraction methods focus on investigating the intra-view or inter-view correlation. However, they fail to consider the sparse reconstruction relationship and the discriminant correlation in multi-view data, simultaneously. In this paper, we propose a novel multi-view feature extraction approach named Multi-view Sparse Embedding Analysis (MSEA). MSEA not only explores the sparse reconstruction relationship that hides in multi-view data, but also considers discriminant correlation by maximizing the within-class correlation and simultaneously minimizing the between-class correlation from intra-view. Moreover, we add orthogonal constraints of embedding matrices to remove the redundancy among views. To tackle the linearly inseparable problem in original feature space, we further provide a kernelized extension of MSEA called KMSEA. The experimental results on two datasets demonstrate the proposed approaches outperform several state-of-the-art related methods. Keywords: Sparse embedding analysis correlation · Orthogonal constraints
1
·
Multi-view
·
Discriminant
Introduction
In computer vision domain, many applications are usually involved with different views of data. With respect to feature extraction, multi-view features can Y. Zhu and X. Jing—The work described in this paper was fully supported by the National Natural Science Foundation of China under Project No. 61272273, the Research Project of Nanjing University of Posts and Telecommunications under Project No. XJKY14016, and the Postgraduate Scientific Research and Innovation Plan of Jiangsu Province Universities under Project No. CXLX13-465. c Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 51–60, 2015. DOI: 10.1007/978-3-662-48570-5 6
52
Y. Zhu et al.
well reveal the inherent property of data. Multi-view feature extraction aims to exploit different characteristics or views of data, which is an attractive and important research direction [1,2]. Existing supervised multi-view extraction methods can be roughly categorized into two types. (1) Shared subspace learning based methods. They focus on learning a common shared subspace, in which the correlation among multiple views can be well revealed. Mostly they are based on canonical correlation analysis (CCA) [3], which is a vital multi-view extraction technique, since it can well utilize
Data Loading...
