Journal of South China University of Technology (Natural Science Edition) ›› 2011, Vol. 39 ›› Issue (7): 156-162.doi: 10.3969/j.issn.1000-565X.2011.07.026

• Computer Science & Technology • Previous Articles     Next Articles

Image Retrieval Based on Spectral Clustering and Multiple Instance Learning

Li Zhan1  Peng Jin-ye1,2  Wen Chao1   

  1. 1. School of Information and Technology,Northwest University,Xi’an 710069,Shaanxi,China; 2. School of Electronics and Information,Northwestern Polytechnical University,Xi’an 710072,Shaanxi,China
  • Received:2010-08-18 Revised:2011-01-11 Online:2011-07-25 Published:2011-06-03
  • Contact: 李展(1973-) ，男，博士，讲师，从事图像检索、机器学习研究． E-mail:lizhan@nwu.edu.cn
  • About author:李展(1973-) ，男，博士，讲师，从事图像检索、机器学习研究．
  • Supported by:

    教育部新世纪优秀人才支持计划项目( NCET-07-0693) ; 陕西省教育厅科学研究计划项目( 10JK852)

PDF

1730

Abstract:

Proposed in this paper is a novel algorithm based on the spectral clustering and the multiple instance learning,which is applied to the object-based image retrieval. In this algorithm,first,the whole image is regarded as a bag and the visual features of the segmented region are regarded as instances. Next,the spectral clustering of the instance set of positive bags is performed according to the principle of maximum the clustering center number to select the potential center and the representation of positive instance. Then,the radial basis function ( RBF) and the pyramid match kernel ( PMK) are respectively used to measure the similarities of the potential positive instances and other instances in bags. Finally,the support vector machine and the relevance feedback are adopted to retrieve images. Experimental results of the SIVAL image set show that the proposed algorithm is an effective solution to the object-based image retrieval.

Key words: image retrieval, multiple instance learning, spectral clustering, pyramid match kernel