Group sparse representation for image categorization and semantic video retrieval

Abstract

Multimedia content analysis and management are a promising and challenging theme. In this paper we develop a novel approach to image representation, which we call group sparse representation (GSR), for image classification and video retrieval. The basic idea is to represent a test image as a weighted combination of all the training images. In particular, we introduce two sets of weight coefficients, one for each training image and the other for each class. Moreover, we formulate our concern as a group nonnegative garrote model. The resulting representations are sparse, and they are appropriate for discriminant analysis. Experiments on Caltech101 and PASCAL VOC2008 image dataset and TRECVID2005 video corpus testify that our proposed approach is efficient and effective.