Employing negative examples in region-based image retrieval

In this paper, the integration of negative examples into an efficient region-based image retrieval framework is investigated. The outputs of a support vector machine that separates positive images from negative ones are used to refine the current retrieval result. Considering the speciality of region-based representation, a kind of new kernel is proposed, which is the generalization of Gaussian kernel. Experimental results on a database of 10,000 general-purpose images demonstrate the effectiveness of employing negative examples and the superiority of the new kernel to Gaussian kernel.