Retrieval with knowledge-driven kernel design: an approach to improving SVM based CBIR with relevance feedback,” in 10th international conference on computer vision

The performance of SVM-based image retrieval is often constrained by the scarcity of training samples. The total number of image samples labelled by users in a retrieval session is very limited, and these small number of labelled samples cannot effectively represent the true distributions of positive and negative image classes, especially for the neg-ative image class. This paper proposes a novel approach to deal with this problem. Instead of treating it as a problem, the mere existence of the small number of labelled images and their desired distribution in the kernel space is con-sidered as prior knowledge from image retrieval to aid the design of the kernel used by SVMs. This is achieved by max-imizing a criterion, such as one based on scatter matrices, through gradient-based search methods, incurring very lit-tle computational overhead to real-time retrieval process. Experimental results on two benchmark image databases demonstrate the improved retrieval performance by the dy-namically designed kernel and hence the effectiveness of the proposed approach for SVM based image retrieval. 1