Content Based Image Retrieval with Relevance Feedback Using Riemannian Manifolds

Abstract

In this paper we propose a novel approach for content-based image retrieval with relevance feedback, which is based on Riemannian Manifold learning algorithm. This method uses positive and negative (relevant/irrelevant) images labeled by the user at every feedback iteration. In this paper, we pre-computed the cost adjacency matrix and its eigenvectors corresponding to the smallest eigen values for effectiveness and efficiency of the retrieval system. Then we apply the Riemannian Manifolds learning concept to estimate the boundary between positive and negative images. Experimental results of the proposed method have been compared with earlier approaches, which show the superiority of the proposed method.