Leveraging Click Completion for Graph-based Image Ranking

Image ranking is a critical component in the image search systems, and graph-based ranking has become a promising way to enhance the retrieval effectiveness. Leveraging the click through data to facilitate the ranking is one of the current trends. However, the sparse and noisy properties of the click-through data make the exploitation of such resource difficult. To this end, this paper proposes a click completion solution for graph based image ranking, which consists of two coupled components. The first one is a click completion algorithm to handle the sparseness. Another one is a soft-label graph ranking solution to exploit the completed click-through data noise-tolerantly. We conduct extensive experiments to evaluate the performance of the proposed scheme for image retrieval, in which encouraging results validate the effectiveness of the proposed techniques.