Social image retrieval based on topic diversity

Abstract Image search re-ranking is one of the most important approaches to enhance text-based image results. Extensive efforts have been dedicated improve accuracy and diversity tag-based retrieval. However, how make top-ranked results relevant diverse still a challenging problem. In this paper, we propose novel method diversify retrieval by latent topic analysis. We first employ NMF (Non-negative Matrix Factorization) Lee Seung (Nature 401(6755):788–791, 1999) estimate initial relevance score query q . Then, fed into an adaptive multi-feature fusion model learn final score. Next, diversification process conducted. group all images semantic clustering distribution each cluster The clusters are ranked based on vector list obtained greedy selection mechanism estimated relevances. Experimental NUS-Wide dataset show effectiveness proposed approach.