Features Fusion for Diversity Gap Reduction

Diversity has been promoted in image retrieval results using clustering algorithms to tackle queries, which refer to multiple information needs, e.g., due to ambiguity. Despite the effective results of diversity-aware methods, the image wealth of large collections and the subjectivity of human perception bring the semantic gap problem. This paper presents multimodal fusion approaches aimed at reducing the diversity gap with ensemble clustering and dimensionality reduction. The applied methods were evaluated by quantifying the clustering effectiveness in comparison to human decisions. The experimental results demonstrate the potential of these approaches to boost diversity-oriented engines and that they could improve state-of-the-art systems.