A kernel-based strategy for exploratory image collection search

This paper proposes a strategy to interactively explore large collections of images. The strategy is based on kernel methods, which offer a mathematically strong framework to address each stage of an exploratory image collection system: image representation, similarity function calculation, summarization, visualization and exploration. This work also proposes a dual form of the well-known Rocchio's algorithm in order to learn from user's feedback. Experiments were performed with real users in order to verify the effectiveness and efficiency of the proposed strategy.