Re-ranking from two image retrieval systems with unknown internal structures

Abstract

Image retrieval has became a huge part of computer vision and data mining. Although commercial image retrieval systems such as Google show great performances, the improvement on retrieval systems are constantly on demand because of the rapid growth of data on web space. To satisfy the demand, many re-ranking algorithms, which improve retrieved results by re-ordering, has been proposed. However, though there is potential to improve the overall performance by cooperation between multiple retrieval systems, the research on the topic has not been done sufficiently. In this thesis, we made the condition that other manner than cooperation cannot improve the ranking result. Also, we proposed a module, which is a key function of cooperation, that convert rank information from each retrieval to a form of pairwise similarity. We evaluate the algorithm on toy model and show that propose module can improve the retrieval results.