Utilizing Context in Ranking Results from Distributed CBIR

Selection and ranking of relevant images from image collections remains a problem in content-based image retrieval. This problem becomes even more visible and acute when attempting to merge and rank multiple result sets retrieved from a distributed database environment. This paper presents findings from a project that investigated if combining text and image retrieval algorithms with the use of image context can help reduce the problem of merging and ranking distributed results [1]. The evaluation of our approach, implemented in a system called CAIRANK (Context-Aware Image Ranking), shows that it returns significantly better results than a more traditional ranking approach based on using DBMS-normalized image similarity scores alone.