REGIMvid at ImageCLEF2012: Improving Diversity in Personal Photo Ranking Using Fuzzy Logic

This paper handles with two main challenges: retrieving the best matching images to a given query and improving diversity in ranking using fuzzy logic. The proposed scheme proceeds as follows: First, an off line module is performed before starting the image retrieval process in order to reduce both, the execution time and the algorithm complexity. This module contains an inter-images semantic similarity graph and an inter-images visual similarity graph. Second, an on-line part implies the relevance-based ranking, the diversity-based ranking and their combination. We deal with the redundancy problem using fuzzy logic. Moreover, the vector of the relevance scores and the vector of the diversity scores are joined in order to have final scores of each image according to a given query. The experiments are conducted on ImageCLEF12 benchmark for the Personal Photo Retrieval task and show satisfying results.