Image Re-ranking and Rank Aggregation Based on Similarity of Ranked Lists

Combining Re-Ranking and Rank Aggregation Methods

“…• RL-Sim Re-Ranking Algorithm: The RL-Sim Re-Ranking [18] algorithm is based on the conjecture that contextual information encoded in the similarity between ranked lists can provide resources for improving effectiveness of CBIR descriptors. An iterative approach is used for improving the effectiveness of ranked lists.…”

“…, A m }. The approach used by the RL-Sim Algorithm [18] combines the set A in a unique matrix A c . For the matrices combination a multiplicative approach is used.…”

“…We call this method as Set Rank Aggregation (SetRA) . We consider the strategy of modelling the ranked lists as sets of different sizes, also used by the RL-Sim [18] algorithm. We use the function ψ to compute the similarity between ranked lists.…”

“…The evaluation of the Cascade Re-Ranking approach considers the MPEG-7 dataset and four different re-ranking methods: the Distance Optimization Algorithm (DOA) [16], the Mutual kNN Graph [9], the Contextual Re-Ranking [15], and RL-Sim [18] algorithms. We also considered the Contextual Re-Ranking [15] and RL-Sim [18] combined with all algorithms. Table 1 presents the results for Recall@40 measure.…”

“…In general, CBIR systems consider only pairwise image analysis, that is, compute similarity measures considering only pairs of images, ignoring the rich information encoded in the relations among several images. In the past few years, there has been considerable research on exploiting contextual information for improving the distance measures and re-ranking images in CBIR systems [9,15,16,18,25,26]. Another approach for improving CBIR systems is based on using rank aggregation techniques [2,5].…”