Combining Re-Ranking and Rank Aggregation Methods

Abstract

Content-Based Image Retrieval (CBIR) aims at retrieving the most similar images in a collection by taking into account image visual properties. In this scenario, accurately ranking collection images is of great relevance. Aiming at improving the effectiveness of CBIR systems, re-ranking and rank aggregation algorithms have been proposed. However, different re-ranking and rank aggregation approaches produce different image rankings. These rankings are complementary and, therefore, can be further combined aiming at obtaining more effective results. This paper presents novel approaches for combining re-ranking and rank aggregation methods aiming at improving the effectiveness of CBIR systems. Several experiments were conducted involving shape, color, and texture descriptors. Experimental results demonstrate that our approaches can improve the effectiveness of CBIR systems.

Combining re-ranking and rank aggregation methods for image retrieval

“…Experimental results demonstrate that our combination approaches can further improve the effectiveness of image retrieval systems. This paper differs from previous work [38,39] as it presents: (i) a formal definition of the proposed methods; (ii) a new meta-agglomerative approach for combining rank aggregation methods; and (iii) extends the experimental protocol, evaluating the proposed method on visual and mutlimodal image retrieval tasks.…”

A soft computing approach for learning to aggregate rankings

“…Given the large amount of existing techniques, each with different properties, recent work (e.g., [43] and [44]) has started to explore the idea of combining the outputs of these methods. For instance, in [44] the authors describe a model to combine in an agglomerative way results of different rank aggregation techniques and show that this model can obtain better results than the individual methods.…”

“…Our proposed model (hereafter called GP-Agg for GP-Based Rank Aggregation) uses genetic programming to search the best combinations of different rank aggregation techniques in an agglomerative way, that is, using the result of one as input of another. The principal reasons to use genetic programming for this task are: (i) the inherent complementarity between the results of different rank aggregation techniques studied in [43]; (ii) the large size of the search space for combination functions; and (iii) previous success of using GP in information retrieval.…”

“…• Experiment including re-ranking algorithms (e.g., [43]) in the structure of an individual as unary operators.…”