Jointly optimising relevance and diversity in image retrieval

Abstract

In this paper we present a method to jointly optimise the relevance and the diversity of the results in image retrieval. Without considering diversity, image retrieval systems often mainly find a set of very similar results, so called near duplicates, which is often not the desired behaviour. From the user perspective, the ideal result consists of documents which are not only relevant but ideally also diverse. Most approaches addressing diversity in image or information retrieval use a two-step approach where in a first step a set of potentially relevant images is determined and in a second step these images are reranked to be diverse among the first positions. In contrast to these approaches, our method addresses the problem directly and jointly optimises the diversity and the relevance of the images in the retrieval ranking using techniques inspired by dynamic programming algorithms. We quantitatively evaluate our method on the ImageCLEF 2008 photo retrieval data and obtain results which outperform the state of the art. Additionally, we perform a qualitative evaluation on a new product search task and it is observed that the diverse results are more attractive to an average user.

A Hybrid Machine-Crowd Approach to Photo Retrieval Result Diversification

“…To determine these properties, authors propose a unified framework of absorbing Markov chain random walks. A different work [8] defines a criterion to measure the diversity of results in image retrieval and propose three approaches to optimize directly this criterion. The proposed methods have been quantitatively evaluated for 39 queries on 20,000 images from the public ImageCLEF 2008 photo retrieval task which incorporates visual and textual information and also qualitatively on a novel product image data set.…”

A hybrid unsupervised image re-ranking approach with latent topic contents

“…Our work is closely related to [11] and [31]. Leuken et al [31] present three methods, called folding, maxmin and reciprocal election, to visually diversify image search results.…”

“…Therefore both these two methods run the risk of little relevant images being at top ranks. Deselaers et al [11] define a criterion to measure the diversity of results and propose a greedy selection method and a dynamic programming method to jointly optimize the diversity and the relevance.…”

“…In this paper, we extend the work in [11,31] and present a novel method to re-rank images from the initial search results. Our approach is also inspired by some notable success in latent topic models in computer vision.…”

“…The contributions of this paper are four-fold: First, we present a novel hybrid approach from the reciprocal election algorithm [31] to the greedy selection algorithm [11]. Different with [31], our approach ranks images depending on not only diversity but also relevance to the query.…”