Title: Pseudo relevance feedback based on majority voting mechanism

Authors: Mawloud Mosbah; Bachir Boucheham

Addresses: Department of Informatics, Faculty of Sciences, University 20 Août 1955 of Skikda, Algeria ' Department of Informatics, Faculty of Sciences, University 20 Août 1955 of Skikda, Algeria

Abstract: The pseudo relevance feedback mechanism has come to improve the performance of the CBIR systems before visualising the final results and without any user assistance. In this paper, we show the superiority of our proposed a pseudo relevance feedback scheme 'majority voting algorithm'. The algorithm is compared to other approaches of the literature of that clustering materialised on two well known clustering algorithms namely: hierarchical agglomerative clustering method (HACM) and K-means and pseudo query reformulation materialised on pseudo query point movement, pseudo standard Rocchio formula and pseudo adaptive shifting query. Experiments are conducted on the heterogeneous Wang (COREL-1K) database and Google image engine using the colour moments as a signature. This work enables us to compare some pseudo relevance feedback techniques of the literature while the obtained results show the clear superiority of our proposed algorithm.

Keywords: content-based image retrieval; CBIR; re-ranking; pseudo relevance feedback; majority voting re-ranking algorithm; hierarchical agglomerative clustering method; HACM; K-means; precision; recall.

DOI: 10.1504/IJWS.2017.088688

International Journal of Web Science, 2017 Vol.3 No.1, pp.58 - 81

Accepted: 14 Jun 2017
Published online: 14 Dec 2017 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article