Balance between diversity and relevance for image search results

  • Authors:

  • Zhong Ji;Jing Li;Yuting Su;Yuqing He;Yanwei Pang

  • Affiliations:

  • School of Electronic Information Engineering, Tianjin University, Tianjin, China,State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China;School of Electronic Information Engineering, Tianjin University, Tianjin, China;School of Electronic Information Engineering, Tianjin University, Tianjin, China;School of Electronic Information Engineering, Tianjin University, Tianjin, China;School of Electronic Information Engineering, Tianjin University, Tianjin, China

  • Venue:
  • IScIDE'12 Proceedings of the third Sino-foreign-interchange conference on Intelligent Science and Intelligent Data Engineering
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Image search reranking has received great attention since it overcomes the drawback of "only textual features utilization" in nowadays web-scale image search engines. Most of existing methods focus on relevance reranking, that is reordering the returned results according to their relevance with the query. However, in many cases, users cannot precisely and exhaustively describe their requirements by several query words. Therefore, relevant results with more diversity are more easily meet the users' ambiguous purpose. To address this problem, in this paper, we proposed a DIR (DDrank-based Image search Rerank) algorithm, which can enrich the topic coverage while keeping the relevance with minimal impact. DIR is based on vertex reinforced random walk and further curbing neighbor items' growing rate. Therefore, DIR can automatically balance the relevance and diversity of the top ranked vertices in a principle way. Extensive experiments are performed in a popular image dataset, and the results demonstrate the superiority against other existing methods in the criterion of both AP and ADP.