Open AccessAdaptive learning region importance for region‐based image retrieval
Author(s) -
Yang Xiaohui,
Lv Feiya,
Cai Lijun,
Li Dengfeng
Publication year - 2015
Publication title -
iet computer vision
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.38
H-Index - 37
eISSN - 1751-9640
pISSN - 1751-9632
DOI - 10.1049/iet-cvi.2014.0119
Subject(s) - computer science , image retrieval , representation (politics) , artificial intelligence , image (mathematics) , relevance feedback , visual word , interference (communication) , adaptive learning , pattern recognition (psychology) , machine learning , computer network , channel (broadcasting) , politics , political science , law
This study addresses the issue of region representation in region‐based image retrieval (RBIR). In order to reduce the user's burden of selecting the region of interest, a statistical index called visual region importance (RI) is constructed to describe the region. By learning from user's current and historical feedback information, visual RI can be automatically updated and semantic RI can be obtained. Furthermore, adaptive learning RI and memory learning RI (MLRI) techniques for RBIR system have been presented. Specifically, the MLRI can mitigate the negative influence of interference regions well. Extensive experiments on the Corel‐1000 dataset and the Caltech‐256 dataset demonstrate that the proposed frameworks are effective, are robust and achieve significantly better performance than the other existing methods.