Feature relevance learning with query shifting for content-based image retrieval

Multimedia Database Retrieval: A Human-Centered Approach

“…Using this paradigm, different weighting schemes and basis functions have also been explored to improve the functionality of the metric adaptation [68][69][70]. More recently, the combination of a query reformulation model and an adaptation metric has shown effective learning with respect to the retrieval accuracy and convergence speed of the learning system [14,[71][72][73][74]. The fundamental concepts derived from this framework have become significant for many newly-proposed retrieval systems that are directed to a more complex level.…”

“…In [14,[71][72][73][74], an adaptive system combines a query reformulation model with the similarity function (e.g. Eqs.…”

<title>Feature-adaptive relevance feedback (FA-RF) for content-based image retrieval</title>

“…Query shifting techniques aim at moving the query towards the region of the features space that contains relevant images and far from the region of non-relevant images 2,4,12 . Feature relevance weighting uses the relevance feedback information to update the weights associated with each feature in order to model user's need and to emphasize the most important characteristics perceived by a user 3,4,12 . Some systems incorporated both techniques 4, 12 . Image description is probably the most important aspect for the retrieval performance, and usually includes the highest possible number of significant parameters.…”

“…Originally developed for Information Retrieval (IR) 9 , it is an active supervised learning technique, which uses the positive and the negative examples provided by the user to improve the retrieval performance. Two main strategies (techniques) for exploitation of the relevance information have been proposed in the literature of CBIR: query shifting and feature relevance weighting 2,3,4,12,13 . Query shifting techniques aim at moving the query towards the region of the features space that contains relevant images and far from the region of non-relevant images 2,4,12 .…”

“…Two main strategies (techniques) for exploitation of the relevance information have been proposed in the literature of CBIR: query shifting and feature relevance weighting 2,3,4,12,13 . Query shifting techniques aim at moving the query towards the region of the features space that contains relevant images and far from the region of non-relevant images 2,4,12 . Feature relevance weighting uses the relevance feedback information to update the weights associated with each feature in order to model user's need and to emphasize the most important characteristics perceived by a user 3,4,12 .…”