Adaptive query shifting for content-based image retrieval

Abstract

Despite the efforts to reduce the semantic gap between user perception of similarity and feature-based representation of images, user interaction is essential to improve retrieval performances in content based image retrieval. To this end a number of relevance feedback mechanisms are currently adopted to refine image queries. They are aimed either to locally modify the feature space or to shift the query point towards more promising regions of the feature space. In this paper we discuss the extent to which query shifting may provide better performances than feature weighting. A novel query shifting mechanism is then proposed to improve retrieval performances beyond those provided by other relevance feedback mechanisms. In addition, we will show that retrieval performances may be less sensitive to the choice of a particular similarity metric when relevance feedback is performed. © Springer-Verlag Berlin Heidelberg 2001.