Independent feature analysis for image retrieval

Content-based image retrieval methods based on the Euclidean metric expect the feature space to be isotropic. They suffer from unequal differential re levance of features in computing the similarity between images in the input feature space. We propose a learning method that attempts to overcome this limitation by capturing local differential relevance of features based on user feedback. This feedback, in the form of accept or reject examples gene rated in response to a query image, is used to locally estimate the strengt h of features along each dimension while taking into consideration the corr elation between features. This results in local neighborhoods that are cons tricted along feature dimensions and that are most relevant, while elongate d along less relevant ones. In addition to exploring and exploiting local p rincipal information, the system seeks a global space for efficient indepen dent feature analysis by combining such local information. We provide exper imental results that demonstrate the efficacy of our technique using both s imulated and real-world data. (C) 2001 Elsevier Science B.V. All rights res erved.