Probabilistic feature relevance learning for content-based image retrieval

Most of the current image retrieval systems use "one-shot" queries to a dat abase to retrieve similar images. Typically a K-nearest neighbor kind of al gorithm is used, where weights measuring feature importance along each inpu t dimension remain fixed (or manually tweaked by the user), in the computat ion of a given similarity metric. However, the similarity does not vary wit h equal strength or in the same proportion in all directions in the feature space emanating hom the query image. The manual adjustment of these weight s is time consuming and exhausting. Moreover, it requires a very sophistica ted user. In this paper, we present a novel probabilistic method that enabl es image retrieval procedures to automatically capture feature relevance ba sed on user's feedback and that is highly adaptive to query locations. Expe rimental results are presented that demonstrate the efficacy of our techniq ue using both simulated and real-world data. (C) 1999 Academic Press.