The PicToSeek WWW image search system

We give an overview of the PicToSeek system for exploring visual information on the World Wide Web. PicToSeek automatically collects, indexes and catalogs visual information entirely on the basis of the pictorial content. PicToSeek allows for content-based image retrieval conducted in an interactive, iterative manner guided by the user by relevance feedback. Relevance feedback can be seen as a method of feature selection and weighting. The PicToSeek system has been implemented based on the client-server paradigm. The client is a Java applet and takes care of interactive query formulation, the display of the results, and the relevance feedback specification given by the user. The server is a Servlet using C-libraries and takes care of the image feature extraction, feature weighting from relevance feedback, k-nearest neighbour feature classification, and image sorting. The system is available at http://www.wins.uva.nl/research/isis/zomax/.