Improve the Access to Image Data by Combining Content-Based Semantic with Common Metadata Image Retrieval in a Zoomable User Interface

Abstract

We would like to present a poster and a demo of our operational semantic image retrieval prototype. The image retrieval system for which we set up a client-server environment and developed a prototype web client, accesses a database with about 600.000 images. We are working on a subset of 13.421 images which we got for test purposes from our cooperation partner. The main problems we are confronted with are the amount of data and therefore the location of an item of interest, the different search strategies that are used, usefull interaction techniques and of course the most interesting issue of combining the content-based image retrieval (cbir) and usual metadata to support the users in their tasks. We combine a zoom based interaction concept with a dynamic query metadata search and a cbir similarity search based on GIFT (Gnu Image Finding Tool) [2] to simplify search, exploration and scalable visualization of the hierarchical data space with search and filter techniques like dynamic queries and direct manipulation to solve the stated problems. We assume that the proposed concept for image retrieval can be also adapted to other data spaces (e.g. videos or documents).