Three Interfaces for Content-Based Access to Image Collections Daniel Heesch and Stefan R¨ uger Department of Computing, Imperial College 180 Queen’s Gate, London SW7 2BZ, England {daniel.heesch,s.rueger}@imperial.ac.uk Abstract. This paper describes interfaces for a suite of three recently developed techniques to facilitate content-based access to large image and video repositories. Two of these techniques involve content-based retrieval while the third technique is centered around a new browsing structure and forms a useful complement to the traditional query-by- example paradigm. Each technique is associated with its own user in- terface and allows for a different set of user interactions. The user can move between interfaces whilst executing a particular search and thus may combine the particular strengths of the different techniques. We il- lustrate each of the techniques using topics from the TRECVID 2003 contest. 1 Introduction Being able to endow systems with the capacity to exhibit intelligent, human-like behaviour has been an early hope of computer scientists. The past fifty years have seen the gradual erosion of this hope and the consensus seems reached that the early optimism of this research program was ill-founded. The general vision problem, that is the problem of being able to describe the content of a visual scene, is among those problems that have as yet been left untouched by the otherwise relentless progress in computer science. To solve it, we will have to come up with answers to deep and fundamental questions about representation and computation that lie at the very core of human intelligence. This is what renders the problem of content-based image retrieval (CBIR) very exciting and challenging at the same time. The increasing interest in human-computer inter- action is testimony of a growing awareness that humans are currently still the most intelligent part of the system and that a tighter integration between hu- mans and machines can lead to results that would otherwise remain unattainable [8]. Unlike in typical computer vision applications, content-based image retrieval (CBIR) systems have an end user seeking information, and thus a dialogue be- tween user and machine seems more adequate from the outset. The presence of a user adds to the problem of image understanding the problem of user under- standing, a problem that can evidently only be resolved by incorporating the user in the retrieval process. P. Enser et al. (Eds.): CIVR 2004, LNCS 3115, pp. 491–499, 2004. c  Springer-Verlag Berlin Heidelberg 2004