Interactive exploration of visual content

t Human-compute rr interaction is a decisive factor in effective content-base d acces to largee image repositories . In current image retrieval systems the user refines his query byy selecting exampl e images from a relevance ranking. Since the top ranked images aree all similar, user feedback often results in rearrangemen t of the presente d images only. . Forr better incorporation of user interaction in the retrieval process , we have developedd the Filter Image Browsing method. It also uses feedback through image selection.. However, it is based on differences between images rather than similarities. Filterr Image Browsing present s overviews of relevant parts of the databas e to users. Throughh interaction users then zoom in on parts of the image collection. By repeat edlyy limiting the information space, the user quickly ends up with a smal number of relevantt images. The method can easil y be extended for the retrieval of multimedia objects. . Forr evaluation of the Filter Image Browsing retrieval concept , a user simulation iss applied to a pictorial databas e containing 10,000 images acquired from the World Widee Web by a search robot. The simulation incorporates uncertaint y in the definition off the information need by users. Results show Filter Image Browsing outperforms plainn interactive similarity ranking in required effort from the user. Also, the method producess predictabl e results for retrieval sessions , so that the user quickly knows if a successfu ll session is possibl e at all. Furthermore, the simulations show the overview techniquess are suited for applications such as hand-hel d devices where screen space is limited. .