Interactive browsing of visual content on the internet

The ability to navigate among large collections of images is essential for systems that perform visual search such as content based image indexing and retrieval. This paper describes an interactive approach, improving on the commonly used approach to human-computer interaction which allows the user to repeatedly choose images similar to the search target. We allow the user to specify complex search directions specified by relations involving the locations of several template images that can be controlled by a mouse. The key idea is a novel indexing scheme, derived from the classic theory of Multi-Dimensional-Scaling (MDS). It takes into consideration the relative positioning of images in addition to their pixel values.