Large online collections of images are attractive in principal, but in in practice they are difficult to use. The comon solution of a search interface is only partly satisfactory because it is hard to compose a good query if you are not already familiar with the contents. Ideally one would be able to navigate a large image collection in a natural way. As a step towards this goal we have implemented an image database browser based on our work on clustering images using features and text
For demonstration we provide a browser for 10,000 images of art courtesy of the Fine Arts Museum of San Francisco.
As far as it is known, the demo currently works on all windows systems with IE. On linux, Netscape can hang, depending on the exact configuration. Mozilla (version 1.0) seems to work fine in conjunction with Red Hat 7.2. Mac users need to use Internet Explorer.
The browser uses the Berkeley digital library GIS viewer to explore the space of clusters. People contributing to this effort include Kobus Barnard, Pinar Duygulu, and David Forsyth.
Note: The demo was constructed while Kobus Barnard, and Pinar Duygulu, were at the University of California, Berkeley, working with the Digital Library Project. Kobus Barnard is now with the University of Arizona computer science department. The project lives on at both institutions (stay tuned for version two and other demos).
Sneak preview of sub alpha version of the rebuild which will become version two (interface is still under construction, image layouts are random for debugging purposes, etc.)