Dynamic overview techniques for image retrieval

One difficulty often overlooked in information retrieval systems is that search criteria themselves are often poorly defined. People describe their information needs in many different ways and frequently change their goals depending on the current results of their search. We have investigated the hypothesis that overviews of the space of available solutions are a good way to remedy this situation. Our overview techniques allow users to get a feel for the meaning of categories through randomly chosen examples, find similar images using content search, and to inspect the global distribution of images according to certain criteria. Users thus organize the retrieval task into an iterative browsing process that makes them specify their queries more accurately. As a result they are more satisfied with what the system retrieves.