The need to gather information about terrorism has highlighted areas of application for digital library technology. Among key issues are the ability to do image analysis, whether face scanning at airports or analysis of satellite surveillance photographs, and the need for search and translation from languages not previously given much study. But perhaps one of the most difficult problems is the ability to merge information from many sources and in many media.

Spoken language processing is a familiar area, but more attention has been given to statistical techniques in recent years, and we now see an interest in handling languages where we do not have large amounts of bilingual text. Fortunately, statistical methods are improving and we now see projects (at CMU and JHU, for example) on transference of information between languages and on statistically based translation.

Image retrieval is an area of great activity right now, including 3-D graphics software and video as well as static 2-D pictures. The basic paradigm for image retrieval is to crawl around the image with some low-level feature extractor, and then use the numbers resulting as classifiers for retrieval. The simplest feature to extract is the color histogram, which is why Robert Wilensky always says to be suspicious of any image retrieval demo which concentrates on finding pictures of sunsets. Although we have moved on from that, we are still well short of something a photo or film librarian would use. About half the queries asked of a film librarian require knowing the names of things in a picture - not just "tower" or "river" but "Eiffel Tower" or "River Seine". Thus it may be as important to pursue projects based on recognizing existing images from a dictionary (e.g. faces) as to do basic feature extraction and classification.

Some interesting projects are the shape recognition and image labeling work at Berkeley (Malik and Forsyth), scale and viewpoint invariance search at Oxford (Zisserman), and face recognition at CMU. Interesting applications include analysis of aerial photographs, analyzing images of geological structures, and (recently) a stress on individual tracking by looking through pictures. The most recent research frontier is in 3-D models and searching, ranging from historic building reconstruction to drug design.

The need to integrate large amounts of information from multiple sources, and to present information in more intelligible ways, are also challenging our researchers. But again, progress is being made: there is now a 100 terabyte public text data base (the Internet Archive) and we are learning more techniques for summarization and filtering. We still need better ways to display information and to exploit it in practice.