Tulane Engineering Forum

Bill P. Buckles

Bill P. Buckles received the Ph.D. degree in Operation Research from the University of Alabama in Huntsville in 1981. Prior to joining the Tulane University, where he is Professor of Computer Science, he was on the faculty at the University of Texas in Arlington. He has been an associate editor of the IEEE Trans. on Parallel and Distributed Systems, Chair of the IEEE Computer Society Technical Committee on Distributed Processing, and recently General Chair of the IEEE Intern. Conference on Distributed Computing Systems. Twice he has been honored with technical achievement awards from NASA. Among approximately 120 papers in national or international media, he has published 27 journal articles. His interests include data mining, data warehousing, evolutionary computation, and image processing.

Presentation Topic: Image Databases

By Bill P. Buckles

Summary

The confluence of increased computer storage capacity and rapid replacement of analog imaging devices with digital ones have lead to a large number of enterprises owning vary large sets of images. NASA's Earth Science Enterprise now collects as many digital images in 90 days as it did during its first 20 years of existence. Hospitals still rely heavily on film but a shift to digital instrumentation is evident. The growth of the internet makes these image sets yet more valuable. The question is: How can images be organized so that they are conveniently accessed by doctors, scientists, farmers, and others? An image cannot be expressed as a concise vector of well-understood symbolic components such as the records in databases of the past. What representation will allow an oceanographer to find all cloud-free images containing a hydrocarbon slick? How can a radiologist fetch the most appropriate images for review prior to consultation with a patient? Both context (when, where, and how was the image obtained) and content information are necessary. Content description is the main issue. One can attach symbolic descriptors (e.g., ``convective storm,'' ``Clark Gable,'' `` cancer lesion'') or generate subsymbolic descriptors (e.g., statistical measures of color or texture). Extensibility, performance, and ease of use are factors affecting which approach to apply. Once solved, the trove of images now available will provide a treasure trove of economic and scientific benefits. Imagine doctors being able to find examples of patients having similar lesions from all over the nation. Imagine land developers being able to find projects similar to their own and from them study the changes in land use and infrastructure development. Imagine an ecologist being able to find all examples of progressive deforestation dating from, say, 1970. Imagine data mining ...