A Physical Approach to Color Image Understanding
Ph.D. thesis, Computer Science Department, Carnegie-Mellon University, May, 1988. Book published by AK Peters, Ltd., Wellesley MA, March 1993.
One of the goals in the field of artificial intelligence is to develop automatic systems to dynamically gather and interpret information from a complex environment. By developing devices that can "sense" as well as living organisms, researchers come that much closer to creating computers with reactions comparable to that of humans. A major subset in this study of sensory input is computer vision where image understanding involves interpreting two-dimensional objects as projections of three-dimensional scenes. Due to the information lost in the projection process, computer vision has not yet succeeded in deriving an appropriate description of an image's surface and illumination properties. "A Physical Approach to Color Image Understanding"
overcomes this problem by presenting a computer vision model that uses color information to interpret the effects of shading and highlighting in a scene. When combined with a camera model, the description of physically feasible color changes on one object can guide an algorithm in interpreting color images. The algorithm finds characteristic color clusters in the image and relates their shape to hypotheses about the object color and degree of shading, as well as the strength, position, and color of the highlights on the object. The physics-based approach of this book may lead to more reliable and useful methods of image understanding than are currently available from more traditional approaches.