Color Algorithms for a General Vision System

We show that an intelligent approach to color can be used to significantly improve the capabilities of a vision system. In previous work, we adopted general physical models which describe how objects interact with light. These models are far more general than those typically used in computer vision. In this report, we use our physical models to derive powerful algorithms for extracting invariant properties of objects from images. The first algorithm is used for the generic classification of objects according to material. The second algorithm provides a solution to the color constancy problem. These algorithms have been implemented and produce correct results on real images. Some examples of experimental results are presented.