Color in Computer Vision: Fundamentals and Applications

Chapter 9: Color Constancy Using Low-level Features

The first type of illuminant estimation algorithms discussed in this book are static methods, or methods that are applied to input images with a fixed parameter setting. Two subtypes are distinguished: a) methods that are based on low-level statistics and b) methods that are based on the physics-based dichromatic reflection model.

9.1 General Gray-World

The best-known and most often used assumption of this type is the gray-world assumption [137]: the average reflectance in a scene under a neutral light source is achromatic. In the original work, the hypothesis is used to derive that the average reflectance for short-wave, middle-wave and long-wave regions is equal, but a stronger definition of achromatic reflectance of a scene is often employed ([139, 138]):

9.1

which avoids making further assumptions. ¹The constant k is between 0 for no reflectance (black) and 1 for total reflectance (white) of the incident light, and the integral is over the domain of the scene. For such a scene with achromatic reflectance, it holds that the reflected color is equal to the color of the light source, since

9.2

9.3

9.4

where the theorem of Fubini ...