Color in Computer Vision: Fundamentals and Applications

Chapter 10: Color Constancy Using Gamut-Based Methods

The gamut mapping algorithm was introduced by Forsyth [162]. ¹It is based on the assumption that in real-world images, for a given illuminant, one observes only a limited number of colors. Consequently, any variations in the colors of an image (i.e., colors that are different from the colors that can be observed under a given illuminant) are caused by a deviation in the color of the light source. This limited set of colors that can occur under a given illuminant is called the canonical gamut , and it is found in a training phase by observing as many surfaces under one known light source (called the canonical illuminant) as possible.

The flow of the gamut mapping is illustrated in Figure 10.1. In general, a gamut mapping algorithm takes as input an image taken under an unknown light source (i.e., an image of which the illuminant is to be estimated), along with the precomputed canonical gamut (see steps 1 and 2 in Fig. 10.1). The precomputed canonical gamut is obtained by aggregating all colors of the training images into one gamut. The training images are acquired under the same illuminant or corrected so that they appear to be acquired under the same illuminant. The combined set of training colors is called canonical gamut. Next, the algorithm consists of three important steps:

1. Estimate the gamut of the unknown light source ...