Dichromatic line

When we now compute chromaticities, the three-dimensional data points are projected onto the plane r + g + b = 1. The linear combination results in a two-dimensional line in chromaticity space. Let Cm be the chromaticity of the matte color and let cs be the chromaticity of the specular color. Then, the dichromatic line on the plane r + g + b = 1... 

Figure 6.31 Two different surfaces have a different diffuse component however, the specular component is the same. The color of the illuminant is located at the intersection of the dichromatic lines.

Figure 6.32 Algorithm of Risson (2003). First, noise is removed by prefiltering the image. Next, the image is segmented. Regions that do not satisfy the assumptions of the dichromatic reflection model are removed. The dichromatic line is computed for the remaining regions. Finally, the most probable point of intersection is computed.

Risson (2003) The segmentation threshold was set to 0.05. A minimum of five pixels were required for each segmented region. Regions with a saturation less than 0.12 were excluded. The curve of the black-body radiator was not used. The intersections were computed in RGB chromaticity space. All intersections between dichromatic lines were computed and the median, separately for the x-and y-direction, was taken as the position of the illuminant. (threshold=0.05, nLimit=5, minSaturation=0.12)... 

---