Color Visualization of 2D Segmentations

This paper deals with the problem of coloring a two-dimensional segmentation result in such a way that each segment has a different color with the constraint that adjacent segments have sufficiently distinct colors to be easily distinguishable by a human viewer. Our solution for this problem is based on a balanced coloring of the neighborhood graph built from the area Voronoi diagram. The balanced coloring with only a limited number of colors is subsequently modified to assign each segment a unique color. For picking the initial palette of evenly distributed colors, we propose a method that is based on the physical analogy of energy minimization in a Wigner crystal. To make the energy formula applicable to arbitrary color spaces and distance measures, we give generalized definitions for the center and radius of a color space with an arbitrary distance metric. The individual steps of our coloring algorithm have been evaluated on the PRIMA dataset.