Highlight and shading invariant color image segmentation using simulated annealing

Color constancy in color image segmentation is an important research issue. In this paper we develop a framework, based on the Dichromatic Reflection Model for asserting the color highlight and shading invariance, and based on a Markov Random Field approach for segmentation. A given RGB image is transformed into a R'G'B' space to remove any highlight components, and only the vector-angle component, representing color hue but not intensity, is preserved to remove shading effects. Due to the arbitrariness of vector angles for low R'G'B' values, we perform a Monte-Carlo sensitivity analysis to determine pixel-dependent weights for the MRF segmentation. Results are presented and analyzed.