A computational model for color assimilation illusions and color constancy

Oguzhan Ulucan, Diclehan Ulucan, Marc Ebner; Proceedings of the Asian Conference on Computer Vision (ACCV), 2024, pp. 630-647

Our visual system unconsciously estimates the objects reflectance in the scene. Even under different illumination conditions, it can discount the effects of the illuminant to recognize the true colors of the objects. Yet, under some circumstances, the perceived color can differ from the actual reflectance. Color illusions can be given as an example of such circumstances. While computer vision studies aim at estimating the scenes illuminant, computational biology studies mostly aim at reproducing our sensation on color illusions. However, as stated in many studies, an algorithm mimicking our perception should be deceived by color illusions, while estimating the reflectance under varying illumination conditions. Yet, to the best of our knowledge, there is no study that mimics our sensation on color illusions and also investigates computational color constancy in detail by using a single method. Based on this motivation, we develop a single method that mimics our behavior on color assimilation illusions and performs color constancy. In particular, we propose a multiresolution color constancy strategy that operates in scale-space. In our approach, we make use of a variant of the local space average color method which we further modify by considering the gradient changes of the scene. According to the experimental results, our algorithm mimics our sensation on color illusions, and it presents competitive results on 4 different color constancy benchmarks.