A Retinal Mechanism Based Color Constancy Model

It has been generally accepted that the retina is the first level that plays critical role in the function of color constancy in human visual system. In this paper we propose a computational model which imitates the neural mechanisms of color information processing in retina. In this model we first compute a simple statistics of reflectance about the real scene by simulating bipolar cells. Then, the ganglion cells (e.g., the midget cells) receive both the computational statistics of reflectance of the scene from bipolar cells and the response from horizontal cells as well as amacrine cells. Subsequently, the ganglion cells provide a stable output independent of the color of illuminant. We tested our model on a commonly used large-scale linear image dataset, and the results demonstrate that our model provides better color constancy performance than those widely accepted color constancy models.