| In the natural environment, the spectral composition of the external illuminant ischanging with the time. The color constancy of visual system is to boycott the alterationtriggered by external illuminant and keep constant perception to the physical property ofnatural scene, e.g. the color of objects.Based on the sufficient findings of physiology and psychophysics in HVS for colorconstancy, this paper proposed several computational models for color constancy.(1) The retina regarded as the door of visual information processing plays veryimportant role in various vision tasks. However, few research probes into the roles ofretina acting in color constancy. In this paper, we assume that the color constancyachieves with some extent during the retina information processing for the first time.The model based on retina mechanisms adequately utilizes the characteristics of variousretina cells, especially the lateral inhibition of horizontal cell and the center-surroundreceptive field of bipolar cell for color information processing, to construct a retinanetwork for color constancy. Thorough evaluation on three commonly used imagedatasets demonstrates the feasibility of color constancy on retina.(2) The double-opponent color cell in primary visual cortex (V1) of the humanvisual system have always been regarded as the physiological basis of color constancy.We introduce a neuronal mechanism based model for color constancy by imitating thefunctional properties of the double-opponent cells. We find that the true sceneilluminant could be easily and accurately estimated from the responses ofdouble-opponent cells with a certain receptive field (RF) structure to the inputcolor-biased images. Experimental evaluations on two commonly used image datasetsshow that the proposed model performs best among the most of existing color constancymodels. |
