A Color-Opponency Based Biological Model for Color Constancy

Color constancy is the ability of the human visual system to adaptively correct color-biased scenes under different illuminants. Most of the existing color constancy models are nonphysiologically plausible. Among the limited biological models, the great majority is Retinex and its variations, and only two or three models directly simulate the feature of color-opponency, but only of the very earliest stages of visual pathway, i.e., the single-opponent mechanisms involved at the levels of retinal ganglion cells and lateral geniculate nucleus (LGN) neurons. Considering the extensive physiological evidences supporting that both the single-opponent cells in retina and LGN and the double-opponent neurons in primary visual cortex (V1) are the building blocks for color constancy, in this study we construct a color-opponency based color constancy model by simulating the opponent fashions of both the single-opponent and double-opponent cells in a forward manner. As for the spatial structure of the receptive fields (RF), both the classical RF (CRF) center and the nonclassical RF (nCRF) surround are taken into account for all the cells. The proposed model was tested on several typical image databases commonly used for performance evaluation of color constancy methods, and exciting results were achieved.