Model for reflectance perception in vision

Human vision routinely compensates for the illumination field and is mostly sensitive to scene reflectance. This paper presents a biologically inspired mathematical model that estimates the illumination field of a scene and compensates for it to produce the output image that is mostly modulated by the scene reflectance. Since the illumination field is responsible for wide variations in images, the present model is seen as an approach to illumination-invariant imaging. The model can be conveniently implemented in an analog silicon retina incorporating modified cellular neural network for the computation of the illumination field. We present several numerically obtained results on scenes with widely varying illumination conditions.