Explaining Low-Level Brightness-Contrast Illusions Using Disinhibition

Conventional Difference of Gaussian (DOG) filter is usually used to model the early stage of visual processing. However, convolution operation used with DOG does not explicitly account for the effects of disinhibition. Because of this, complex brightness-contrast (B-C) illusions such as the White’s effect cannot be explained using DOG filters. We discovered that a model based on lateral disinhibition in biological retinas allows us to explain subtle B-C illusions. Further, we show that a feedforward filter can be derived to achieve this operation in a single pass. The results suggest that contextual effects can be processed through recurrent disinhibition. Such a context sensitive structure might be employed to improve network robustness of visual capturing or displaying systems. Another potential application of this algorithm could be automatic detection and correction of perceived incoherences where accurate perception of intensity level is critical.