Abstract
Many studies have shown that the apparent contrast of a visual target varies as a function of the physical contrast of the surround. This perceptual phenomenon has generally been attributed to the inhibitory effects of the context on the neural response corresponding to the target. Based on the assumption that visual system must instantiate the statistics of spectral light experienced in natural visual environments to guide appropriate behavior, here we examine an alternative idea, namely that this perceptual effect is generated by the locally co-occurring spectral variations of target and context that observers have experienced in natural scenes. We tested this hypothesis using a large database of natural scenes in which each image was carefully calibrated to insure that the values measured for all the three color channels were linear responses to spectral light. By sampling ∼109 center-surround patches in these images, we determined in both achromatic and achromatic axes (i.e., the L−M and S−[L+M] axes) the probability distribution functions of the contrast at the center of the patches, given the contrast of the surround (this analysis was repeated for different center-surround areas of 1–2, 1–4, 2–4, and 2–8 degree). The results of the analysis show that these statistical functions accurately predict the way the perceived contrast of a visual target changes as a function of contextual contrast. We conclude these contrast effects are generated by an instantiation in neural circuitry of the statistical structure of the co-occurring contrasts that human observers have always experienced in natural environments.
This work was supported by the NIH and the Geller endowment.