Abstract
The highest luminance anchoring principle (HLAP) asserts the highest luminance surface within an illumination field appears white and the lightness of other surfaces are computed relative to the highest luminance. HLAP is a key tenet of the anchoring theories of Gilchrist and Bressan, and Land's Retinex color constancy model. The principle is supported by classical psychophysical findings that the appearance of incremental targets is not much affected by changes in the surround luminance, while the appearances of decremental targets depends on the target-surround luminance ratio (Wallach, 1948; Heinemann, 1955). However, Arend and Spehar (1993) showed that this interpretation is too simplistic. Lightness matches made with such stimuli are strongly affected by instructions regarding either the perceptual dimension to be matched (lightness versus brightness) or the nature of illumination when lightness judgments are made. Rudd (2010) demonstrated that instructional effects can even transform contrast effects into assimilation effects. To model these results, I proposed a Retinex-like neural model incorporating mechanisms of edge integration, contrast gain control, and top-down control of edge weights. Here I show how known mechanisms in visual cortex could instantiate the model. Feedback from prefrontal cortex to layer 6 of V1 modulates edge responses in V1 to reorganize the edge integration properties of the V1-V4 circuit. Filling-in processes in V4 compute different lightnesses depending on the V1 gain settings, which are controlled by the observer's conscious intention to view the stimulus in one way or another. The theory accounts for the instruction-dependent shifts between contrast and assimilation.
Meeting abstract presented at VSS 2013