Abstract
Current models suggest that the visual system does not partition image luminance into multiple layers to compute lightness (Gilchrist et al., 1999; Adelson, 2000). Anderson and Winawer (2005) presented illusions that they argue are contrary to this view. However, I show that their displays and analyses are identical in all relevant respects to Metelli's (1974) transparency work, although Metelli did not use slow gradients or emphasize the term ‘lightness illusion.’ Indeed, I show that Adelson's effects are robust to many modifications that reduce or eliminate perceptual transparency, whereas Anderson and Winawer's effects appear to depend strongly on conscious transparency perception. Also, parts of the dark and light targets in Anderson and Winawer's displays are perceived to be in ‘plain view’, and thus under identical conditions of transparency/illumination, making transparency ‘discounting’ unnecessary. In contrast, the dark and light targets in Adelson's displays are everywhere perceived to be under different conditions of transparency/illumination. Moreover, I show that effects similar to Anderson and Winawer's occur in ordinary opaque perceptual occlusion displays, suggesting that layered luminance decomposition is not critical.
Although Anderson and Winawer's displays suggest a role for layered representations in lightness, the visual system need not compute background lightness using the kind of quantitative intrinsic images analysis that Anderson and Winawer advocate. In such displays, background lightness could be computed by 1) simply comparing the luminances of regions perceived to be in ‘plain view’ (as determined by simple heuristics), or applying a frameworks model to these regions, 2) spreading these relative lightness values up to abrupt real or illusory image boundaries (overrunning slow gradients), and 3) combining the previous result in a weighted average with a 2D frameworks or image statistics computation to determine the final lightness percept, with the relative weights determined by the perceptual salience of the layered decomposition.