Abstract
There is a widespread view that higher-level mechanisms, which help to parse the visual scene into components of reflectance and illumination, are required to explain a variety of brightness illusions, and that low-level spatial filtering explanations are inadequate. This idea arises, in large part, from the erroneous assumption that controlling for local (border) luminance differences rules out low-level spatial filtering explanations. This assumption would be valid if spatial filtering was performed solely by high-frequency “edge-detectors”; however, it is now understood that this is not the case and that filtering occurs at multiple spatial scales (Hess, 2003; Wilson & Wilkinson, 2003). This means that areas remote from the test patches can influence their brightness due to multiscale filtering alone and that holding local luminance constant is an inadequate control. Here we investigate the degree to which low-level multiscale filtering can account for brightness perception in a set of visual illusions often cited as evidence for the necessity of higher-level mechanisms. Although local luminance remains essentially unchanged in these stimuli the larger contexts within which these identical targets are embedded do not. It is thus possible that the observed brightness effects can be accounted for by multiscale filtering mechanisms. Psychophysical brightness and lightness matches were obtained at key locations within the stimuli and were compared with predictions from the ODOG multiscale filtering model (Blakeslee & McCourt, 1999). The ODOG model predicted the brightness effects in the snake illusion (Somers & Adelson, 1997), in simultaneous brightness contrast stimuli with multiple illumination cues (Williams, McCoy, & Purves, 1998), in the paint/shadow illusion (Hillis & Brainard, 2007), and within photographs of natural scenes (Gilchrist, 2006). We conclude that although higher-level mechanisms may enhance the magnitude of these illusions under some circumstances, multiscale filtering provides a unified and parsimonious explanation of these effects.
NIH COBRE Grant P20 RR020151.