Abstract
Achromatic perceptual transparency has been studied extensively over the past 40 years. Recent models proposed by Robilotto and Zaidi (2002; 2004) and Anderson et al (2006; 2008) suggest that perceived image contrasts determine the degree of perceived achromatic transparency (or opacity) of a visual surface. However, Robilotto and Zaidi suggested that perceived transparency is determined by the perceived contrast of the image region occupied by the perceived transparent surface alone, whereas Anderson et al proposed that it is determined by the ratio of the perceived contrasts in the transparent and ‘plain-view’ regions.
Anderson et al (2008) argued that observers equate perceived-contrast ratios when asked to match opacity, but that whether or not observers are also able to explicitly match perceived-contrast ratios (when asked to match contrast ratios, rather than opacities) is irrelevant to their opacity theory. In fact, if observers are unable to match ratios of perceived contrasts, then Anderson et al's theory would be logically incoherent, since no unique ratio structure for perceived contrast would exist (two patterns could have equal contrast ratios according to one measure and unequal contrast ratios according to another). In my experiments, observers' matches of perceived-contrast ratios (to the extent that observers were able to make them) were very different from their opacity matches using the identical experimental setup (Albert, 2008).
Recent evidence suggests that both perceived filter contrasts and filter-to-background relative properties (contrast ratios, filter boundary contrast, and mean luminance ratios) influence opacity matching (Albert, 2008). Observers tend to equate perceived filter contrasts in matching opacity so long as filter-to-background relative properties remain qualitatively similar in the target and match patterns. Otherwise, making filter-to-background relative properties qualitatively similar will be more important than equating perceived filter contrasts. I propose a new model that is consistent with the existing data.