Abstract
We probe visual information processing in humans using a well-known shape-dependent color-after-image (Van Lier et al., 2009), in which the elicited after-image is contingent upon a shape outline presented after adaptation. We investigated the extent to which the induced after-image transfers across mismatches between the adapting stimulus and the subsequent test outline. Each participant (N=12) completed two sessions of three blocks. We varied the test outline in size (0 to 100% increase in diameter), rotation (0 to 24 deg), or both, relative to the adapting stimulus. During each trial, a 1-second view of the adapting stimulus was followed by the presentation of a test outline. In a 3-AFC, participants reported whether the after-image either fit the outline exactly, did not fit the outline exactly, or no after-image was present. Fitting psychometric functions to these reports separately for each observer and condition, we determined 50% thresholds where (1) observers reported seeing an after-image, and (2) the after-image fit the outline exactly. For (1), observers reported after-images 50% of the time with a rotation of up to 7 deg and a size increase of 66% from the adapting stimulus. For (2), the threshold was at a rotation of 4 deg and a size increase of 16% from the adapting stimulus. Our results indicate that the color after-image is robust to differences between the adapting stimulus and the test outline. To gain understanding into the underlying perceptual mechanism, we developed a general computational model architecture that emulates visual information processing via interconnected recursive filters. With this, we implemented constraints between processing of shape and color. This model qualitatively reproduced our psychophysical results and generated novel predictions about perceptual consequences of shape-contingent color adaptation, which were validated by our experiments. Models of interconnected recursive filters thus hold promise for gaining insights into perceptual mechanisms.