Abstract
Purpose. Recent classification image (CI) studies have revealed the important role of pixels near interpolated contours in the perception of illusory shapes. However, it is unclear if the direction/magnitude of this influence owes only to contour interpolation, or at least partly to induced surface lightness. Method. We addressed this issue by having subjects discriminate fat and thin noise-corrupted rectangles, the tops and bottoms of which were connected by illusory contours or luminance-defined contours. Inducers were either all black, all white, or mixed (half black / half white). Within twenty days, two naïve participants each performed 10,000 trials for each of the six polarity/contour pairings. An additional twenty naïve subjects each performed exactly 200 trials of the six conditions. A Quest staircase adjusted inducer contrast to yield 70% performance for each observer. Results. First, CIs deriving from the white and black inducers had features of opposite polarity, and the mixed inducers produced the weakest CI features. Second, we used the average noise field of one response category with the mixed inducer as a baseline condition, and compared it with the average noise field of the alternative response category with either the white or black inducers. In the illusory condition, the resulting CI revealed a lighter and darker center region when the inducers were black and white, respectively. Finally, for both illusory and luminance-defined contours, thresholds with mixed inducers were lowest. Conclusions. These results suggest that a) dominant CI features (direction/magnitude) systematically depend on inducer polarity; b) the luminance of the central surface of illusory shapes affects discrimination, despite being far-removed from contours; and c) shape discrimination is more efficient when lightness induction is reduced, perhaps by causing greater reliance on contours. These findings, taken together, indicate that contour interpolation and lightness induction mechanisms interact to produce fat/thin CI characteristics.
Supported by National Eye Institute Grant EY13518 to PJK.