Neural Computation of Perceived Relative Size and Depth in Complex 2D Image Configurations

摘要

The neural networks of the human visual brain are capable of extracting 3D structure from specific 2D cues available in planar images. Many of the functional principles governing this ability are still not fully understood. Neural models backed by psychophysical data predict how local differences in either luminance contrast or physical size of local boundaries in 2D images may determine the perception of 3D structure, but do not generate predictions relative to the role of color in this process. To further clarify the potential contribution of color to 3D perceptual organization, we created 2D image configurations with multiple surface representations where the relative physical size of local boundaries between contrast regions was held constant. The only potential cues to 3D available in the images were specific local combinations of color and luminance contrast. Psychophysical experiments with human observers were run to test for selective local effects on the subjective relative depth and the subjective relative size of image regions. It was found that response probabilities for subjective depth and subjective size are systematically and consistently determined by local surface colors and their immediate backgrounds. The results show consistently varying perceptual judgments with a statistically significant correlation between subjective depth and subjective size. Moreover, there is a color specific effect on both dependent variables, and this effect depends on the polarity of the immediate surround of the reference surface rather than local center-surround contrast intensity. These findings are not predicted by any of the current neural models and suggest that the perceptual mechanisms generating 3D effects from 2D visual input selectively exploit specific color and background cues to enable the intrinsically coherent 3D perceptual organization of otherwise ambiguous 2D images with multiple surface representations.