Abstract
Image defocus has been studied extensively as a depth cue, but little is known about how the visual system disentangles image defocus from environmental sources such as diffuse surface shading. Here, we show that the perception of defocus can arise even when viewing fully focused, smoothly shaded bumpy surfaces. We further show that introducing sharp contours adjacent to the shaded surface can completely eliminate this percept of defocus, but only if the contours are perceived as self-occluding edges attached to the gradients. In Experiment 1, we generated sharp-edged 'masks' by taking level cuts of a smoothly shaded bumpy terrain and compared the perception of blur and 3D shape of the remaining gradients to images in which the level cut masks were rotated 180 degrees. The geometrically correct level cuts induced a dramatic decrease in perceived defocus and highly reliable 3D shape measurements, whereas the shading gradients bordered by rotated masks exhibited little change in perceived defocus and elicited poor measurements of 3D shape. In Experiment 2, we presented only a narrow strip of the gradients immediately adjacent to the mask edges to determine what led to the dramatic differences in 3D shape and defocus observed in Experiment 1. Our results show that the reduction in defocus and enhancement of 3D shape generated by the 'attached' contours of the level cut masks arises from the perception of surface relief and consistent light field direction that can be derived using information from these contour-adjacent shading strips alone; coherent percepts of 3D shape and illumination direction are experienced only with the strips adjacent to the unrotated masks. Our results demonstrate that the computation of defocus in images of shaded surfaces is intrinsically coupled to the information about 3D shape and illumination that can (or cannot) be derived from photogeometric constraints along sharp contours.
Meeting abstract presented at VSS 2016