Abstract
Over the last few years, I have shown that images of 3D objects contain highly organized patterns of orientation and spatial frequency information (‘orientation fields’), which are systematically related to 3D shape. Here we present a novel illusion and an adaptation experiment that provide the first direct evidence that orientation fields are sufficient to drive 3D shape perception.
The logic of the illusion is as follows. If orientation fields play an important role in 3D shape estimation, then it should be possible to synthesize 2D patterns of orientation that elicit 3D shape percepts. We did this by ‘smearing’ random noise along specific directions in the image (derived from a 3D model). The result is a 2D texture pattern–generated entirely through filtering operations–that appears vividly like a 3D object. We call this illusion “shape from smear”. A depth discrimination task showed that naïve subjects reliably perceive specific 3D shapes from such stimuli.
Because ‘shape from smear’ is an entirely 2D process, we can modify the orientations and scales in the image and measure the effects on perceived 3D shape. The more we ‘smear’ the noise pattern, the more 3D the image appears, suggesting that we are manipulating the image information that the visual system uses for estimating shape. We also find that orientation variations are more important than spatial frequency variations.
Most importantly, we used shape from smear to induce 3D shape percepts by adaptation. We created ‘anti-shape’ textures by smearing noise along directions orthogonal to the correct directions for a specific 3D shape. Prolonged viewing induces local orientation adaptation, which makes a subsequently presented neutral isotopic noise pattern appear like a specific 3D shape. Thus, for the first time we can show that local orientation detectors are directly involved in the perception of 3D shape.
RF supported by DFG FL 624/1-1.