The interpretation of visual motion: evidence for surface segmentation mechanisms

G R Stoner,T D Albright
DOI: https://doi.org/10.1016/0042-6989(95)00195-6
Abstract:The independent motions of objects in a visual scene are commonly manifest as overlapping retinal motions. A consequence of this overlap is the creation of spurious retinal image features--such as corners and terminated contours--that bear no direct relation to the motions of the objects that give rise to them. To reconstruct object motions, these emergent features must be distinguished from the retinal motions of real object features. This process can be studied using visual stimuli known as plaid patterns, which provide a laboratory archetype for the ubiquitous real-world circumstance of two surfaces with overlapping retinal projections. By adjusting luminance relationships in a plaid pattern it is possible to influence the perceptual interpretation of image features, such that they are seen as either an emergent consequence of occlusion or as real variations in surface reflectance. In the former case, the plaid is most likely to be to perceived as two independently moving surfaces, whereas the latter generally elicits a percept of a single moving surface. This dependence of motion perception on luminance configuration can be viewed as evidence for the involvement of surface segmentation mechanisms, which distinguish between real and emergent image features by promoting a depth-ordered neural representation of surfaces. An alternative interpretation, which does not demand such depth-ordering and feature classification, asserts that the effect of luminance configuration can be accounted for by attendant variations in the distribution of moving Fourier components. To evaluate these two proposed mechanisms, we designed novel plaid stimuli in which surface segmentation cues could be varied independently of changes in the distribution of Fourier components. Perceived motion was found to be highly correlated with the presence of appropriate segmentation cues and uncorrelated with the distribution of Fourier components. These results refute the Fourier components hypothesis, and they support our proposal that surface segmentation plays a critical role in the interpretation of visual motion signals.
What problem does this paper attempt to address?