Scenes composed of surfaces with similar colors and textures (for example, forest scenes) may appear as a continuous mass to a stationary observer. Once the observer starts moving, however, the surfaces segregate from one another. How does our visual system use motion to organize such scenes? There is considerable evidence that observers use motion discontinuities to locate the boundaries between surfaces. But this approach can only cut the scene into patches. A partially occluded surface may appear as many disconnected patches in the image and to recover this surface, the visual system needs a way to group patches. Recent evidence indicates that the visual system may group patches of motion by fitting the visual stimulus with a global motion pattern. Motions that are consistent with the pattern are grouped and segregated from motions that are inconsistent with the pattern. The goal of this research is to characterize such global segmentation strategies. One set of experiments examines whether the visual system treats motion patterns in much the same way that it treats static patterns. These experiments look for evidence that the visual system applies the Gestalt grouping rules of `good continuation` and `bilateral symmetry` to motion stimuli. A second set of experiments considers global motions not as simple 2D patterns, but rather as the result of imaging 3D surfaces. These experiments compare performance on patterns that have similar 2D image properties but very different 3D interpretations. And finally, since the vast majority of natural flow patterns evolve over time, a third set of experiments examines the segmentation of motion patterns that vary over time and space. The building of a surface representation is thought to be the fundamental goal of mid-level vision. Thus to understand human vision it is essential to understand how the visual system organizes scenes with multiple surfaces. This research is important because it examines an unexplored and potentially powerful approach to this segmentation problem.
