Lab Home | Phone | Search | ||||||||
|
||||||||
We seek to understand the principles underlying visual information processing as an abstraction in computational neuroscience. A theory of early computational vision, from boundary detection to color, shading and stereo, is sketched. The theory is geometrical. It is derived by identifying the columnar organization prevalent in primate neurobiology with the tangent bundle. We consider in particular how long-range horizontal connections relate to connection forms (curvatures). For example, edge elements are identified with a local approximation to tangents and curvature is necessary to introduce consistency between them. Local circuitry for these computations in primate V1 is described. The resultant models quantitatively predict the anatomy and physiology of long-range horizontal connections to second-order, and provide a foundation for inferences about perceptual surfaces. Steven W. Zucker Dept. of Computer Science and Program in Applied Mathematics Yale University New Haven, CT 06520-8285 steven.zucker@yale.edu Host: Eddy Timmermans/T-CNLS |