The PI has developed a methodology to group token elements in 2-D or 3-D, in the presence of noise and outliers. These tokens can take form of points, edgels, surface elements, or any combination of these. Each site collects all the votes cast at its location and encodes them. A local, parallel routine then simultaneously detects junctions, curves, regions and surfaces. The method is non-iterative, requires no initial guess, and the only free parameter is scale. This research will extend the foundation work in three directions: formalization, validation, and extensions. Formalization will unify the different elements using tensor calculus for representation, and non-linear polling for data communication, and produce integrated descriptions for features. Validation will be performed via controlled experiments, with various inputs and known amounts of noise, and results will be compared with imagery and methods from other sites. Extensions will be applied to stereo, motion segmentation, and shape description.