Grouping based on coupled diffusion maps

Systems of coupled, non-linear diffusion equations are proposed as a comput ational tool for grouping. Grouping tasks are divided into two classes - lo cal and bilocal - and for each a prototypical set of equations is presented . It is shown how different cues can be used for grouping given these two b lueprints plus cue-specific specialisations. Results are shown for intensit y, texture orientation, stereo disparity, optical flow, mirror symmetry, an d regular textures. The proposed equations are particularly well suited for parallel implementations. They also show some interesting analogies with b asic architectural characteristics of the cortex.