Low level image segmentation with high level `emergent properties´: color based segmentation

A method is presented which incorporates a discontinuity detection process into a multigrid relaxation algorithm, with the goal of recovering significant discontinuities at different scales. Line processes are activated in a deterministic way, depending on local properties of both neighboring line processes (at different scales) and neighboring continuous variables. Computational complexity is O(n ) for an image with n pixels, and convergence time is a small multiple of that required by one relaxation step at the finest grid. The scheme is applied to the problem of image segmentation based on color differences. These dissimilarities are detected by considering changes in the relative intensity of the red, green, and blue components of the pixels adjacent to a given discontinuity. A final relaxation step restricted within the detected boundaries is then suggested as a way of `coloring´ the delineated regions in a uniform way. The algorithm has been implemented with high efficiency on a MIMD parallel computer with distributed memory. A coarse grain decomposition is found to be useful for this and other multiscale problems