Multiscale relaxation labeling of fractal images

Multiscale relaxation labeling for segmentation of fractal images is described. The images used are of pavement distress, on which simple edge detection schemes perform poorly. Relaxation labeling is used to improve upon initial edge-based segmentation. A multiscale relaxation technique is used in a pavement distress detection system. To better model pixel interactions, nonlinear terms are included in the relaxation process. Symmetry arguments and careful engineering allow a 93% reduction in the complexity of this approach. To demonstrate the necessity of the multiscale approach, examples with and without multiscale relaxation are shown. It is found that performance is greatly improved by multiscale relaxation