A renormalization group approach to image processing problems

A method for studying problems in digital image processing, based on a combination of renormalization group ideas, the Markov random-field modeling of images, and metropolis-type Monte Carlo algorithms, is presented. The method is efficiently implementable on parallel architectures, and provides a unifying procedure for performing a hierarchical, multiscale, coarse-to-fine analysis of image-processing tasks such as restoration, texture analysis, coding, motion analysis, etc. The method is formulated and applied to the restoration of degraded images. The restoration algorithm is a global-optimization algorithm applicable to other optimization problems. It generates iteratively a multilevel cascode of restored images corresponding to different levels of resolution, or scale. In the lower levels of the cascade appear the large-scale features of the image, and in the higher levels, the microscopic features of the image