Image processing based on Gibbs models

Gibbs models are being applied in image processing due to easiness of taking into account of pixel interactions. An iteration class of algorithms known as stochastic relaxation is the most powerful tool for solving a great many tasks in image processing. In spite of the fact that many researchers concern themselves with Gibbs model application to image processing, there are problems in this field that are not solved up to date. In this work, authors present some new results obtained recently. In particular new methods of finite-state Gibbs model parameter estimation based on sufficient statistics and on conditional moments, algorithms of texture segmentation, algorithms of grayscale and color picture reconstruction are presented. These algorithms are suitable for reconstructing not only artificially distorted model images, as in most cases of published works, but for reconstructing the images obtained by plain cameras as well. For the purpose of increasing the quality of grayscale and color picture reconstruction a new hierarchical two-level Gibbs model using as a lower layer a line process describing edges of regions is suggested. Experiments give a demonstration of increasing the visually perceptible quality of recovered images. Moreover, this algorithm as a by-product extracts edges of picture regions.