Based On Markov Random Fields For Sar Image Restoration And Segmentation

With its ability to image any targets on the earth,together with high resolution under nearly all weather conditions, synthetic aperture radar (SAR) has been widely used. However, because of its coherent imaging principle, the speckle phenomenon strongly hinders data interpretation with standard image analysis tools.This task is dedicated for two aspects:1) Study the algorithms for despeckling, enhancing, edge extraction and region segmentation. The emphasis is image restoration and segmentation based on Markov random field (MRF).2) Implement those algorithms with VC++ under Windows 2000 platform.Selecting a proper neighborhood system and using the ability of Markov random field to describe spatial dependence, MRF can be used to model the structural and textural behavior of images. Selecting a appropriate model and making use of the optimal algorithm —simulated annealing to estimate the parameters, wonderful image restoration can be achieved.Watershed transformation, a mathematical morphological method, is used for initial segmentation. Regions description and merging are done with Gaussian MRF and ideal results can be attained.In this article, Markov random field models, simulated annealing and watershed transformation used together for SAR images processing are discussed internally for the first time; a simple and effective method is presented for deducting textural information in SAR images; J. Schou's method for image restoration is extended from polarimetric SAR images to usual SAR images; two methods for reconstruction of gradient images, based on which watershed transformation is done, are put forward; furthermore, a new method is presented for quantizing 16-bit SAR data.