A Study On SAR Image Segmentation Algorithm Base On Markov Random Field

Synthetic Aperture Radar (SAR) is widely used in the area of military surveillance and civil area because of its capability of detecting and investigating to earth interface and penetrating the foliage. Image segmentation is a foundational problem in the area of low level computer vision which is also the one of core technologies realizing the automation interpreting of SAR image, and also is the global problem of SAR research. The application of tradition optical image segmentation method on the SAR image will lead to many unresolved problems. For example, the low accuracy of segmentation, the unsmooth of segmentation frontier and the difficult to fuse the priori information come from high level understand mechanism etc, so it is hard to satisfy the demand of SAR image segmentation and is urgent to create a new segmentation frame which could process in a fast speed and combine the attribution of image low level vision with the priori information of under division section. The application of Markov Random Field image segmentation module tackles the problem much better and makes a great progress in the real SAR image segmentation. But this module has many insurmountable problems, for examples, the introduce of image local correlations which is accompanying with data volumes inflating which leads to a inferior computing speed, the relate optimal algorithm is in a dilemma between the computing efficiency and quality of optimal; the unexpected coherent noise which come from the simple frontier module and image make the segmentation algorithm be hard to balance between frontier protection and anti-noise.This article focuses on the research of Markov Random Field image segmentation module to tackle the problems proposed before which could improve the accuracy of segmentation, the efficiency of algorithm computing and robust. This article would propose a series of improvement measures to MRF module under SAR images, and we also get the satisfying experiment results. The detail of research innovations includes(1) Imposing Graph Cuts technology on target function comes from the Markov Random Field image segmentation module to get the optimized cuts results. The alpha-expansion algorithm in GC can shoot a higher speed under the situation of less noise, but the computing efficiency of this algorithm cann't satisfy the need of actual demanding. Because when the alpha-expansion algorithm is configuring the network chart, meanwhile, we need add auxiliary acmes among the adjacent pixels of different class which would leads to the decreasing of algorithm efficiency. We all know that the coherent image would bring to speckle noise which would lower the locality relation, so we need to add many auxiliary acmes for image and this also do harms to computing efficiency. So I propose a new optimized algorithm to tackle this problem, this new optimized algorithm could improve the efficiency under a good performance of partition accuracy effectively.(2) In the MRF image segmentation module, the key of improving efficiency is determine the proportionate relationships between smooth energy terms and data energy terms. This article proposes a strategy of increasing smooth energy terms step by step which improves the efficiency effectively in the SAR image segmentation.(3) To improve the speed and efficiency of segmentation, we propose a new segmentation framework: Firstly, we partition the original image into many small segments in which has similar grayscale and good boundary. Secondly, in order to enhance the ability of reducing noise of the method, we set each image pixel gray value as the average of the segment to which the pixel belongs. Finally, in iterative optimization stage of the algorithm, each segment in place of pixels was used as the processing units, which reduce the amount of processing unit and improve the efficiency of the algorithm.