Active Contour Model For SAR Image Segmentation And Its Algorithm

Due to the synthetic aperture radar(SAR)can avoid being influenced by objective condition such as light,time and climate,these advantages make it can be widely used in many areas.However,the specificity of the SAR system imaging mechanism and the diversity of different object backscattering,which make the work of SAR image processing become very difficult.The fuzzy extent of object edges is further aggravated by the inherent speckle noise in SAR images.Therefore,the object edge detection is the important research content in SAR image processing areas.The quality of SAR image interpretation task will be affected directly by the level of contour extraction accuracy.For this reason,it is very necessary to design a high efficient algorithm which can be used in contour extraction of SAR image.However,the existing algorithms have many disadvantages such as high complexity of algorithms,bad noise elimination and low accuracy.To alleviate these problems,the following meaningful research works are carried out in this paper:1)Firstly,the statistical features of SAR image are modeled by the Gamma distribution.Then,according to the characteristic of Ayed model which is low computational complexity but the poor ability of noise reduction,and the characteristic of high computational complexity but the good ability of noise reduction of local Gaussian distribution fitting energy(LGDF)model,a region-based active contour model is proposed by combining the merits of Ayed model and LGDF model,the proposed model not only has excellent noise suppression but also has low computational complexity.Finally,the validity of the proposed method is verified by experimental results for synthetic and real SAR images.2)Due to the poor ability of existing methods in terms of processing the sharp and weak edges,an edge-region active contour model is proposed based on the content above.Firstly,the proposed model consists of two main energy terms: an edge-region term and a regularization term.The edge-region term is derived from a Gamma model and gradient term model,which can process the speckle noises and drive the motion of the curves toward desired locations.The regularization term is not only able to maintain a desired shape of the evolution curves but also has a strong smoothing curve effect and avoid the occurrence of small,isolated regions in the final segmentation.Then,the gradient descent flow method is introduced for minimizing the proposed energy functional.Experimental results not only show that the proposed method is not sensitive to the contour initialization location,but also demonstrate the fast convergence speed and desirable property to process inhomogeneous regions in SAR images compared with other methods.