The Research On Active Contour Model For SAR Image Segmentation

Synthetic aperture radar(SAR)is a microwave imaging system with full-time,allweather and penetrating characteristics.It is an important techniques for ground observation and has been widely used in military and civilian fields.SAR image segmentation is the basis of SAR image interpretation which becomes a hot spot in the field of remote sensing.Therefore,SAR image segmentation plays a key role in subsequent image understanding.Due to the imaging mechanism of SAR images,SAR images contain inherent multiplicative speckle noise,which leads to SAR images having gray unevenness,weak/blurred edges,irregular target edges and complex texture information.Because of the presence of speckle,SAR image segmentation brings great challenges and the optical image segmentation methods are not suitable for SAR image segmentation.To address the above problems in SAR image,based on the summary and analysis of existing methods,the thesis studies the description of local statistical information in SAR images,texture feature extraction and fusion mechanism,the active contour model modeling for image segmentation.The research results were summarized as follows:(1)SAR image segmentation method is proposed based on local region information by summarizing the existing SAR image segmentation methods.The method first provide a Gamma statistical distribution of SAR image for the variations of backscattering intensities in SAR images.Then,a modified region mean estimation formula is proposed on Gamma distribution.The energy function of the active contour model is further constructed.Finally,Gaussian filtering is employed to regularize the level set function and obtained SAR image segmentation.The experiment results verify the efficiency of the proposed method.(2)This thesis presents an active contour model based on a reaction–diffusion(RD)theory for a SAR image segmentation algorithm to address the problem of intensity inhomogeneous in SAR images and re-initialization in active contour model.The well-known reaction diffusion theory consist of two main parts:reaction and diffusion terms.Firstly,we constructed the reaction terms in an energy function integrating the Gamma statistical distribution and edge information for SAR images to simultaneously suppress speckle noise and drive contour toward the object boundaries.Then,partial differential equation is utilized to compute the proposed energy function.Finally,to ensure stability of the level set function and regularize the segmented region,diffusion term is introduced into the level set equation(LSE).The experimental results on both synthetic and real SAR images show that the proposed model has good robustness against speckle noise as well as higher segmentation efficiency and accuracy than some existing models.(3)The thesis proposes a matrix factorization active contour model based on fused features for SAR image segmentation.Due to the complex texture features and scenes characteristics of SAR image,firstly,feature maps(matrix)are constructed by combining wavelet textual features,Do G filter features and Gabor filter features via local spectral histogram,which improves spatial pattern and express image structure.Secondly,region information is obtained via matrix factorization theory on the feature matrix.Edge information is obtained by modified the ratio of exponentially weighted averages(ROEWA)operator.Combining region and edge information,an energy function is constructed.Moreover,to avoid the local minimum,a convex energy function is proposed.A fast dual formulation is introduced for the evolution of the contour.Finally,Synthetic and real SAR data are used for verification.The experimental results demonstrate the proposed algorithm is effective for SAR image segmentation.(4)An active contour model based on wavelet domain is proposed for SAR image segmentation.In order to suppress the influence of speckle noise and solve the problem that the weak edge of SAR image is difficult to extract,the SAR image based on wavelet transform is first to be preprocessed.Then,based on the image obtained by wavelet transform,a new energy function is constructed by embedding fuzzy region competition into the region term and edge term.Furthermore,the fuzzy clustering results is used as initial contour for reducing the computing time.Finally,the experimental results verify the robustness and the effectiveness of the proposed method.