Sar Image Segmentation Based On The Level Set

Image segmentation is a fundamental problem for SAR image processing, current level set methods are usually proposed for noncoherent images, thus they cannot correctly model the edge-based and region-based information for coherent SAR images, which limits their applications to SAR images.To address these problems, this thesis presented some studies concentrated in three innovations:(1) A level Set method for SAR image segmentation method combined local and global region information was proposed, for the characteristics of widespread intensity in homogeneity in SAR image, we proposed a local region model, which is based on the local binary fitting model, the exponent kernel function was applied to the LBF model instead of the gaussian kernel function, at the same time we improved the global region-based model with the Gamma probability density function, finally combining these two models.(2) A level set method for SAR image segmentation based on the combing of edge and the region information was proposed, on the analysis of the existing geodesic active contour model (GAC) and Chen-vese model(CV), we combined the unique statistics of characteristics of SAR images to improve the model of edge-based and global region-based level set model. The SAR image edge detection operator ROEWA was applied to the GAC model instead of the edge gradient information, at the same time we improved the global region-based mode with the Gamma probability density function, finally combining these two models.(3) A level set method for SAR image segmentation based on the region statistics information level set method was proposed,this method firstly applied the local gray mean to seek Gamma probability density function,then we construct local statistics information model, in order to solve the slow convergence rate and complex computation of the existing level set methods,we applid the global Gamma probability density difference to improve global piecewise contant (PC) model, finally combining these two models.This work was supported in part by the National Natural Science Foundation of China (No.60971128,61072106,61173090); the Fundamental Research Funds for the Central Universities (No. JY10000902001).