基于全局优化方法的SAR图像快速分割算法

为解决变分水平集分割模型能量泛函的非凸性及其易陷入局部极小值解的问题,研究变分水平集分割模型的全局优化问题.基于Aubert-Aujol (AA)去噪模型和变分水平集方法,提出一个局部统计活动轮廓模型;然后通过凸松弛技术将提出的模型转换成全局优化模型;再利用分裂Bregman技术将全局优化模型转化为两个易于计算的Shrinkage算子和Laplace算子.通过对合成图像和Envisat SAR图像的分割实验,提出的全局分割模型不仅能够快速地得到全局最小值,而且比经典模型更准确地得到图像分割边缘. 

Fast SAR Image Segmentation Algorithm Based on Global Optimization Method

In order to cope with the non-convexity of energy functional of variational level set segmentation model and its easily getting stuck in local minima, a global optimization problem of the variational level set segmentation model had been studied. A locally statistical active contour model (LACM) was proposed based on Aubert-Aujol (AA) denoising model and variational level set method. Then, the proposed model was transformed into a global optimization model by using convex relaxation technique. Finally, the split Bregman technique was applied to transform the global optimization model into two alternating optimization processes of Shrinkage operator and Laplace operator. The segmenting experiments of synthetic images and Envisat SAR images show that, the proposed globally segmentation model can not only obtain a stationary global minimum quickly, but also get the image segmentation boundary more accurately than classic models.