一种基于Graph Cuts的SAR图像分割方法

A New Algorithm Based on Graph Cuts for SAR Image Segmentation

在最小化由马尔科夫随机场(MRF)图像分割模型建立的能量函数方面,基于Graph Cuts的alpha-expansion是一种比较有效的算法。但是,由此算法构建的s/t图中边的数目非常多,运算速度很慢。为了减少alpha-expansion算法的计算量,本文在标号为alpha的像素向其它像素膨胀的过程中,先隔离非alpha类间的联系,而只考虑alpha类与非alpha类之间的关系,从而避免了alpha-expansion算法需要构造辅助结点的问题,减少了s/t图中边的数目,提高了算法的计算效率。因放松了非alpha类间的关系对alpha膨胀的约束,使得算法可以更容易得跳出能量函数的局部极小点而获得更优的分割结果。实验中将改进的算法与传统的基于Graph Cuts的算法做了对比,显示了新算法在运算时间和最小化能量方面的有效性。

Alpha-expansion algorithm based on Graph Cuts is a useful method for minimizing energy function established by Markov Random Field model of image segmentation. However, the number of edge of the graph constructed by the algorithm is large, so the speed is rather slow. In order to reduce the complexity of the alpha-expansion algorithm, the relationship between the pixels labeled alpha and pixels labeled non-alpha was only considered and the relationship between the pixels labeled non-alpha was omitted. This idea avoided adding auxiliary nodes which were one of main factor to affect the efficiency of the alpha-expansion algorithm. Since the constraints on the relationship between pixels set labeled different non-alpha was loosed,it was easier for the new algorithm to escape some local minimum of energy function so as to gain more optimal segmentation results. In experiments, compared with the standard algorithms based on Graph cuts, the algorithm had the better performance in the running time and the minimum energy.