图像分割问题的梯度投影算法  

Gradient Projection Algorithm for Image Segmentation

作　　者：江成顺[1] 崔颖[1] 

机构地区：[1]中南财经政法大学武汉学院,武汉430079

出　　处：《系统仿真学报》2013年第7期1446-1450,共5页Journal of System Simulation

基　　金：湖北省自然科学科技计划项目(2011CDC143);国家863项目子课题(2009AA012200)

摘　　要：针对Berkels提出的无约束严格凸的图像分割模型,在Bermudez-Moreno对偶算法的基础上提出了一种不仅易于实现并能快速得到全局极小值的梯度投影算法。首先,解释了Bermudez-Moreno对偶算法可用于模型问题的求解并推导出迭代计算格式;然后,证明了迭代格式的收敛性并分析了其收敛速度;最后,仿真实验表明了所提出算法能够有效地、快速地求解图像分割问题。A gradient projection algorithm was algorithm for Berkels＇ unconstrained strictly given relying on Bermudez and Moreno＇s duality-based convex image segmentation model, which is easy to implement and allows us to get a fast global minimization. Firstly, Bermudez-Moreno algorithm could be used for Berkels＇model and the iterative scheme was deduced. Secondly, the convergence of the proposed scheme was proved and its convergent rate was analyzed. At last, the study was illustrated with numerical examples, which shows the scheme is efficient and powerful for image segmentation.

关 键 词：图像分割 全局极小值 对偶算法 梯度投影算法 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]