基于全变分模型的图像分割方法及其算法-计算数学专业论文.docxVIP

I I PAGE PAGE VI 摘 要 近年来, 基于变分偏微分方程的图像分割模型已经成为图像分割领域内的重要研究 课题. 本文主要是基于T Vg + L1模型和MS模型提出有效的数值算法, 主要工作安排如下: 第一章介绍本论文用到的一些符号和相关定义以及相关的模型和算法, 包括全变分 模型, 最优化理论, Bermudez-Moreno(B-M)算法, Chambolle算法以及初始对偶方法等. 第二章, 针对T Vg + L1模型中的不可微项导致的数值困难, 我们分别提出用投影梯度 法(也叫经典的半隐式梯度下降法)和增广拉格朗日方法求解. 同时, 我们指出投影梯度法 可以看作为BM算法的特殊形式, 并证明该算法收敛性. 通过实验结果比较了这两种算法 的分割结果. 第三章, 基于MS分割模型提出快速图像分割模型和算法—投影梯度算法. 数值实验 表明该算法比Chambolle算法有较快的收敛速度, 具有较好的抗噪性. 此硕士论文得到了河南省科技厅项目(No.132300410150)的资助. 关键词: 图像分割; 全变分; 投影梯度算法; ALM算法. 基于全变分模型的图像分割方法及其算法 基于全变分模型的图像分割方法及其算法 PAGE PAGE III PAGE PAGE IV ABSTRACT Image segmentation recently becomes an important research topic in the ?eld of image processing. Our main work is based on the T Vg ? L1 model and the MS model and then propose some e?cient numerical methods to solve them. The main results of this dissertation are arranged as follows: In the ?rst chapter, we list some symbols and their de?nitions which will be used in this paper. We also give the basic knowledge of some related models and numerical methods such as the total variational model, optimization theory, the B-M algorithm and the Chambolle algorithm, the primal-dual method, etc. In the second chapter, we propose the semi-implicit gradient descent method (Cham- bolle’s method) and the augmented Lagrangian method to solve the T Vg ? L1 model in order to overcome its numerical di?culties due to the non-di?erentiablity. We also deduce that the semi-implicit gradient descent method can be regarded as the special form of the BM method and then prove the convergence of the proposed algorithm. In the third chapter, based on the MS segmentation model, we propose to use the projected algorithm for solving it. Some numerical experiments show that this method has faster convergence speed than Chambolle algorithm as well as good noise resistance. This master thesis is funded by the project of Science and Technology Agency, Henan province. (No. 132300410150). KEY WORDS: Image segmentatio