基于变分的CT图像环形伪影校正  

Variation-based Ring Artifact Correction in CT Images

作　　者：霍其润[1] 李建武[2] 陆耀[2] 秦明 HUO Qi-Run;LI Jian-Wu;LU Yao;QIN Ming(College of Information Engineering, Capital Normal University, Beijing 100048;School of Computer Science, Beijing Institute of Technology, Beijing 100081;China Electric Power Research Institute, Beijing 100192)

机构地区：[1]首都师范大学信息工程学院,北京100048 [2]北京理工大学计算机学院,北京100081 [3]中国电力科学研究院有限公司,北京100192

出　　处：《自动化学报》2019年第9期1713-1726,共14页Acta Automatica Sinica

基　　金：北京市教委科研计划(KM201810028016)资助~~

摘　　要：有效去除CT图像中环形伪影是医学图像处理领域的一个重要研究方向,现有的方法在去除环形伪影的同时,对CT图像的边缘及细节保留存在困难和挑战.本文采用变分优化的思想,将环形伪影的去除问题建模为一个能量最小化问题,来缓解保持图像信息和去除伪影之间的矛盾,提出了一种后处理的伪影校正算法.根据环形伪影产生机理和特性表现构造有针对性的变分模型,一是从环形伪影的几何特性入手,设计更为合理的梯度保真形式,增强模型对图像细节信息的保护;二是从环形伪影的边缘特性入手,构建具有伪影辨识能力的相对全变分正则项,降低模型对图像结构性信息的影响.基于构造的变分模型,采用高效的优化求解算法,实现环形伪影的有效去除.对比实验表明,无论在视觉观察还是定量分析方面,本文算法均体现出了较好的性能.Effective removal of ring artifacts in CT images is an important research area in the field of medical image processing. The existing methods have some difficulties and challenges in preserving the edges and details of CT images while removing ring artifacts. In this paper, we model the removal of ring artifacts as a problem of energy minimization based on the idea of variational optimization and propose a post-processing method to alleviate the conflict between maintaining image information and removing artifacts. The variational model has been constructed according to the analysis of the mechanism and characteristics of ring artifacts. On the one hand, we design a more suitable gradient fidelity term based on the geometrical characteristics of ring artifacts to enhance the protection on image details. On the other hand, we construct a relative total variational regular term with the ability of artifact identification based on the edge characteristics of ring artifacts, so as to reduce influences on the structural information of images. The designed variational model is to be solved by an efficient optimization algorithm to effectively remove ring artifacts. Experimental results show that the proposed method presents better performance on both visual inspection and quantitative assessment.

关 键 词：CT图像 环形伪影 后处理 变分模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]