【作者】 游嘉；

【导师】 何中市；

【作者基本信息】 重庆大学 ， 计算机软件与理论， 2011， 硕士

【摘要】 眼底图像融合是将对取自不同时间、不同传感器或不同视角的关于眼底图像或者图像序列加以综合的过程。由于眼底图像在眼科是一个客观、标准的诊断方法,图像融合技术在眼底图像中的应用可提供更强的信息解译能力,对分析和记录各种眼睛疾病及其发展有非常重要的作用。图像融合过程可分为三个阶段:图像预处理、图像配准与图像融合,各阶段的处理算法很多,目前还没有适用于所有图像的方法,往往需要根据图像本身特征,在处理过程中尽量找到在准确性、速度和鲁棒性上平衡的方案。基于此,本文采用层层递进的结构,在预处理阶段着重考虑了眼底血管的增强问题,提出了基于Hessian矩阵的血管增强方案;考虑到优化算法对图像配准有重要的影响,在配准阶段,本文提出了基于互信息和PSO-NMS的图像配准方案;在融合部分,主要关注基于小波的融合方法,提出了一种基于空间频率和区域最大值的小波变换融合方法,同时整合了眼底图像融合的全过程方案,并将该方案应用到眼底图像融合系统的实现中,以此证明本文方案的实用性。本文具体工作如下:①阐述了图像融合的定义、理论、方法及流程,分别从图像融合处理阶段、配准阶段、融合阶段阐明其中的关键步骤对融合结果的影响,并介绍了当前主要的图像增强、配准及融合方法。②分别在融合的三个阶段提出了针对眼底图像的解决方案。在预处理阶段,为保留眼底图像中血管结构信息,同时降低图像配准的计算复杂度,本文在介绍了几种常见的眼底图像增强方法基础上,提出了一种基于Hessian矩阵的血管增强方法,加上图像校正和平滑,形成眼底图像融合的预处理方案。在配准阶段,重点介绍了基于互信息的相似性测度,以及用于求解复杂问题的粒子群优化算法及其改进,在此基础上,本文应用了一种单纯形法结合粒子群算法的混合优化算法来求解变换参数。结合配准框架,形成了系统的配准方案。在融合阶段,重点介绍了基于小波的图像融合方法,并详细叙述了小波分解和重构过程,在此基础上,本文提出了一种基于空间频率和区域最大值的小波融合方案。③综合预处理、图像配准与融合三个阶段的处理方案,提出了一套眼底图像融合的整体方案。在此基础上形成了眼底图像融合的基本系统架构及功能模块,并根据本文所设计的图像融合方案和步骤,开发了一个基于Visual C++的眼底图像融合系统。更多还原

【Abstract】 Fundus image fusion is a process of integrating the fundus images taken by different sensors or perspectives from the same patient. Since the fundus image is an objective and standard diagnostic tool in ophthalmology, the application of image fusion technology in the fundus image can provide better capacity of information interpretation, which takes a very important role in analyzing and recording of all kinds of eye diseases and their progression.The process of image fusion can generally be summarized as three phases: image pre-processing, image registration and image fusion. There are various algorithms for each stage, however, the method of best applicability has not been found yet. It usually needs to consider about the characteristics of the image itself and try to find the best solution on the balance of accuracy, robustness and low time consuming. Therefore, this thesis is written with a progressive structure: In the phase of pre-processing, we focus on the retinal blood vessel enhancement, and a new approach to vascular enhancement based on Hessian matrix is proposed. In the registration phase, this thesis proposes an image registration scheme based on mutual information and the PSO-NMS optimization algorithm. As for the fusion phase, we focus on the wavelet-based image fusion method, and describe the wavelet decomposition and reconstruction process of image in detail, and furthermore, we propose a wavelet-based image fusion scheme using spatial frequency and regional maximum. The proposed schemes are applied in the implementation of a system for fundus image fusion, which would prove the practicality of our proposed schemes.Specific work of this thesis is described as follows:①Introduce the definition, theory and methods of image fusion, and explain several essential steps that illustrate results of fusion. Discuss the current methods for image enhancement, registration and fusion.②Solution schemes of fundus image fusion have been proposed respectively in the three phases of image fusion. At the pre-processing phase, in order to preserve the structure information of retinal blood vessels, while reducing the computational complexity of the image, this thesis proposes an approach to vascular enhancement based on Hessian matrix after introduce several common image enhancement methods. Adding to image smoothing and correction, the pre-processing of fundus image is presented. In the registration phase, after introduce similarity measure based on mutual information and improved particle swarm optimization algorithm for complex problems solving, a hybrid PSO combined with simplex method is used to solve the transformation parameters, and furthermore, a registration scheme has been formed based on the registration framework. The fusion phase of this thesis is focusing on the wavelet-based image fusion methods and a wavelet-based fusion scheme using spatial frequency and regional maximum for fundus image is presented after detailed descriptions of the wavelet decomposition and reconstruction is introduced.③Composing the three solution schemes of preprocessing, image registration and fusion, we present an overall scheme for fundus image fusion, which forms the basic system structure and function modules of a system. Therefore, we developed a fundus image fusion system according to the proposed schemes.更多还原