视网膜图像配准算法的研究与实现
文献类型：学位论文
| 作者 | 邢雨 |
| 学位类别 | 工学硕士 |
| 答辩日期 | 2010-05-29 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 田捷 |
| 关键词 | 医学图像配准 视网膜图像 多模态 基于局部特征描述 medical image registration retinal image multimodal based on local feature description |
| 其他题名 | Research and Implement of Retinal Image Registration |
| 学位专业 | 计算机应用技术 |
| 中文摘要 | 随着成像技术和计算机科学的发展，医学图像处理已成为现代医学的重要组成部分。医学图像配准是医学图像处理技术的一个重要方面和关键步骤，是图像对比、数据融合、变化分析和目标识别的前提。 很多的视网膜疾病（如青光眼、与年龄相关的黄斑退行性改变等）和很多的普发性非眼科疾病（如糖尿病、脑中风等）都会在视网膜图像中呈现出病理改变。视网膜图像的处理能够使医生更加方便准确地对患者进行诊断和治疗。而视网膜图像的配准是视网膜图像处理与分析中的一个重点和难点。因此，视网膜图像配准对于眼科疾病的治疗具有重大意义。 本文首先简要介绍了医学图像配准的背景和研究意义。医学图像配准方法可分为两种类型：基于灰度的配准算法和基于特征的配准算法，其次，详细分析了基于特征的图像配准算法的原理、方法，并重点介绍了经典的尺度不变特征变换算法。然后，本文以基于特征的图像配准算法为基础，特别针对视网膜眼底图像配准的实际问题，提出一种基于局部特征描述的多模态视网膜图像配准算法。最后，介绍了该算法在医学影像算法平台中的实现。 多模态视网膜图像配准对很多视网膜疾病的诊断和治疗具有十分重大的意义。本文提出的基于局部特征描述的多模态视网膜图像配准方法采用圆环结构划分关键点周围区域，通过局部梯度方向直方图构造特征描述子，并对所提取的特征向量进行规范化。实验证明，该算法在多模态视网膜数据集上提高了配准的成功率，相比于经典算法提高了算法的速度和鲁棒性。 |
| 英文摘要 | With the development of imaging technology and computer science, medical image processing has been an important component of modern medicine. Medical image registration is an important aspect and committed step of medical image processing technology and it is the premise of data comparison, data fusion, variation analysis and target recognition. Many retinal diseases and usual non-retinal diseases reveal pathological changes in retina, such as glaucoma, age-related macular degeneration, diabetes and stroke. The processing of retinal images can help doctors diagnose and cure patients more conveniently and quickly. Retinal image registration is an important and difficult aspect in the processing and analysis of retinal images. Accordingly, retinal image registration has great significance for therapy of ophthalmic diseases. This paper firstly introduces the background and research significance of medical image registration. Medical image registration method can be divided into two kinds: intensity-based and feature-based. Secondly, this paper introduces the feature-based method and the classic scale invariant feature transform algorithm in detail. And then, based on feature-based method, this paper brings forward a method of based on local feature description multimodal retinal image registration for the problem of retinal image registration. In the end, the implementation of this algorithm in MITK platform is introduced. Multimodal retinal image registration has great significance for diagnose and therapy of many retinal diseases. This paper proposes a multimodal retinal image registration method based on local feature descriptor which divided surrounding area of key point by loop structure and constructed normalized feature descriptor by local gradient histogram. Experiments prove that the method enhances the success rate of registration on the dataset of multimodal retinal image. In comparison with classic algorithm, the method increases the rate and robustness of multimodal retinal image registration algorithm. |
| 语种 | 中文 |
| 公开日期 | 2015-09-08 |
| 其他标识符 | 200728017029236 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/7541]  |
| 专题 | 毕业生_硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 邢雨. 视网膜图像配准算法的研究与实现[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2010. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。