中文题名: | 针对视网膜血管图像的超分辨率重建及保持连通性分割 |
姓名: | |
保密级别: | 公开 |
论文语种: | 中文 |
学科代码: | 080901 |
学科专业: | |
学生类型: | 学士 |
学位: | 理学学士 |
学位年度: | 2021 |
学校: | 北京师范大学 |
校区: | |
学院: | |
第一导师姓名: | |
第一导师单位: | |
提交日期: | 2021-06-16 |
答辩日期: | 2021-05-11 |
外文题名: | Super Resolution Reconstruction and Segmentation with Connectivity Guarantee for Retinal Blood Vessel Images |
中文关键词: | |
外文关键词: | SRGAN ; U-net ; super resolution image reconstruction ; image segmentation ; pixel connectivity |
中文摘要: |
超分辨率生成对抗网络(SRGAN)是一种通过对抗训练的方式来生成高分辨率图像的网络模型,在图像重建领域中有非常多的应用。然而,SRGAN虽然可以提高图像的分辨率,但它并不能有效地恢复出连通的像素路径。近年来,血管像素连通性在医学图像重建与分割领域得到了越来越多的关注。为了提高图像分辨率的同时保持血管的单连通特性,本文提出了一种基于贪心算法的保持连通性后处理算法,命名为AutoConnet算法。本文首先使用SRGAN对DRIVE视网膜数据集进行超分辨率图像重建,经U-net分割后,再利用AutoConnet算法对非连通路径进行处理。实验表明, AutoConnet算法作用后的结果图像具有更高的召回率,使SRGAN生成的高分辨率图像更加符合真实的人体视网膜结构,能够帮助医师进行更准确的诊断,具有较大的实践意义。 |
外文摘要: |
Super resolution generative adversarial network (SRGAN) is a neural network that generates high-resolution(HR) image through adversarial training. It has many applications in the field of super-resolution image reconstruction. However, although SRGAN can improve the image resolution, it can not effectively restore the connected pixel path. In recent years, blood vessel pixel connectivity has attracted more and more attention in the field of medical image reconstruction and segmentation. In order to improve the image resolution and maintain the single connectivity of blood vessels, this thesis proposes a connectivity preserving post-processing algorithm based on greedy algorithm, named AutoConnet algorithm. In the experiment, firstly, use SRGAN to improve the resolution of the DRIVE data set. After U-net, use AutoConnet algorithm to process the unconnected path. The experimental results show that the result images have higher recall rate, and the HR image generated by SRGAN is more consistent with the real retinal structure, which can help doctors to diagnose more accurately.
|
参考文献总数: | 20 |
插图总数: | 9 |
插表总数: | 2 |
馆藏号: | 本080901/21024 |
开放日期: | 2022-06-16 |