基于数据驱动的马尔科夫蒙特卡洛视网膜血管分割

发布时间：2018-06-06 04:36

本文选题：马尔科夫蒙特卡洛 + 数据驱动　；参考：《南京航空航天大学》2011年硕士论文

【摘要】：眼底视网膜血管作为人体非创伤观察的重要器官,其不同程度的变化能够反映出高血压、动脉硬化等心血管疾病的症状,特别当与血管相关的器官发生病变时,眼底血管的直径、曲率等特征的改变在一定程度上可反映病变的程度。故定量和定性地自动分析视网膜血管具有非常重要的临床应用价值,而视网膜血管分割和提取则是分析血管的首要任务。现有多数视网膜血管分割方法尽管对非病变的血管图像具有较好的分割效果,但对病变图像的分割效果仍不理想,尤对光照不均、病灶等敏感。为此,本文提出了一种相对鲁棒的血管分割方法。该方法首次尝试利用计算机视觉中的Top-down和Bottom-up两种层次化分割框架相结合实现视网膜图像的分割。具体而言,首先在彩色视网膜图像的绿色通道上利用Curvelet变换进行血管增强。然后在贝叶斯统计框架下,采用可逆跳转的马尔科夫蒙特卡洛算法搜索参数空间,从而求得不依赖于初始分割的近似全局最优的分割,同时利用数据驱动的均值漂移聚类算法和Canny边缘检测算子来加速马尔科夫链的动态变化。本文在MATLAB环境下,采用标准STARE视网膜图像库中四幅图像进行了实验,实验结果表明该方法不仅对非病变的图像而且对病变的图像都具有较好的鲁棒分割效果,并且通过模式识别中的没有免费午餐定理在理论上对该方法进行了理论分析。
[Abstract]:The retinal vessels in the fundus of the eye, as an important organ for non-traumatic observation, can reflect the symptoms of cardiovascular diseases such as hypertension, arteriosclerosis and so on, especially when the organs associated with blood vessels are changed. Changes in the diameter and curvature of the fundus vessels can reflect the extent of the lesion to a certain extent. Therefore, quantitative and qualitative automatic analysis of retinal vessels has very important clinical application value, and retinal blood vessel segmentation and extraction is the primary task of vascular analysis. Although most of the existing retinal vascular segmentation methods have good segmentation effect on non-pathological vascular images, the segmentation effect of the diseased images is still not ideal, especially sensitive to uneven illumination and focus. Therefore, a relatively robust blood vessel segmentation method is proposed in this paper. This method is the first attempt to realize retinal image segmentation by combining Top-down and Bottom-up in computer vision. Firstly, the Curvelet transform is used to enhance the blood vessels in the green channel of the color retinal image. Then under the Bayesian statistical framework, the reversible jump Markov Monte Carlo algorithm is used to search the parameter space, and the approximate global optimal segmentation independent of the initial segmentation is obtained. At the same time, the data-driven mean shift clustering algorithm and the Canny edge detection operator are used to accelerate the dynamic change of Markov chain. In this paper, four images in the standard STARE retinal image library are used in MATLAB environment. The experimental results show that the proposed method has a good robust segmentation effect not only for non-pathological images but also for diseased images. The method is theoretically analyzed by the free lunch theorem in pattern recognition.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：R774.1;R318.0;TP391.41

【参考文献】