基于参数估计的多层M-S模型杯盘分割方法

摘要

The hierarchical Mumford-Shah model can segment the optic cup and disk of optic nerve head images very well gotten from color fundus imaging. But because of poor illumination and low contrast, the images have many unwanted noise, thus segmentation of images based on the hierarchical Mumford-Shah model needs different length parameters. To solve this problem, an adaptive model based on the parameter estimation is presented. Vascular occlusion will also make cup and disk edge cave and part. Using a method combined prior knowledge of the constraint method, the effective edge curve is extracted, then, the complete edge of optic cup and disk is gotten by using curve fitting. The experimental results show that this method can effectively solve the above problem of segmentation of optic cup and disk of optic nerve head images.%多层Mumford-Shah模型能够很好地分割通过彩色眼底成像得到的视乳头杯盘图像,但由于光照不均、对比度小等因素,图像中存在很多无用的噪声点,对于不同的图像,多层Mumford-Shah模型需要不同的长度参数,为了解决这个问题,给出了一个自适应的长度参数估计模型.另外,血管遮挡也会使得杯盘边缘曲线的凹陷和断裂,从而造成图像分割困难,采用结合了先验知识的约束方法来提取有效的边缘曲线,并通过曲线拟合来得到完整的杯盘边缘.实验证明,该方法能有效地解决上述视乳头杯盘分割问题.