结合改进型GBVS模型和眼底血管结构特征的视盘检测方法研究

发布时间：2018-02-15 03:16

本文关键词： 眼底图像视觉特征血管辅助 Gbvs C-V模型　出处：《天津工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：彩色眼底图像的处理与分析不仅可以用于诊断某些眼科疾病,还可以帮助医生进行糖尿病、高血压等全身性疾病的诊断,跟踪病情的发展。在眼底视网膜图像中,视盘是一个类圆形的近黄色或白色的亮斑,同时是眼底血管的发源地,汇聚着大量较粗的血管,其形状、大小和深度等参数是衡量眼底健康状况的重要指标。准确的视盘检测不仅可以辅助定位血管、黄斑等重要的眼底组织结构,还可辅助确定渗出物、微动脉瘤等病变的位置,对眼底图像分析具有重要的意义。本文方法充分利用视盘的亮度、对比度、相位一致性三类视觉特征以及主血管结构特性,提出一种结合改进型基于图的显著性模型(Gbvs)和眼底血管结构特征的视盘检测方法。该方法首先对Gbvs模型进行改进,将其中的颜色、亮度、方向三类特征改为亮度、对比度、结合相位一致性(PC)三类特征,并利用改进后的Gbvs模型构造眼底图像的显著图;然后提静脉血管轮廓线,进行抛物线拟合,通过比较抛物线顶点邻域内的显著性与整幅眼底图像的平均显著性的大小确定视盘位置;最后,消除视盘局部区域的血管,利用C-V水平集方法来确定视盘边界,获得视盘分割结果。在四个公开的眼底图像数据集(DRIVE、MESSIDOR、STARE和DIABETEDO)上对该方法进行了测试,平均定位准确率分别为100%、99.25%、90.12%、96.1%,高于现有代表性方法。
[Abstract]:Color fundus image processing and analysis can not only be used to diagnose some eye diseases, but also to help doctors to diagnose diabetes, hypertension and other systemic diseases, tracking the development of the disease. The optic disc is a round, nearly yellow or white bright spot, and is the origin of the fundus vessels, which gather a large number of thicker blood vessels, the shape of which, The parameters such as size and depth are important indexes to measure the fundus health. Accurate optical disk detection can not only assist in locating the blood vessels, macula and other important ocular fundus tissue structure, but also help to determine the location of exudates, microaneurysms and other lesions. This method makes full use of three visual features, such as brightness, contrast, phase consistency, and main vascular structure. In this paper, a new method of visual disk detection based on improved graph-based saliency model (Gbvs) and fundus vascular features is proposed. Firstly, the Gbvs model is improved to change the color, brightness and directional features to luminance and contrast. Combining with three kinds of features, the improved Gbvs model was used to construct the salient image of the fundus, and then the contour of the levator vein was fitted with parabola. The position of the disc is determined by comparing the significance in the parabola vertex neighborhood with the average significance of the whole fundus image. Finally, the blood vessels in the local area of the disc are eliminated, and the boundary of the disc is determined by using the C-V level set method. The results of visual disk segmentation were obtained. The method was tested on four open fundus image data sets, namely, DRIVE / MESSIDORSTARE and DIABETEDO. the average accuracy of the method was 100 and 99.250.12.1, which was higher than that of the existing representative methods.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R770.4;TP391.41

【相似文献】