摘要
利用不同波长的视网膜图像可以对视网膜血管血氧饱和度进行计算,但需进行配准处理.提出一种基于视网膜图像血管分割的互信息图像配准方法.为了充分利用血管的轮廓信息和灰度信息,提高配准精度,首先对配准图像进行图像分割,提取视网膜图像中的血管轮廓信息;然后对分割后图像中的血管进行相似度计算,并采用Powell优化算法中的黄金分割法一维搜索算法来提升运算速度;最后根据计算的相似度值来完成不同波长图像的配准.研究中算法配准获得变换参数(角度、X方向、Y方向)的误差的均值分别为2.00%、2.53%和2.52%,误差的方差分别为0.57、2.09和0.34,均优于直接互信息配准方法.实验表明:算法可以自动、有效地对不同波长的视网膜血管图像进行配准,并具有良好的可重复性和稳定性.
In order to calculate the blood oxygen saturation of retinal images, different wavelength images should be registered. This paper presents an image registration method based on the segmentation of blood vessd image and mutual information. In the study, in order to reduce the impact of the information on the registration result, the registration of image segmentation, extraction of retinal vessels information in the image; calculating the vascular similarity in the segmented image, and using the Powell optimization algo- rithm 0.618 one-dimensional search algorithm to improve the speed of operation; the different wavelength of the image registration based on the calculated similarity value. In the study, the error average of parameters （ angle, X direction, Y direction） calculated from the registration algorithm is 2.00%, 2.53% and 2.52%, and the variance of the error is 0.57, 2.09 and 0.34, were better than the direct mutual information registration method. Experiments show that the algorithm can automatically and effectively register retinal images with different wavelengths, and has good repeatability and stability.
出处
《四川师范大学学报（自然科学版）》
CAS
北大核心
2017年第4期554-560,共7页
Journal of Sichuan Normal University（Natural Science）
基金
国家自然科学基金(81301286)
四川省科技支撑项目(2014GZ0005)
关键词
视网膜图像
图像分割
配准
互信息
retinal image
image segmentation
registration
mutual information
