基于Corvis ST医学影像的角膜轮廓提取算法

目前已有的轮廓提取算法在提取可视化角膜生物力学分析仪(corneal visualization scheimpflug technology, Corvis ST)影像的角膜轮廓中,由于角膜边缘的局部图像灰度分布相近这一特点,提取出的角膜轮廓普遍不完整或者提取的角膜轮廓边缘出现细小突出。这会使得角膜轮廓的完整性遭到破坏,提取到的角膜轮廓与实际的角膜原始图像严重不符。本文针对Corvis ST采集的角膜图像的轮廓提取问题,基于最大类间方差法(OTSU)算法设计一种高效的图像处理方法。首先,将角膜图像进行除杂、灰度化以及图像降噪等处理,达到减少图像计算量和降低数字图像噪声干扰的目的;其次,基于OTSU算法对图像进行分割,并在此基础上加入数学形态学运算,达到平滑图像边界和填充细小“孔洞”的目的;最后,采用Canny边缘检测算法提取图像中的角膜轮廓,达成提取出高完整性角膜轮廓的目的。在相同的图像数据集上,与最新的纽扣轮廓瑕疵检测系统中设计的轮廓提取算法(B-OTSU algorithm)进行了对比实验。实验结果表明,从轮廓完整性以及准确性的角度,应用本文方法提取的角膜轮廓明显优于最新的纽扣轮...

Corneal contour extraction algorithm based on Corvis ST medical image

The currently existing contour extraction algorithm extracting the corneal contour of corneal visualization scheimpflug technology (Corvis ST) image,due to the close gray distribution of the cornea edge local image,the corneal contour extracted generally is incomplete or corneal contour edges extracted appear small extenuation. This will destroy the integrity of the corneal contour,and the extracted corneal contour is seriously inconsistent with the actual original corneal image. In this paper,a high-efficient image processing method based on OTSU algorithm is designed for contour extraction of corneal images collected by the Corvis ST.Firstly,the corneal image is processed by impurity removal, gray scale and image denoising to reduce the amount of image computation and digital image noise interference.Secondly,the image is segmented based on OTSU algorithm,and mathematical morphology operation is added on the basis to smooth the image boundary and fill small "holes".Finally,Canny edge detection algorithm is used to extract corneal contour in the image,achieving the purpose of extracting high integrity corneal contour. On the same image data set,the contour extraction algorithm (B-OTSU algorithm) designed in the latest button contour defect detection system is compared with this algorithm in the experiment.The experimental results show that from the perspective of contour integrity and accuracy,the corneal contours extracted by the proposed method are obviously superior to the corneal contours extracted by the B-OTSU algorithm in the button contour defect detection system.