The Research And Implementation Of Optic Disc Segmentation Method In Fundus Image

Optic disc is one of the most important anatomical features of fundus images.The detection of optic disc is not only conducive to the automatic positioning of ocular fundus,such as blood vessels and macula,but also helps to locate the affected areas such as exudates and microvascular aneurysms.Aiming at the problem that the lesions in the diabetic retinopathy affect the localization of the optic disc,based on current domestic and foreign research status and the defects of the mainstream algorithms in the localization and segmentation of the optic disc,this thesis summarizes a set of optic disc based on visual saliency detection algorithm framework.The algorithm uses visual saliency detection technology to extract candidate region of optic disc.Afterwards,the region of optic disc is determined by analyzing the vessel structural features of the fundus image.Finally,the optic disc contour is segmented.In the realization of each step,this thesis has done the following work:In the extraction stage of the candidate region of optic disc,this thesis presents a visual saliency detection method based on graph theory.In this method,multiscale down-sampling is performed on the fundus image to establish three characteristic pyramid models based on color,brightness and texture direction.The Markov matrix is constructed in each of the feature pyramid and the feature map is calculated.Then,the feature maps are fused to obtain the saliency probability graph of the candidate region of the optic disc.Finally,the final optic disc candidate region is obtained based on ROI automatic detector.In the candidate region of the optic disc screening stage,this thesis mainly based on morphological hat transform blood vessels segmentation.The original image is reduced dimension reconstruction by PCA method,and then based on the background of the CLAHE elimination algorithm and the Otsu threshold segmentation algorithm are used to obtain the vessel segmentation images.After that,the final region of the optic disc is determined by counting the number of vertical blood vessels in each candidate region.In the stage of accurate segmentation of the optic disc,we combine the Hough transform and the Grow-cut method to preliminarily divide the optic disc contour and use it as the initial contour line,and then use GVF model and balloon force model respectively to further study the optic disc contour calibration.Comparative experiments show that the segmentation method based on Grow-cut and GVF models is better.The method of the present invention extracts the optic disc area with high accuracy and has a stable effect on fundus images with complicated lesions.