Research On Localization And Segmentation Of Optic Disc In Fundus Images

Fundus is just like the window to observe the systemic disease.Hypertension,hyperlipidemia,nephritis,diabetes,central nervous system and many other diseases will lead to fundus lesions.Therefore,the disease early screening and detection based on fundus images have obtained more and more attention.Because OD(optic disc)is an important feature in the fundus image,for disease diagnosis with fundus image,how to get the location and segmentation of OD accurately and quickly is very significant.Considering the accuracy of localization and segmentation,the algorithm implemented in this thesis focuses on the computation speed to meet the requirements of real-time system.The specific studies are as follows:(1)In preprocessing part,Otsu threshold is used in the red component of fundus image to generate ROI mask,and bilinear interpolation method is used to adjust ROI length to 540 pixels.(2)In vessel segmentation part,four contrast enhancement algorithms are compared in the fundus image green channel.The CLAHE algorithm is selected for its fast speed and good effect.After comparison between adaptive histogram equalization(AHE)and morphological processing,the latter one is selected.Otsu threshold method is used to obtain blood vessel segmentation result.(3)In OD localization part,based on vessel segmentation result,the half-splitting operation on ROI is raised to reduce the calculation amount.DLC(Directional Local Contrast)is used to take the bright spot area in fundus image as OD candidate.OD candidates with the most vascular branches are chosen as OD area.The gravity center of this area is taken as the center of OD.(4)In OD segmentation part,the morphological and interpolation method are combined to erase the vessels effectively.Canny operator is used to detect the edge of OD,and then Hough transformation is used to detect circle to obtain OD segmentation result.Finally,the tests of MESSIDOR database(1200 fundus images)with different fundus lesions are completed.The test results show that the proposed algorithm has good localization of OD with 99% accuracy and average test time of 0.777 s.Then good segmentation result is obtained based on the correct localization result with average time of 0.295 s.Therefore,it can be said that,compared with other methods,the proposed OD localization and segmentationalgorithm could meet the system requirements of real-time processing for its low complexity,high speed and high accuracy.