| Human retinal vascular morphology has a very close connection with a variety of diseases,and it provides the base of early diagnosis and prevention for diseases of the eye disease and systemic system.The traditional artificial screening is not only workload-needed and time-consuming,but low accuracy-acquired,and even it is likely to cause misdiagnosis.With the development of digital image processing technology in the field of medical image as well as the computer is widely used,retinal vascular structure of the automatic detection has become a trend.Now researchers have proposed many automatic segmentation algorithm of retinal blood vessels,and made some achievements,but the accuracy of blood vessels segmentation and the impact of the retina itself organization structure remains to be further improved.Based on this problem,this paper puts forward two new segmentation algorithms of retinal blood vessels and a large number of experiments have been made in the library of STARE and DRIVE,the Last the experiment results have been compared with the segmentation algorithm.The main work is as follows:1)According to the low contrast with background pixels of the small blood vessels,the segmentation accuracy is low,which brings the problems such as the loss of the small blood vessels and vascular rupture,this paper proposes a combination of local and global enhancement based on multi-scale of retinal blood vessels segmentation algorithm.First,multi-scale linear detector is divided into two parts of small scale and large scale;Second,the image of small scale could be dealt with the local enhancement and the image of large scale could be dealt with the global enhancement;Third,the response function of the small scales,and the response function of the large scales could be fused with its corresponding enhance image respectively;Finally it could obtain the fusion image for subsequent processing with removing the start-point and isolated points getting the final segmentation images of retinal blood vessels.Theoretical analysis and experimental results show that the algorithm of segmentation could have more small blood vessels.In the image library of STARE and DRIVE,its accuracy is 0.9662 and 0.9645 respectively,which is higher than that of the traditional algorithm of segmentation.2)The structure of optic disk is similar with blood vessels for retinal blood vessels,which causes the segmentation problem,this paper proposes a segmentation algorithm based on the Zernike moment coefficient of different template of retinal blood vessels.First,calculate the Zernike moment with four different moment coefficient of the template and study the influence of coefficient of different templates for different width of the blood vessels;Second,the four template above convolution result could be dealt with multi-scale line detection,obtaining the different response function of the different template;Third,theresponse function could be fused with moment image respectively;Finally it could obtain the fusion image for subsequent processing with removing the start-point and isolated points getting the final segmentation images of retinal blood vessels.Theoretical analysis and experimental results show that the accuracy of the algorithm in the image library of STARE and DRIVE is 0.9659 and 0.9643 respectively,and effectively for the optic disk to deal with. |
