Retinal Blood Vessel Extraction Method And System Based On Dynamic Scale Allocation

With the rapid development of medical image processing technology,medical staff can use the computer automatic processing system to diagnose and treat the human eye diseases as early as possible,it can effectively avoid further deterioration of the disease.Among the numerous capillaries in the human body,the retinal blood vessels are the only ones that can be obtained in non-traumatic form.Some of the early vascular disease may change its morphology,so the retinal blood vessels can be used as an indicator of disease diagnosis.Therefore,it has important application value to establish a rapid,accurate and reliable automatic extraction system of retinal blood vessels in the field of medical research.However,due to the influence of illumination and noise in the process of imaging and transmission,the image quality is reduced,and the analysis is very difficult.Therefore,it is the bottleneck to restrict the development of the existing algorithms that how to extract the blood vessels with complex width in the extremely low contrast retinal images.In this paper,according to the gray level information of blood vessels,a new algorithm for retinal blood vessel extraction based on dynamic scales allocation is proposed.The main tasks are described as following:1.A dynamic scales allocation scheme based on the gray level information of image is proposed.The existing blood vessel extraction scheme based on multi-scale matched filter using uniform scale throughout the retinal image.For this kind of blood vessel extraction algorithm,there are some local regions of the image scale distribution is unreasonable.In this paper,we proposed an algorithm of retinal blood vessel extraction based on dynamic scales allocation,the above problems can be overcome effectively.Because of the contrast of the blood vessels and background is more uniform in local image regions,the image is divided into equal size sub images firstly.Secondly,the gray distribution of each sub image is analyzed,to determine the type of blood vessels,the filter with matching scale is allocated for processing.2.A dynamic threshold processing scheme is proposed,which to adjust thresholds based on the gray histogram of the image.In this part,we are still the histogram of each sub image is analyzed,judged the contrast level and the presence of non-vascular structures,and according to this result,we use different threshold to segment the blood vessels.Finally,the sub images of the segmentation are merged to obtain the complete retinal vascular network.Compared with the previous algorithms based on matched filter,the proposed method detects many fine vessels drowned by noise and avoids an over-estimated of the thin vessel.The test result of all illustrations of DRIVE database test set shows that,the algorithm can automatically extract the retinal blood vessel structure with the highest accuracy of 0.7526 and the average segmentation accuracy of 0.9393.