Diabetic retinopathy is one of the most epidemedical diseases in the humanvisual system.The cause of this diasease is the increaseofglucose level rate in the blood. In this thesiss weintroduce a novel technique to estimate blood flow velocity in main retinal arteries by using florescent angiography video clips.For preprocessingvideo frames, we apply Gabor filter to enhance contrast of them then we usestatistically local thresholding (SLT) and support vector machine (SVM) image processing approaches to extract retinal vessels. It was necessary to preprocess videos to eliminate sensitivity changes applied by imager operator. The next step was processing the intensity vs. time curves of retinal vessel points in order to obtain the cross time (CR) of dye. We processed curves to find corresponding points (The points where all curves show the same behavior). Knowing CR between points, the next step was to compute on-vasculature distance (OVD) between pixels and after that blood flow velocity was estimated in vessels by CR and OVD. Finally estimated speed was evaluated with blood flow simulation that was made with MATLAB software. .We got 94% accuracy, forSLT based method and 91% accuracyfor SVM vessel extraction method. Our results showed that none proliferative diabetic retinopathy patients had a range of 5.23 to23.63 , moderate diabetic retinopathy patients had a range of 1.09 to 8.44 , proliferative diabetic retinopathy patient had 1.32 and healthy data had 4.77 mm/sec blood flow velocity. . Keywords: Diabetic Retinopathy,Fundus Fluorescent Angiography,Statistically Local Thresholding, Support Vector Machine (SVM) method, Intensity vs. Time Curves.
