Detection Of Diabetic Retinopathy

In modern people's life,diabetes mellitus,as a kind of metabolic disease already more and more get the attention of the people,among them,retinopathy is one of the major complications of diabetes,and early on a regular basis for a person with diabetes retina pathological change detection,helps the doctor early detection,against the disease with early treatment,avoid blindness status,but because of the complexity of the process of diagnosis,and the doctor manpower is limited,can't quickly on the diagnosis to patients with retinal conditions,and exist in the process of artificial diagnosis misdiagnosis due to the deviation of subjective consciousness,so you need to establish a method of objective,to achieve rapid diagnosis of diabetic retinal conditions.Diabetes mainly affects the blood vessels in the retina of the eye to the whole retina pathological change,with the change of the retinal vascular blood sugar content,viscosity,blood vessels,curvature of radius,and the bifurcation angles will change,and with the increase of the degree of retinopathy,constantly appear in the retina bleeding spots,hard exudate and microaneurysm some shapes and colors are different from normal secretion of the retinal image,the image is based on the pathological changes,various features change in a person's blood vessels,as well as a variety of secretion of retinal image color change,so the retinal image texture feature extracting and the color of the retinal image features as the main characteristics of diabetic retinopathy recognition.In this dissertation,through a series of pretreatment technology makes the contrast of retinal blood vessels in the image and the background has been greatly improved,the mask processing technology background region that is independent of the removal of the original image using histogram equalization to enhance the image,and through the median filtering for image denoising,image after pretreatment by optimal threshold segmentation method,maximum entropy method,the global threshold iteration method,image threshold segmentation of image and image after image segmentation,can clearly show the distribution of blood vessels.This dissertation on retinal image recognition based on support vector machine,the selection of features for image texture features and color features of fusion after comprehensive characteristic,Using the gray level co-occurrence matrix of image texture feature is extracted,in order to overcome the traditional support vector machine(SVM)recognition rate is low due to improper parameter selection problem,this paper uses the particle swarm optimization(PSO)algorithm for penalty coefficient and the gaussian kernel function of SVM nuclear parameter optimization,improve the recognition rate of the algorithm and the lesion detection,practical strong,in the experimental stage,respectively for three different to the retinal image after image segmentation method for feature extraction,and then use the optimized SVM is used to detect the pathological changes of image segmentation method to choose the recognition rate is higher.