AN EFFICIENT DETECTION AND CLASSIFICATION OF DIABETIC RETINAL FUNDUS IMAGES USING FEATURE EXTRACTION

Diabetic Retinopathy (DR) is caused by high sugar level diabetes is an eye disorder. The DR is detected by analyzing retinal fundus images at early stage and treating the diseases reduces the risk of vision. The present methods for DR detection takes more amount of time for detecting and preventing. In this paper, the proposed hybrid classification method is effective for image processing in detecting DR from fundus images. The proposed method includes DR image preprocessing, DR image blood vessel segmentation and removal, DR feature extraction and selection and DR image classification steps. These methods are effective in detecting and preventing the disease very early stages with minimum amount of time and also increases the detection performances. Diabetic retinopathy is complicated disease between diabetic patience's. Many diabetic patients are affected with various disease and which is more complicate to detect the types of disease. The DR is categorized by severity of lesions and lesions produces sequence of changes in the hard exudate, micro aneurysms and also soft exudate. The simulation is performed using MATLAB for MESSIDOR and DIARETDB1 datasets and results are validated with different parameters. MATLAB tool is more suitable for images processing and datasets are well defined which is extracted from the reliable sources. In this experiments, two algorithms are used to detect and classify the fundus images such as Support Vector Machines (SVM) and Fuzzy C-Means Clustering (FCM). The result for the experiment is compared with SVM and FCM which shows greater efficiency and effectiveness. The proposed method DR sensitivity is 98%, DR specificity is 92% and DR accuracy 89.1%.