| Popis: |
Diabetic retinopathy is the main cause of vision loss among working-age adults around the world. Early detection and timely treatment are essential to reduce impairment and prevent blindness. In order to enable people to easily check their own conditions, we develop a decision support system for detecting diabetic retinopathy in early stage. The system extracts two important features, blood vessels and hard exudates, by using a combination of image processing techniques including black top hat, morphological opening and closing, canny edge detection, blob function, convert label matrix, convex-Hull method and color thresholding. In addition, an adaptive algorithm is proposed to cope with the variety of illuminations in different images. These extracted features are essential for estimating the risk of developing diabetic retinopathy. The system has been tested on 1379 images from DIARETDB0, DIARETDB1, and MESSIDOR databases. A qualified optometrist verified that our system achieved 92% and 90.3% accuracy for blood vessel and hard exudates extraction respectively. The system takes only 0.6 seconds to analyze one image, and therefore it is suitable for real-time implementation in mobile devices. |
