Diabetic retinopathy (DR) is a microvascular problem of long-term diabetes which is the primaryroot of visual lossdue to theabnormalities in blood vessels of the retina. The detection of DR in early stages is the most important to prevent visual impairment. In this paper, a Fuzzy Naive Bayesian (FNB) classifier is proposed to classify the four types of retinal abnormalities such as Hordeolum, Seborrheic keratosis, Xanthelasma and squamous Cell Carcinoma. The proposed method includes four main stages to detect DR. At first, Discrete Wavelet Transform (DWT) based Bayes Shrink thresholding method is applied to the retinal images for reducing noise and improving the quality of the image. In the second stage, a hybrid retinal image segmentation algorithm is used which is the integration of active contour with Fuzzy C-Means (FCM) algorithm. In the third stage, texture based features are extracted from the segmented image using Gray-Level Co-occurrence Matrix (GLCM). Finally, based on FNB model four abnormities are detected.The classification of the proposed FNB classifier achieves an accuracy of 98.7%, sensitivity of 98.2%, and specificity of 96% which is better than the other existing techniques.
Keywords : Diabetic Retinopathy (DR),Fuzzy Naive Bayesian (FNB) classifier, Discrete Wavelet Transform (DWT),Fuzzy C-Means (FCM), Gray-Level Co-occurrence Matrix (GLCM).