388 DETECTION OF EXUDATES AND FEATURE EXTRACTION OF RETINAL IMAGES USING FUZZY CLUSTERING METHOD P.Hosanna Princye 1 and V.Vijayakumari 2 1 Department of EIE, Adhiyamaan college of Engineering Hosur, Tamilnadu, India 2 Department of ECE Sri Krishna College of Technology, Coimbatore, Tamilnadu, India 1 hprincye@gmail.com and 2 ebinviji@rediffmail.com Abstract: Diabetic retinopathy (DR) is the micro vascu- lar changes that cause detectable changes in the optic disc. This paper aims at the detection of retinal exu- dates and other features such as blood vessels and optic disc from fundus image. The two methods are imple- mented for the detection of exudates they are morpho- logical method and FCM clustering method. Contrast limited adaptive histogram equalization (CLACHE) is used to extract the green component in the image. In blood vessel extraction, blood vessels are extracted by top hat transformation followed by connected compo- nent analysis. The optic disc centre is found using Cir- cular Hough Transform (CHT) and propagation through radii method is employed and the entire optic disc re- gion is blackened and removed. Exudates detection is the important characteristics of diabetic retinopathy and its varies depends upon the severity of the DR. The FCM method used to detect exudates. The overall sensi- tivity, specificity and accuracy are calculated and 98% accuracy obtained Keywords: Retinopathy, Exudates, Optic disc, Blood vessels, Hough transform, histogram. INTRODUCTION Diabetic Retinopathy (DR) is a lingering disease which eventually leads to blindness. Diabetic Retinopathy is damage the retina caused by diabetics mellitus. Contin- uous screening is necessary to prevent blindness. During the screening colour images are obtained by fundus camera and are required for manual analysis and diag- nose the diabetic retinopathy. Sometimes fundus images are not clear to see because of abnormality of eyes, non- illumination and noise in a fundus image. It is suitable for automatic screening system. In an automatic system the normal features like optic disc, blood vessel and exudates in the retinal images are automatically detect- ed. Depending upon the stages of the disease the effect of diabetic retinopathy on vision varies. There are two types of retinopathy non-proliferative diabetic retinopa- thy and proliferative diabetic retinopathy. Non- proliferative diabetic retinopathy consists of cotton- wool spots, intraregional haemorrhages, hard exudates, micro aneurysms. Proliferative diabetic retinopathy causes visual impairment where there may be sudden haemorrhage from the unstable new vessels resulting in total or partial visual loss or from preretinal haemor- rhages. Many techniques have been previously em- ployed in this work. In the work of Sophrak et.al the method uses feature selection and exudates classifica- tion using naive Bayes and support vector machine (SVM) classifiers [3]. Morphological methods using watershed transformation has been studied by Thomas Walter in order to localize the optic disc [11]. The coun- ters of the optic disc were detected using Watershed transformation. S.Kavita et.al proposed a method which uses an automatic detection of diabetic retinopathy exu- dates in color fundus retinal images. Candidate exudates employing a multi-scale morphological process were identified by Fleming A.D. et al. [7]. Based on local properties, the probability of a candidate being a mem- ber of classes’ exudates, drusen or background was es- timated. Ahmed Wasif Reza et al. have presented an approach to automatically segment the Optic Disc and exudates [1]. Akara Sopharak et. al have proposed an automatic method to detect exudates from low-contrast digital images of retinopathy patients with non-dilated pupils by a Fuzzy C-Means (FCM) clustering [3]. An effective framework to automatically segment hard exu- dates (HEs) in fundus images was proposed by Guoli- ang et.al based on a coarse to fine strategy, as a coarse result is obtained frost allowed of some negative sam- ples, then eliminate the negative samples step by step . A new method for the detection of exudates using adap- tive thresholding and classification is proposed by Hussain et.al in which the retinal structures are used to remove artefacts from exudates detection results [8]. The location of the optic disc is an important issue in retinal image analysis as it is a significant landmark feature, and its diameter is usually used as a reference length for measuring distances and sizes. A deformable contour model (or Snake) with gradient vector flow (GVF) (Viranee et.al, 2009) can be used as an external force for optic disc detection using segmentation [12]. S. Sekhar et.al used the morphological characters of an image. Morphology is used to locate the brightest region within the image and a Hough Transform is used to de- tect circular features within the gradient image of the resulting region of interest [10]. A geometrical paramet- ric model was proposed by Foracchia et.al to describe the general direction of retinal vessels at any given posi- tion in the image. In this two of the model parameters are the coordinates of the OD centre [9]. Blood vessel