EXUDATE DETECTION FROM RETINAL IMAGES USING IMAGE PROCESING TECHNIQUES

One of the popular research areas in image processing is analysis of medical images. Digital images are analyzed to diagnose different medical imaging problems. Diabetic retinopathy is the most common form of diabetic eye disease, which occurs when the level of insulin increases in blood so the damage takes place on retina, which may further lead to blindness. The research is carried out in the in the following manner. In first phase, preprocessing method is developed to segment the retinal images. This method uses the mean and variance based segmentation and ratio of hue and intensity channel is used to remove the noise in retinal images. In second phase, post-processing technique is created to detect the optic disc and exudate in the retinal images. This method uses the Hough transform technique to detect optic disc. Morphological closing is to extract the blood vessels, and applying adaptive contrast enhancement technique to the extracted blood vessels in order to improve the contrast of exudate. OTSU algorithm is used to calculate threshold value to detect exudates. The proposed work applies a Hidden Markov Model (HMM) method for classification of exudate and non exudate region. The proposed system is tested and evaluated on publicly available HOOVER, STARE retinal image database using performance metric such as sensitivity, specificity and accuracy. Through adoption of HMM classifier accuracy was 97.5% within a period of 0.9 seconds.