Diabetic Retinopathy is the major cause of blindness in many diabetic patients. Automatic detection of exudates in retinal images can assist in early screening of Diabetic Retinopathy. Several techniques can achieve good performance on a good quality retinal images. But when the image is of low quality, we need a new method. In this paper, we presented a novel method for the detection of exudates in low quality retinal images. The colour retinal images are pre-processed by a hyperbolic median filter and then segmented using fuzzy c-means clustering algorithm. After segmenting the images, a set of features based on colour, size and texture are extracted. Then these features are optimized using Particle Swarm Optimization (PSO) technique. Finally the features are classified using a recursive Support Vector Machine (SVM) Classifier. The proposed method achieves an accuracy of 98% and predictivity of 98.5% for the identification of exudates.