Diabetic retinopathy is a serious disease diabetic eye disease as a leading cause of blindness. In this paper, we propose an automated algorithm of detecting exudates in diabetic retinopathy image. The algorithm merges regions using
nearest neighbor graph, and identifies exudates according to color information and pixel locations. The algorithm is quite fast but achieves an average specificity of 95.42% with the average sensitivity of 91.08% in detection of the exudates in a test case with 8 diabetic retinopathy images. This method can be implemented as a prototype for automatic exudates detection in a clinical environment.