Exudates dynamic detection in retinal fundus images based on the noise map distribution

In this paper a new and reliable method to detect and segment the exudates in retinal fundus images is presented. The introduced approach is based in the computation of the noise map distribution, and on the use of morphological operators and adaptive thresholding. The proposed method reveals a good resilience to contrast changes, non-uniform illumination and variable background, resulting in a correct detection of exudates. A sensitivity of 97.49%, a specificity of 99.95% and an Accuracy of 99.91% is achieved for the exudates detection. Moreover, the method is very effective when applied in the segmentation of the exudates. A comparison is made with recent works. The application of the method shows a strong potential that can be applied in automatic detection and follow-up studies in diabetic retinopathy.