Automatic Segmentation of Hard Exudates Using LAB Color Space Contours Edge Detection and Morphological Operation

Diabetic retinopathy is a serious medical condition that affects people across the world and diminishes the quality of life. Hard exudates are one of the most common and early signs of diabetic retinopathy. Segmentation of hard exudates is a challenging task due to the wide diversity in features such as irregular shape, size, and location. This paper proposes an efficient and accurate approach for the segmentation of hard exudates from RGB (Red, Green, Blue) fundus images. An unsupervised approach partitions the fundus image into disjoint and mutually exclusive regions. With this approach, the system is flexible enough to use different methods in diverse regions. Luminance and A-axis component generates a new image and adaptive threshold extracts the exudates in one region. In other regions, the unsupervised multi-stage algorithm detects the region of exudates. Morphological operation followed by contour features enhances the exudates region and removes the outliers. The proposed approach obtains 0.913, 0.981, 0.975 of recall, specificity, and accuracy on the ISBI IDRiD dataset.