Optic cup segmentation based on extracting blood vessel kinks and cup thresholding using Type-II fuzzy approach

A novel technique has been developed to segment the optic cup from a 2D colored fundus image. Cup segmentation is the most challenging part of image processing the optic nerve head (ONH) due to the complexity of its structure. The cup size is used for diagnosis of glaucoma. Blood vessels densely cover the cup boundary in some cases and in other cases, they form the boundaries. Therefore, extracting the vessels were conducted by using a top hat transform and Otsu's function in order to detect the curvature of the blood vessels (kinks), which indicates the cup boundary. Then, an Interval Type-II fuzzy entropy procedure was applied to cup thresholding. Finally, the Hough transform was applied to approximate the cup boundaries. The algorithm was evaluated on 100 fundus images from the RIGA database, where the cup was manually marked by 6 Ophthalmologists. The cup detection accuracy was 72.5%.