Fully automatized parallel segmentation of the optic disc in retinal fundus images

Abstract

This paper presents a fully automatic parallel software for the localization of the optic disc (OD) in retinal fundus color images. A new method has been implemented with the Graphics Processing Units (GPU) technology. Image edges are extracted using a new operator, called AGP-color segmentator. The resulting image is binarized with Hamadani’s technique and, finally, a new algorithm called Hough circle cloud is applied for the detection of the OD. The reliability of the tool has been tested with 129 images from the public databases DRIVE and DIARETDB1 obtaining an average accuracy of 99.6% and a mean consumed time per image of 7.6 and 16.3 s respectively. A comparison with several state-of-the-art algorithms shows that our algorithm represents a significant improvement in terms of accuracy and efficiency.