A Novel Method to Detect Fovea from Color Fundus Images

The computer-aided diagnosis technology in retinal image analysis requires localization of different fundus structures. Efficient detection and localization of fovea are essential in the analysis of diabetic macular edema. This paper demonstrates a novel technique for detection of fovea from color fundus images based on image enhancement by adaptive manifold filter and further mathematical morphological operations for final foveal center localization. The major advantage of the proposed technique is that it does not need a spatial relationship of optic disc and vessels for the detection of fovea. It is robust to illumination changes and interference caused by retinal pathologies. Experiments show encouraging results that are analyzed on five publically available databases DRIVE, HEI-MED, DIARETDB1, HRF, and MESSIDOR with an accuracy of detection as 100%, 99.40%, 98.88%, 100%, and 98.66%, respectively. Comparative analysis of results indicates that the proposed method achieves better performance than other earlier methods present in the literature.