Automatic and efficient detection of the fovea center in retinal images

A fast and automated method of the MF (macula fovea) center detection in retinal images is presented, which is a prerequisite for DR (diabetic retinopathy) screening project or the computer aided diagnosis of DME (diabetic macular edema). The approach is based on foregone anatomical structures of retina constraints on the relative locations and the intensity information within the OD (optic disk) circle boundary. Firstly, OD circle boundary is obtained accurately using adaptive thresholding, morphological bottom-hat transform and region-based active contour model techniques. And then the fovea region is selected by analyzing the intensity information within the OD. Lastly, MF center is detected using mathematical morphology. The public DRIVE, HEI-MED and DIARETDB1 databases are used to test the performance of the algorithm. The MF center is detected correctly in 100% of the 35 images cases in the DRIVE database, in 98.8% of 169 images cases in the HEI-MED dataset, and in 93.3% of the 89 cases in the DIARETDB1 database. It takes only about 12 seconds for per image because of not relying on blood vessel segmentation in advance, which is the main advantage of the presented approach. The experimental results show that the method potentially should get a better performance than comparable methods presented in other literatures.