Automatic optic disc detection using retinal background and retinal blood vessels

This paper presents an automatic optic disc detection technique that locates the optic disc through retinal background surface estimation and retinal blood vessel analysis. In the proposed technique, a retinal background surface is first estimated through an iterative Savitzky-Golay smoothing procedure. Multiple optic disc candidates are then detected from the difference between the retinal image under study and the estimated retinal background surface. Finally, the real optic disc is identified through the incorporation of the directional retinal blood vessel information. Experiments over four public datasets show that the proposed technique is accurate (acc. at 97.94%) and tolerant to retinal lesion and imaging artifacts.