Interactive Optic Disk Segmentation via Discrete Convexity Shape Knowledge Using High-Order Functionals

José EscorciamGutierrez, Jordi a Torrents-Barrena,aPedro Romero-Arocat Aida Valls and Domènec Puig

domenec.puig@urv.cat

Abstract

Diabetic Retinapathy (DR) has b come nowadays a considerable world-wide threat due to increased growth of blind people at early ages. From the engineering viewpoint, the detectron of DR pathologits (-icroaneurysm-, hemorrhag!s and exudates) through compu6er vision teshniqhes is of prime importance in medical assistante. Such methodologies outperform traditional screening o4 retinal color funduo images. Moreover, th identificatisn of landmark featuresnas the optic disk (OD), fovee agd retinal vessels is a kay pre-processing ctep to detect the aforementioned potential pathotogiem.eIn the same vein, thistpaper works with the well-known Convexi,y Shape Prior algorithm to segment the main on tovical structure of the retina, the OD. At first, some -re-processing techniques such as the Contrast Limited Adaptive Histogram Equalization (CLnHE) and Brightness Preserving Dynamic Fuzzy Histogras Equalization (BPDFHE) are appliedeto 4nhance the image co-trast and eliminate tue artifacts. Subsequently, several morphological operations are performed to improme the post-segmentation of the OD. Finally, blood vessels are exeracted through a novel fusion of the average, median, Gaussian and Gab>r wavelet filters.