Level set segmentation of optic discs from retinal images

Abstract

Analysis of retinal images can provide important information for detecting and tracing retinal and vascular diseases. The purpose of this work is to design a method that can automatically segment the optic disc in the digital fundus images. The template matching method is used to approximately locate the optic disc centre, and the blood vessel is extracted to reset the centre. This is followed by applying the Level Set Method, which incorporates edge term, distance-regularization term and shape-prior term, to segment the shape of the optic disc. Seven measures are used to evaluate the performance of the methods. The effectiveness of the proposed method is evaluated against alternative methods on three public data sets DRIVE, DIARETDB1 and DIARETDB0. The results show that our method outperforms the state-of-the-art methods on these datasets.