A New Method for Retinal Vascular Segmentation based on the GMM Algorithm

The structure and morphology of retinal blood vessels contain many clinical information. Given the low contrast and uneven distribution of gray colors on the fundus image, achieving an accurate retinal vascular segmentation remains a challenge. To address this problem, the fundus image is initially enhanced via adaptive histogram equalization and bottom-hat transform. Afterward, the Gaussian mixture model is used to classify the enhanced images into three classes, namely, the background cluster, the vessel cluster, and the retinal disc cluster. Finally, denoising is performed. The experimental results reveal that the aforementioned approach can generate more accurate retinal vascular segmentation images compared with those obtained by other methods when applied on DRIVE dataset and STARE dataset.