Automatic detection of multiple oriented blood vessels in retinal images

Abstract

Automatic segmentation of the vasculature in retinal images is important in the detection of diabetic retinopathy that affects the morphology of the blood vessel tree. In this paper, a hybrid method for efficient segmentation of multiple oriented blood vessels in colour retinal images is proposed. Initially, the appearance of the blood vessels are enhanced and background noise is suppressed with the set of real component of a complex Gabor filters. Then the vessel pixels are detected in the vessel enhanced image using entropic thresholding based on gray level co-occurrence matrix as it takes into account the spatial distribution of gray levels and preserving the spatial structures. The performance of the method is illustrated on two sets of retinal images from publicly available DRIVE (Digital Retinal Images for Vessel Extraction) and Hoover’s databases. For DRIVE database, the blood vessels are detected with sensitivity of 86.47±3.6 (Mean±SD) and specificity of 96±1.01.