Retinal Vessel Segmentation Using Non- Subsampled Directional Filter Bank and Hessian Multiscale Filter Enhancement

Automated extraction of retinal vessels play an important role in the diagnosis of a wide range of retinal diseases and also for diagnosing complications due to cardiovascular diseases, stroke and hypertension. The blood vessels of the retina are a complex network and manual segmentation of them is a prolong and tedious task which requires high skills and training. In this paper, a novel method is proposed to segment the retinal vessels using Non-Subsampled Directional Filter Bank (NSDFB) and improved Hessian filter. Vessel enhancement is an important preprocessing step for vessel diagnosis and further processing. The second order geometrical structures are exploited for local shape properties in Hessian-based methods. But the classical Hessian filters are sensitive to noise and suffer from well-known drawbacks such as intensity inhomogeneity, vessel junction suppression. To resolve these issues, a new vessel enhancement approach based on NonSubsampled Directional Filter Bank and Hessian multiscale filter is used to enhance the vessels. An image grayscale factor is added to the vesselness function computed by Hessian matrix eigen value to reduce the pseudo vessel structures. This technique locates and segments the blood vessels using entropy thresholding and morphological operations. This proposed segmentation method is tested on publicly available STARE, DRIVE, CHASE_DB, HRF database. These public databases have manually labeled images which have been established to facilitate comparative studies on segmentation of blood vessels in retinal images. The performance of the proposed algorithm is evaluated on the basis of sensitivity, accuracy and specificity. The performance measures are also compared with the manually segmented results of publically available databases. The proposed technique achieves high mean accuracy and sensitivity while compared with the several previously proposed algorithms.