Retinal vessel segmentation using non-subsampled contourlet transform and multi-scale line detection

We present an effective method for automatically extracting blood vessels from colour images of retinal. The proposed method is based on non _ subsampled contourlet transform (NSCT) and line detectors. In the first step we enhance green channel of retinal image by using the non-subsampled contourlet transform at five levels and 25 directions. This process improves discriminating vessels from background. In the next step we use a line detector at eight scales and twelve directions to extract proper features for detecting vessel centerlines. Finally, vessel width is measured at each pixel on a vessel centerline. The proposed method is evaluated and compared with several recent methods using images from the DRIVE database. As reported the performance of this method is very promising.