Blood vessel segmentation in retinal images based on the nonsubsampled contourlet transform

This paper presents a method for blood vessel segmentation in retinal images based on the nonsubsampled contourlet transform (NSCT). The method uses a line detector and different directions of the NSCT subbands to detect the direction of blood vessel for each NSCT level. Then, the method computes three kinds of features using the orthogonal direction of blood vessel of the NSCT coefficients. After pixel classification, we present three kinds of post-processing procedures to correct the optic disk, the field of view mask, lesion areas, and noise. The performance is evaluated on the DRIVE database. The average accuracy of the DRIVE databases evaluation is 0.9423.