Phase congruency based retinal vessel segmentation

Detection of blood vessels in a retinal fundus image is the preliminary step to diagnose several retinal diseases. There exist a number of methods to accomplish this task automatically. However, all of these methods suffer from lengthy processing time. The other major use of retina scanning is biometric authentication, for which a real time vessel detection system is required. In this work we describe a method that acquires binary vessel image from a color retinal fundus image in near real time. This method first generates the phase congruency image of the green channel of a color retinal image, and then thresholding is applied on the phase congruency image to obtain the blood vessels. Our method is able to acquire the blood vessels from a retinal fundus image within 10 seconds on a PC with the values of dasiaaccuracypsila and dasiaarea under ROCpsila: (0.92, 0.94) for the standard testing database called DRIVE. However, a (0.90, 0.91) accuracy and area under ROC can be acquired in less than 2 sec.