Moment Invariants based feature techniques for segmentation of retinal images using supervised method

Blood vessel extraction from retinal fundus images is an important task in developing the computer-aided diagnostic system for ophthalmologists. In this paper we have presented an algorithm for extraction of blood vessels of retinal fundus images and comparison of different moment invariants used for the extraction of features for the vessel pixels. The algorithm uses neural networks for distinguishing between vessel pixels and the non-vessel pixel. The moment invariants used are geometric moment invariants, complex moment invariants, Legendre moment invariants and Zernike moment invariants. The contribution of research work presented in this paper is in the experimentation and performance evaluation of different Moment Invariant techniques which concludes that accuracy of vessel identification and segmentation is relatively higher in Legendre Moment Invariant technique when compared to Hu´s MI used in the referred literature and GMI, CMI, and ZMI presented in this work. The accuracy of legendre MI is 1.142% higher than the accuracy of Hu´s MI. The complete algorithm was developed and implemented using the Matlab tools on the publicly available DRIVE database.