Retinal Identification System based on Optical Disc Ring Extraction and New Local SIFT-RUK Descriptor

Abstract : Personal recognition based on retina has been an attractive topic of scientific research. A common challenge of retinal identification system is to ensure a high recognition rate while maintaining a low mismatching FMR rate and execution time. In this context, this paper presents a retinal identification system based on a novel local feature description. The proposed system is composed of three stages, firstly we enhance the retinal image and we select a ring around the optical disc as an interest region by using our recently proposed Optical Disc Ring ODR method. Secondly, in order to reduce the mismatching rate and speed up the matching step, we propose in this paper an original alternative local description based on the Remove of Uninformative SIFT Keypoints, that we call SIFT-RUK. Finally, the generalization of Lowe’s matching technique (g2NN test) is employed. Experiments on the VARIA database are done to evaluate the performance of our proposed SIFT-RUK feature-based identification system. We show that we obtain a high performance with 99.74% of identification accuracy rate without any mismatching (0% of False Matching Rate FMR) and with a low matching processing time compared to existing identification systems.