Neural Network Algorithm for Biometric Analysis of Human Retina Image

Abstract

Identity recognition is an important process because many systems require a valid user identity for security and access control. Identity recognition such as passwords, signatures, id cards have some weaknesses that are they can be duplicated, stolen, forgotten, and even lost. Identity recognition using biometric techniques is known to be more reliable. Biometric technique is a recognition and classification technique that uses human behavior and physical atributes. In this research, a non-realtime simulation system is designed to identify a person by biometric of retina image. The system can identify one's identity through pattern of retinal blood vessels. The processes of this system divided into two stages that are training stages and testing stage. The identification process begins with prepocessing retinal photo. Biometric features extracted by using Discrete Orthonormal S Transform (DOST). Biometric classification by using Adaptive Resonance Theory 2 (ART 2) with unsupervised learning process that can recall previously learned patterns . The results obtained from this study showed 65% of accuracy  for the right retina image and 50% of accuracy for the left retina image. Computing time is about 6 seconds. Further development is needed to improve the accuracy of the system as a security and access control systems.