期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2016
卷号：93
期号：2
出版社：Journal of Theoretical and Applied
摘要：Iris recognition technique is now regarded among the most trustworthy biometrics tactics. This is basically ascribed to its extraordinary consistency in identifying individuals. Moreover, this technique is highly efficient because of iris distinctive characteristics and due to its ability to protect the iris against environmental and aging effects. The Problem statement of this work is that the study presented an effective Iris recognition mechanism that is dependent on video. In this paper, it includes the data created based on best frame selected from iris video, then iris segmentation, normalization and feature extraction based on Hough Transform approach, Daugman rubber sheet modal and 1D Log-Gabor filter respectively. Flowed this proposed iris matching was proposed on the basis of two Deep Neural Networks models separately: Stacked Sparse Auto Encoders (SSAE) and Bi-propagation Neural network Models. Results are executed experimentally on the MBGC v1 NIR Iris Video datasets from the National Institute for Standards and Technology (NIST). The results displayed that Bi-propagation was the most efficient training algorithm for the iris recognition system based on video, and Stacked Sparse Auto Encoders (SSAE) is faster than Bi-propagation. Finally, that can be conclude an iris recognition system based on video that produces based on both separately models: Stacked Sparse Auto Encoders (SSAE) and Bi-propagation Neural network Models was achieves very low error rates was mean the both models are successfully for iris recognition system based on video.
