Feature quality fusion based multimodal eye recognition
Abstract
Sclera vessel patterns in the human eye are unique and have both genetic and developmental components that determine their structure. The vessel patterns can achieve better accuracy for human identification than other traditional methods using visible light. However, poor quality sclera images can significantly affect recognition accuracy. In this thesis, we propose a comprehensive approach for sclera image quality measurements and a multi-angle sclera recognition system including frontal and side-looking sclera images. Results presented in this thesis show that these proposed methods can be used to improve the performance of sclera recognition systems. To further improve the recognition accuracy for human identification, we propose a multimodal eye recognition system to fuse sclera and iris recognitions for human identification. Iris recognition is shown to be one of the most reliable approaches for automatic human recognition with frontal high quality near-infrared (NIR) images. However, a dark brown eye can only reveal a rich and complex iris pattern using NIR light, which can dramatically affect the accuracy of iris recognition in visible light. Sclera recognition can achieve better accuracy than other traditional methods in visible light. Results in this thesis show that our proposed multimodal eye recognition method can achieve better performance compared to iris or sclera recognition. Non-ideal eye images are still challenging for eye recognition and can significantly affect the accuracy of eye recognition systems because they cannot be properly preprocessed and/or they have poor image quality. In order to eliminate the effect of image quality, we propose a feature quality fusion based multimodal eye recognition system. Our quality measure evaluates the entire eye image including the iris area and the sclera area. The experimental results show that our overall iris and sclera quality scores are highly correlated to recognition accuracy. Furthermore, our quality fusion based eye recognition can improve the performance of eye recognition systems.
Degree
Ph.D.
Advisors
Delp, Purdue University.
Subject Area
Engineering|Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.