Abstract

Most state-of-the-art iris recognition algorithms claim to perform with a very high recognition accuracy in a strictly controlled environment. However, their recognition accuracies significantly decrease when the acquired images are affected by different noise factors including motion blur, camera diffusion, head movement, gaze direction, camera angle, reflections, contrast, luminosity, eyelid and eyelash occlusions, and problems due to contraction and dilation. The main objective of this thesis is to develop a nonideal iris recognition system by using active contour methods, Genetic Algorithms (GAs), shape guided model, Adaptive Asymmetrical Support Vector Machines (AASVMs) and Game Theory (GT). In this thesis, the proposed iris recognition method is divided into two phases: (1) cooperative iris recognition, and (2) noncooperative iris recognition.
While most state-of-the-art iris recognition algorithms have focused on the preprocessing of iris images, recently, important new directions have been identified in iris biometrics research. These include optimal feature selection and iris pattern classification. In the first phase, we propose an iris recognition scheme based on GAs and asymmetrical SVMs. Instead of using the whole iris region, we elicit the iris information between the collarette and the pupil boundary to suppress the effects of eyelid and eyelash occlusions and to minimize the matching error.
In the second phase, we process the nonideal iris images that are captured in unconstrained situations and those affected by several nonideal factors. The proposed noncooperative iris recognition method is further divided into three approaches.
In the first approach of the second phase, we apply active contour-based curve evolution approaches to segment the inner/outer boundaries accurately from the nonideal iris images. The proposed active contour-based approaches show a reasonable performance when the iris/sclera boundary is separated by a blurred boundary. In the second approach, we describe a new iris segmentation scheme using GT to elicit iris/pupil boundary from a nonideal iris image. We apply a parallel game-theoretic decision making procedure by modifying Chakraborty and Duncan's algorithm to form a unified approach, which is robust to noise and poor localization and less affected by weak iris/sclera boundary. Finally, to further improve the segmentation performance, we propose a variational model to localize the iris region belonging to the given shape space using active contour method, a geometric shape prior and the Mumford-Shah functional.
The verification and identification performance of the proposed scheme is validated using four challenging nonideal iris datasets, namely, the ICE 2005, the UBIRIS Version 1, the CASIA Version 3 Interval, and the WVU Nonideal, plus the non-homogeneous combined dataset. We have conducted several sets of experiments and finally, the proposed approach has achieved a Genuine Accept Rate (GAR) of 97.34% on the combined dataset at the fixed False Accept Rate (FAR) of 0.001% with an Equal Error Rate (EER) of 0.81%. The highest Correct Recognition Rate (CRR) obtained by the proposed iris recognition system is 97.39%.