Elsevier

Pattern Recognition Letters

Volume 82, Part 2, 15 October 2016, Pages 132-143
Pattern Recognition Letters

A study of how gaze angle affects the performance of iris recognition

https://doi.org/10.1016/j.patrec.2015.11.001Get rights and content

Highlights

  • Eye structures related with iris recognition affects the performance of iris biometrics.

  • Images from same gaze angle shows the lowest Hamming distance.

  • As angle difference increases, the Hamming distance increases in intra-class distribution.

  • As angle difference increases, the average Hamming distance decreases in the inter-class.

Abstract

In traditional iris recognition systems, majority of researches improves the recognition performance for frontal images by ignoring the challenging issues including corneal refraction, 3D iris texture, limbus occlusion, and blur. When comparing images from the same angle such as frontal, all the challenging effects have similar distortions (i.e., corneal refraction and 3D texture) or minimal effects (i.e., limbus effect and blur) on all iris images. However, in off-angle iris recognition they have a significant negative impact on the accuracy and performance and they require additional treatments. In this paper, we first investigate how eye structures related iris recognition affects the performance of iris biometrics for different gaze angles and then quantify the effect of gaze angle on the inter-class and intra-class Hamming distance distributions. Based on our results from real images, as gaze angle of the probe image increases, the Hamming distance scores increases in intra-class distribution. We further found that average Hamming distance of the inter-class decreases as image captured from stepper angles and the distribution moves towards the intra-class distribution that causes an increment in the false-match-rate due to the existence of less area in the segmented off-angle images compared with the frontal images.

Introduction

The rapid increase in human population, revealed the need for the use of security systems for the quick identification and verification. So today, the use of biometric systems such as fingerprint, face, voice, signature, hand and vessel geometry, iris recognition are becoming increasingly common. Recent studies on biometric research confirm that iris recognition is one of the most truthful, unique, and consistent biometric methods for identity verification of an individual [1].

Existing commercial iris recognition systems are different versions of the Daugman's initial iris recognition method in [2] with the same methodology. They follow five main procedures including iris image capturing, iris segmentation, iris normalization, iris encoding and iris matching with the previously recorded iris database. After iris image is captured with a near infrared sensitive camera, iris region is segmented between inner (pupil) and outer (sclera) iris boundaries. In traditional iris segmentation, edge detection algorithms (such as Canny and Sobel) find edge points at the inner and outer iris boundary and circle detection algorithms (such as Hough transform) fits a circle for each boundary. In order to compare iris images with others independently from the difference in pixel size due to the angle and distance between camera and iris plane, zoom factor of camera, and pupil dilation, segmented circular iris texture is transformed into a dimensionless rectangular shape by using a pseudo polar coordinate system. Then, Gabor wavelet filters converts normalized iris images into binary iris code. The resultant iris code is compared with the previously enrolled iris codes in the database by calculating the Hamming distances between two iris codes. If the Hamming distance between two iris codes is less than the defined threshold, it is accepted that those iris codes belong to the same subject, or otherwise. It is expected that Hamming distance between same irises is close to zero and randomly selected two irises has 0.5 Hamming distance.

The identification performance of iris biometric systems relies on the quality of images and similarity of the data acquisition conditions. In order to get the highly accurate results from the traditional iris recognition systems, capturing the high quality frontal iris images are required. Existing iris recognition systems are undesirably affected by some challenging issues including gaze angle, pupil dilation, reflections, and occlusion that cause capturing non-ideal iris images [3]. These non-ideal iris images result in faulty decisions at the traditional iris biometric systems. Even with a trained operators and existence of cooperative subjects, it is possible to capture iris images in low quality, such as off-angle. In commercial iris biometrics systems, these non-ideal images are rejected by the system so capturing procedure is repeated until getting a high quality ideal frontal image.

Furthermore, recent improvements in camera and computer technologies have started transforming the iris biometrics systems from being a well-established and well-controlled setup, where an individual stands in front of a camera for identification, to being a smart standoff modality. Standoff iris biometric systems support to recognize the identity of both cooperative and non-cooperative individuals from an image or video sequence. The iris images captured in a standoff system are likely to be off-angle due to the much less constrained setup compared with a traditional system. Therefore, off-angle iris recognition is an emerging research topic in biometrics that focuses on addressing several challenging issues including corneal refraction, complex 3D iris texture, limbus occlusion and depth of field blur.

An example for frontal and off-angle iris images of same eye with same dilation level and their normalized images are shown in Figs. 1 and 2, respectively. The visual axis of the eye and the optical axis of the camera are perpendicular to each other in a frontal iris image, but in an off-angle image this angle is different. Since the dilation level difference between two images affects the Hamming distance, both iris images are selected with the same dilation levels to eliminate the dilation effect from the compared iris images. Their normalized images are shown below the corresponding iris images. It is easy to see that these two iris images have obvious differences. A traditional iris biometrics system based on the elliptical unwrapping measures the score of Hamming distance between these frontal and off- angle images as 0.45, close to score between randomly selected two irises.

Gaze angle difference between two compared iris images is one of the main degradation factors for the performance of the iris recognition not only in standoff systems but also in constrained traditional systems. Therefore, several researchers have proposed different methods to address the challenging issues in the off-angle iris images. However, there is no detailed prior research that quantifies how change in gaze angle affects the performance of iris recognition systems. This paper first explains how eye structures related iris recognition affects the performance of iris biometrics in different gaze angles and then quantifies the effect of gaze angle on the inter-class and intra-class distributions of Hamming distances.

The rest of the journal paper is organized in different sections as follows: The existing research in off-angle iris recognition literature is presented in Section 2. In Section 3, we explain the effect of eye structures related iris recognition and their impact on iris biometrics. Section 4 presented the off-angle iris dataset and software used to process these iris images. Effect of gaze angle on the performance of the iris recognition is presented and the potential approaches to solve these issues are discussed in Section 5. Finally, we conclude in Section 6.

Section snippets

Related works in off-angle iris recognition

In biometrics literature, John Daugman is known as the founder of the iris recognition system with his first practical implementation [4]. Existing iris recognition systems are different variations of his method. Recent works in literature have mainly focused on new algorithm design for frontal images that primarily affected from occlusion, refraction, variations in illumination, and blur [3]. These algorithms can be grouped into different categories including eyelid localization with complex

Effect of eye structures related iris recognition

Besides known degradation factors in iris recognition including pupil dilation, occlusion, image quality, focus, motion, blur, specular reflections, and illumination variations [3], [16], [17], the image acquisition angle (i.e., gaze angle) causes challenging problems including corneal refraction of light, change in appearance of complex three-dimensional iris texture, depth-of- field out of focus blurring, and limbus occlusion in iris recognition systems. In traditional iris recognition

Experimental off-angle iris dataset and algorithms

In order to quantify how gaze angle affects performance of iris biometrics, we used Oak Ridge National Laboratory Off-angle Dataset that includes iris images captured from 50 different individuals. In order to collect off-angle iris images, data captures setup in Fig. 9 is established where a camera is attached on a 50 cm in length movable arm support. A stepper motor rotates the arm with a constant speed from –50° to 50°. Dataset has been collected from subjects with diverse iris colors,

Effect of gaze angle on performance of iris recognition

Previous section shows that comparison of the frontal and off-angle iris images degrades the performance of the iris biometrics. However, this does not show the details about the effect of the gaze angle on performance of iris recognition based on the degree of angle difference. In order to investigate the details about the impact of gaze angle on iris recognition depending on the gaze angle differences between the compared irises, we first separate the datasets in different subsets according

Conclusion

Gaze angle difference between two iris images is one of the main degradation factors for the performance of the iris recognition not only in standoff systems but also in constrained systems. This paper first explained how eye structures related iris recognition affects the performance of iris biometrics for different gaze angles and then quantified the effect of gaze angle on the inter-class and intra-class distributions in iris recognition.

We observe that as gaze angle difference between of

Acknowledgment

This project was made possible by support from The Scientific and Technological Research Council of Turkey (TUBITAK) under the project number 113C024.

References (24)

  • SchuckersS.A.C. et al.

    On techniques for angle compensation in nonideal iris recognition

    IEEE Trans. Syst. Man, and Cybern.

    (2007)
  • J. Zuo et al.

    A robust iris segmentation procedure for unconstrained subject presentation

  • Cited by (0)

    This paper has been recommended for acceptance by Maria De Marsico

    View full text