A fusion approach to unconstrained iris recognition

https://doi.org/10.1016/j.patrec.2011.08.017Get rights and content

Abstract

As biometrics has evolved, the iris has remained a preferred trait because its uniqueness, lifetime stability and regular shape contribute to good segmentation and recognition performance. However, commercially deployed systems are characterized by strong acquisition constraints based on active subject cooperation, which is not always achievable or even reasonable for extensive deployment in everyday scenarios. Research on new techniques has been focused on lowering these constraints without significantly impacting performance while increasing system usability, and new approaches have rapidly emerged. Here we propose a novel fusion of different recognition approaches and describe how it can contribute to more reliable noncooperative iris recognition by compensating for degraded images captured in less constrained acquisition setups and protocols under visible wavelengths and varying lighting conditions. The proposed method was tested at the NICE.II (Noisy Iris Challenge Evaluation – Part 2) contest, and its performance was corroborated by a third-place finish.

Highlights

► Analysis of different biometric recognition techniques. ► Fusion of such methods to achieve recognition on noisy iris data. ► Individual and global performance, with analysis of the improvement. ► Evaluation on both verification and identification modes. ► Conclusions show the applicability of our study on non-cooperative scenarios.

Introduction

The use of the iris as main biometric trait has emerged as one of the most recommended methods due not only to the possibility of noncontact data acquisition and to its circular and planar shape that facilitates detection, segmentation and compensation for off-angle capture but also for its predominately randotypic appearance. Although these factors contribute to high effectiveness in the currently deployed iris-recognition systems, their typical scenarios are quite constrained: subjects stop and stare relatively close to the acquisition device while their eyes are illuminated by a near-infrared light source, enabling the acquisition of high-quality data. As reported in the study conducted by Aton Origin for the United Kingdom Passport Service,1 imaging constraints are a major obstacle for the mass implementation of iris-based biometric systems. Notably, several researchers are currently working on minimizing the constraints associated with this process, in a way often referred to as noncooperative iris recognition, referring to several factors that can make iris images nonideal, such as at-a-distance imagery, on-the-move subjects, and high dynamic lighting variations.

In this study, we stress multiple recognition techniques, each one based on a different rationale and exploiting different properties of the eye region. Furthermore, we show how their fusion can increase the robustness to the degraded data typically captured in unconstrained acquisition setups.

The recognition techniques used in our proposition can be divide in two main categories. In one approach, we use wavelet-based iris-feature-extraction methods, complemented with a zero-crossing representation (Hoyle et al., 2010, Hoyle et al., 2009) and the analysis of iriscode-matching bit distribution (Santos and Proença, 2010). Complementarily, we expanded the extraction of features to the ocular region outside the iris, as recent studies (Savvides et al., 2010, Miller et al., 2010, Park et al., 2009) have suggested using these data, which appear to be a middle ground between iris and face biometrics and incorporates some advantages of each.

The performance of the fusion method we propose is highlighted by its third-place finish at the NICE.II (Noisy Iris Challenge Evaluation – Part 2), an international contest involving almost seventy participants worldwide.

The remainder of this paper is structured as follows: Section 2 describes the steps for iris-boundary localization and normalization, feature extraction and matching for the different approaches, and how their outputs are joined; Section 3 details the experimental process followed by a discussion of the obtained results; finally, Section 4 states the conclusions.

Section snippets

Proposed methodology

This section describes the five steps of our approach: iris-boundary detection, iris normalization, feature extraction, matching and decision ensemble (as schematized in Fig. 1). Furthermore, for feature extraction and matching, five recognition techniques are detailed.

Analysis of results

To assess the performance of the proposed method, experiments were conducted using 1,000 iris images from the UBIRIS.v2 (Proença et al., 2010) database used for the NICE.II4 contest, and their respective segmentation masks. Although this contest was based only on identification mode (performance was ranked through the decidability measure), our experiments were carried out in two modes: verification mode (one-to-one matching) and identification mode

Conclusions

In this study, we presented a novel fusion of different recognition approaches to address the issue of noncooperative iris recognition using nonideal visible-wavelength images captured in an unconstrained environment.

We tested several different autonomous approaches; their individual performances were evaluated in identification and verification modes and then the methods were fused, resulting in improved accuracy. We also showed that combining features extracted from the iris region itself

Acknowledgments

We acknowledge the financial support provided by “FCT-Fundação para a Ciência e Tecnologia” and “FEDER” in the scope of the PTDC/EIA/69106/2006 “BIOREC: Non-Cooperative Biometric Recognition” and PTDC/EIA-EIA/103945/2008 “NECOVID: Negative Covert Biometric Identification” research projects.

References (24)

  • D. Ballard

    Generalizing the Hough transform to detect arbitrary shapes

    Pattern Recognition

    (1981)
  • A.B.M. Cantor

    Understanding logistic regression

    Evidence-based Oncology

    (2002)
  • T. Tan et al.

    Efficient and robust segmentation of noisy iris images for non-cooperative iris recognition

    Image Vision Comput.

    (2010)
  • A. Agresti

    Categorical Data Analysis (Wiley Series in Probability and Statistics)

    (2002)
  • Boles, W., 1997. A security system based on human iris identification using wavelet transform. In: KES’97: Proc. First...
  • W. Boles et al.

    A human identification technique using images of the iris and wavelet transform

    IEEE Trans. Signal Process.

    (1998)
  • J. Canny

    A computational approach to edge detection

    IEEE Trans. Pattern Anal. Machine Intell. PAMI-8

    (1986)
  • I. Daubechies

    Ten Lectures on Wavelets

    (1992)
  • J.G. Daugman

    How iris recognition works

    IEEE Trans. Circuits Systems Video Technol.

    (2004)
  • Daugman, J., Williams, G., 1996. A proposed standard for biometric decidability. In: Proceedings of the...
  • D.W. Hosmer et al.

    Applied logistic regression (Wiley Series in probability and statistics)

    (2000)
  • Hoyle, E., Feitosa, R., Petraglia, A., 2009. Iris recognition using one-dimensional signal analysis. In: Proc. 8th...
  • Cited by (96)

    • Periocular biometrics: A survey

      2022, Journal of King Saud University - Computer and Information Sciences
    • Future Iris Imaging with Advanced Fuzzified Histogram Equalization

      2024, International Journal of Advanced Computer Science and Applications
    View all citing articles on Scopus
    View full text