Elsevier

Signal Processing

Volume 93, Issue 1, January 2013, Pages 230-241
Signal Processing

Reliable iris localization using Hough transform, histogram-bisection, and eccentricity

https://doi.org/10.1016/j.sigpro.2012.07.033Get rights and content

Abstract

The iris technology recognizes individuals from their iris texture with great precision. However, it does not perform well for the non-ideal data, where the eye image may contain non-ideal issues such as the off-axis eye image, blurring, non-uniform illumination, hair, glasses, etc. It is because of their iris localization algorithms, which are developed for the ideal data. In this paper, we propose a reliable iris localization algorithm. It includes localizing a coarse iris location in the eye image using the Hough transform and image statistics; localizing the pupillary boundary using a bi-valued adaptive threshold and the two-dimensional (2D) shape properties; localizing the limbic boundary by reusing the Hough accumulator and image statistics; and finally, regularizing these boundaries using a technique based on the Fourier series and radial gradients. The proposed technique is tested on the public iris databases: CASIA V1, CASIA-IrisV3-Lamp, CASIA-IrisV4-Thousand, IITD V1.0, MMU V1.0, and MMU (new) V2.0. Experimental results obtained on these databases show superiority of the proposed technique over some state of the art iris localization techniques.

Highlights

► Robust iris region localization in the eye image using Hough transform and image gray statistics. ► Pupillary boundary localization using a bi-valued adaptive threshold and 2D-shape properties. ► Fast limbic boundary localization by reusing Hough accumulator and image gray statistics. ► Conversion of the circular iris contours to non-circular contours. ► Tolerance to off-axis eye images, glasses, contact lenses, eyelashes, eyelids, and hair.

Introduction

Traditional security systems have long been in service to identify and verify the humans at public and other numerous restricted zones; for example, airports, banks, citizen registration department, health care departments, and the border crossing control are to name a few [1]. These systems rely on the knowledge (e.g., passwords and the personal identification numbers) and tokens (e.g., keys and identity cards) which could be shared, lost, and/or hacked [2]. Due to these imperfections, the research community struggled to devise other security measures for the human verification and identification; for instance, the biometric technology. Biometric technology utilizes the physiological and physical traits, such as the earprints, iris, retina, face, voice, fingerprints, smell, DNA, and so on, to verify and identify individuals [3], [4]. Among these biometric traits, the iris has long been remained a hot research area for the research community owing to its stable, unique, and non-invasive properties. The iris has very complex structure that includes ridges, crypts, furrows, corona, arching ligaments, freckles, etc. [5], [6]. This structure is so unique that it remains stable over the entire life of a subject, except some negligible changes that occur in the early stages of a child [5], [6]. Due to these properties, the iris recognition technology has great potential for its applications in diverse areas; for example, the auto teller machines, access to a network of computers, border crossing control, etc.

A typical iris recognition system [5] includes the eye image acquisition, iris segmentation, features extraction, and matching and recognition. Among these fundamental modules, the iris segmentation module plays rather a critical role in the overall system performance because it isolates the valid part of an iris in the input eye image. It first localizes the pupillary (iris inner boundary/pupil circle) and limbic (iris outer boundary/iris circle) boundaries and then detects and excludes any other superimposed noise, such as the eyelids, eyelashes, and specular reflections in the iris region [1], [3], [4], [7]. However, most state of the art iris segmentation algorithms perform well for ideal data, which is acquired under very constrained environment [1]. In constrained environment, the image data is acquired in a very controlled way, such as the subject wearing no contact lenses and/or glasses, should stand at short distance and gaze directly into the camera [1]. On the other hand, the non-ideal data is acquired is the less constrained and/or un-constrained environment [1]. In the un-constrained environment [8], [9], the image data is captured with a subject at-a-move and/or at-a-distance using the visible wavelength (VW) or the near infrared (NIR) illumination source. Here, the eye image may contain numerous non-ideal issues, such as the blurring, non-uniform illumination, low contrast, off-axis and the off-angle eye images, specular reflections; eyelids, eyelashes, hair, glasses, and contact lenses occlusions. As the less-constrained environment imposes from minimum to maximum constraints on a subject, therefore the resultant eye image may contain the non-ideal issues, as mentioned earlier, but it is not necessary for the subject to be at-a-distance and/or at-a-move [10]. Researchers [1], [11] have contributed to the un-constrained systems, but this area is still immature and needs more efforts from the research community as reported by experts from the United Arab Emirates [12]. In this paper, we focus our work for the less constrained systems using NIR illumination sources. The NIR band reveals relatively more iris texture, especially for the dark irises, compared to the VW illumination sources [1].

The researchers used diverse techniques to segment an iris in the eye image; for example, the gradients and edge-detectors [5], [6], active contours models [13], [14], histogram and thresholding [3], [4], and so on. Literature reveals that the histogram and thresholding based techniques are relatively fast [1], [4]. Ibrahim et al. [3] used the histogram, thresholding, and standard deviation of the xy-coordinates of the pixel belonging to a binary object to localize the pupillary boundary. Here, the decision is based on the standard deviation only and no other supplementary parameters are adopted herein to make this technique more robust. Therefore, this technique may fail for a binary image containing other binary objects, which are caused by the low intensity regions, e.g., eyebrows, eyelashes, hair, and possible the black-frame of glasses. Similarly, Khan et al. [4] used the histogram, thresholding, and eccentricity to extract the pupillary boundary. However, similar to the method of Ibrahim et al. [3], they also did not propose any scheme to handle with multiple objects in the binary image. Therefore, this eccentricity-based decision could also be misleading, because a small object (other than the pupil object) may have the smaller eccentricity. Moreover, Basit and Javed [15] used a type of the image integral projection function [16] to first localize a coarse pixel in the pupil region and then extract the pupil circle via a thresholding and gradient-based techniques. This scheme may not work for an eye image containing the other low intensity regions, as mentioned before.

Besides, another major problem in most of the iris segmentation algorithms is of the off-axis eye image (eye not oriented horizontally). It is because that most of the researchers [1], [2], [3], [4], [15] biased their localization technique in the horizontal direction while localizing the limbic boundary. To resolve these issues, we propose a reliable iris localization algorithm. It comprises preprocessing the input eye image; localizing a coarse iris region (CIR) using the Hough transform and image gray level statistics; localizing the pupillary boundary within the CIR using a bi-valued adaptive threshold and the 2D shape properties; localizing the limbic boundary by reusing the Hough space and image statistics; and finally, regularizing these boundaries using the Fourier series and gradients. In this paper, we focus our work on localizing the pupillary and limbic boundaries only, i.e., the resultant localized iris would contain the eyelids and eyelashes occlusions, as in [3], [4], [9], [17], [18], [19]. However, to remove these occlusions, we refer to the published work [1].

The remaining of this paper is organized as follows: Section 2 details the different modules of the proposed algorithm while experimental results and discussion are given in Section 3. Limitations of the proposed technique are given in Section 4 and finally, Section 5 concludes the paper.

Section snippets

Proposed iris localization algorithm

The proposed technique involves the following basic modules: Pupillary Boundary Extraction, Limbic Boundary Extraction, and the Iris Boundaries Regularization. Fig. 1 shows the block diagram of the proposed method, which is explained in the following sections.

Experimental results and discussion

To evaluate performance of the proposed technique, we used the MATLAB version 7.1 installed on a PC with 2.33 GHz CPU and 1 GB RAM and a set of the public iris databases: CASIA V1 [31], CASIA-IrsiV3-Lamp [31], CASIA-IrsiV4-Thousand [31], IITD V1.0 [32], MMU V1.0 [21], and MMU (new) V2.0 [21]. All these databases have been developed with the NIR illumination source. As mentioned earlier, the NIR range exposes more iris texture and is suitable for the dark irises [1]. The set of these databases

Limitations of the proposed iris localization algorithm

The proposed algorithm does not perform well for an eye image where the pupil region is severely occluded by a strip of eyelashes, which have gray level intensity similar to the pupil region. Fig. 10 shows two different eye images where the proposed algorithm is failed.

Conclusion

To conclude this paper, we summaries its key points: For fast processing, we first localize the coarse location of iris in the eye image using a robust technique based on the Hough transform and image gray level statistics. Secondly, we use a reliable technique comprising a bi-valued adaptive-threshold based on the Histogram-bisection and image gray level statistics; eccentricity, area, and the 2D-object geometry to localize the pupillary boundary within the coarse iris region. Thirdly, we

Acknowledgment

The authors highly acknowledge COMSATS Institute of Information Technology, Islamabad campus Pakistan, for its in-house Ph.D. program. We also thank the following: Malaysia Multimedia University; Indian Institute of Technology Delhi; Department of Computer Science, SOCIA Laboratory; and the Chinese Academy of Sciences for granting free access to their respective developed standard iris databases.

References (36)

  • P. Li et al.

    Robust and accurate iris segmentation in very noisy iris images

    Image and Vision Computing

    (2010)
  • A. Basit, Iris Localization Using Graylevel Texture Analysis and Recognition Using Bit Planes. Ph.D. Thesis. Department...
  • J.G. Daugman

    High confidence visual recognition of persons by a test of statistical independence

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (1993)
  • R.P. Wildes

    Iris recognition: an emerging biometric technology

    Proceedings of the IEEE

    (1997)
  • H. Proenca et al.

    The UBIRIS.v2: a database of visible wavelength iris images captured on-the-move and at-a-distance

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (2010)
  • S. Venugopalan, M. Savvides, Unconstrained iris acquisition and recognition using COTS PTZ camera, EURASIP Journal on...
  • UBIRIS iris database, 〈http://iris.di.ubi.pt/〉 (last accessed:...
  • Title: “Iris at a distance not yet mature enough, says UAE”....
  • Cited by (0)

    View full text