Using iris and sclera for detection and classification of contact lenses☆
Introduction
Iris-based biometric authentication systems are becoming extremely popular thanks to their high reliability in both identification and verification tasks [5], [11], [27]. This is essentially due to the properties of the iris pattern, which is unique even for identical twins, and so rich of distinctive features that a casual wrong identification is seldom observed. Moreover, as an internal organ of the eye, the iris is well protected from the environment and stable with age [10].
The performance of iris recognition systems, however, can sharply degrade when the users wear contact lenses. The impairment is quite significant in the case of cosmetic lenses, used to modify the appearance of the iris, so as to alter its color and possibly also its texture [9]. However, it may be non-negligible also for transparent lenses, as recently shown in [2], due to induced artifacts. For example, the lens generally moves around on the eye’s surface, causing several artifacts that prevent a correct matching and increase the false rejection rate in the identification task. Hence, to improve iris recognition, it is important that the system be able to recognize whether the user wears contact lenses, and of which type, cosmetic or transparent.
Some examples of such images coming from the IIIT-D contact lens database [21] are shown in Fig. 1. It can be seen that cosmetic lenses have, typically, a very marked texture, more visible than that of natural irises, suggesting that they can be easily recognized based on the analysis of the iris region. Indeed, the iris with its complex structure, provides abundant textural information to exploit for classification purposes. On the contrary, transparent lenses, and in particular soft lenses, do not alter the characteristics of the iris. However, unlike in the no-lens case, a boundary of circular shape is always present, more or less visible, in the sclera region. We argue that these differences can be well captured by a fine-scale statistical analysis conducted over small patches of the images, provided that also part of the sclera region is examined.
In particular, in this work we use a Bag-of-Features paradigm where dense local features are extracted from all patches of the region of interest and jointly quantized for the classification task. This approach has proven very promising for detecting the biometric spoofing of different traits [15]. The regions of interest are extracted by means of an ad hoc segmentation algorithm. We take into account information coming both from the iris and from part of the sclera region, in order to detect both textured and transparent lenses. Moreover, in order to prevent any type of texture distortion, we avoid any normalization.
Preliminary results of this work were presented in [14]. With respect to that paper, the current version is largely improved under several points of view: (i) new versions of the proposed algorithm have been devised and analyzed; (ii) a more thorough assessment of the segmentation algorithm has been carried out, through the inclusion of new specific datasets; and (iii) the comparative assessment of classification results, conducted on the same datasets used in [48], includes a further technique very recently published. Such new data provide a better insight into the effectiveness of the proposed method.
The rest of the paper is organized as follows. In Section 2 the current literature on this topic is reviewed. Then, in Section 3, our approach is described in detail, while the experimental results are reported in Section 3.1. Eventually, Section 4 draws some conclusions.
Section snippets
Related work
Most of the research on this topic focuses on colored lenses, since they can be also used by a malicious user to attack the system. Fakes made by cosmetic lenses are the most challenging to detect, since, differently from printed-iris attacks [39], the alteration is localized in a very small part of the image.
A pioneering work in this direction is due to [10] where irises are detected by means of the periodicities left by the dot-matrix printers used in the fabrication process. Another method,
Proposed approach
In this section we describe the proposed method, which comprises three major steps: segmentation of the iris and part of the sclera region, dense feature extraction and classification by means of the Bag-of-Features paradigm.
Experimental results
To assess the classification performance of the proposed method a number of experiments were performed on the publicly available datasets used in [48] . Since these are thoroughly described in the reference paper, we report here only the major characteristics of interest for our experiments. The images of the NotreDame-I and II datasets, as already said, are provided with the associated ideal segmentation of the iris, while this is not available for the IIIT-D Cogent, and IIIT-D Vista datasets.
Conclusions
In this work, we propose a novel method to detect and classify cosmetic and soft contact lenses. A preliminary parametric segmentation of the image allows us to accurately identify the iris region. However, we also include part of the sclera in the region under analysis, since this area turns out to be very discriminative when dealing with transparent lenses. To perform classification we use the Bag-of-Features paradigm with features associated to the rotation and scale-invariant local
References (49)
- et al.
Degradation of iris recognition performance due to non-cosmetic prescription contact lenses
Comput. Vis. Image Underst.
(2010) Generalizing the hough transform to detect arbitrary shapes
Pattern Recognit.
(1981)- et al.
Image understanding for iris biometrics: a survey
Comput. Vis. Image Underst.
(2008) The importance of being random: statistical principles of iris recognition
Pattern Recognit.
(2003)- et al.
A comparative study of texture measures with classification based on feature distributions
Pattern Recognit.
(1996) - et al.
Efficient iris segmentation method in unconstrained environments
Pattern Recognit.
(2013) - et al.
Iris boundaries segmentation using the generalized structure tensor. a study on the effects of image degradation
Proceedings of the IEEE Conference on Biometrics: Theory, Applications and Systems
(2012) - et al.
Representing shape with a spatial pyramid kernel
Proceedings of the ACM International Conference on Image and video retrieval
(2007) A computational approach to edge detection
IEEE Trans. Pattern Anal. Mach. Intell.
(1986)Pupil dynamics for iris liveness detection
IEEE Trans. Inf. Forensics Secur.
(2015)
High confidence visual recognition of persons by a test of statistical independence
IEEE Trans. Pattern Anal. Mach. Intell.
Demodulation by complex-valued wavelets for stochastic pattern recognition
Int. J. Wavel. Multiresolut. Inf. Process.
New methods in iris recognition
IEEE Trans. Syst. Man Cybern. Part B
Automatic detection of non-cosmetic soft contact lenses in ocular images
SPIE Biometric and Surveillance Technology for Human and Activity Identification
Fingerprint liveness detection using binarized statistical image features
Proceedings of the IEEE International Conference on Biometrics: Theory, Applications and Systems
Contact lens detection and classification in iris images through scale invariant descriptor
Proceedings of the Workshop on Insight on Eye Biometrics in Conjunction with SITIS
An investigation of local descriptors for biometrics spoofing detection
IEEE Trans. Inf. Forensics Secur.
Efficient iris spoof detection via boosted local binary patterns
Advances in Biometrics
A ground truth for iris segmentation
Proceedings of the International Conference on Pattern Recognition
Detection of contact-lens-based iris biometric spoofs using stereo imaging
Proceedings of the Hawaii International Conference on System Sciences
Aggregating local descriptors into a compact image representation
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition
Bsif: Binarized statistical image features
Proceedings of the International Conference on Pattern Recognition
Revisiting iris recognition with color cosmetic contact lenses
Proceedings of the International Conference on Biometrics
Cited by (30)
Enhanced iris presentation attack detection via contraction-expansion CNN
2022, Pattern Recognition LettersCitation Excerpt :The classifiers based on type of input information can be described based on what and how many information is used. For example, [8–10] extracted the features from several local and overall global regions, and in the end, the decision is taken by combining the scores computed over each image region. [11] utilizes the 2D and 3D information for iris presentation attack detection.
A Less-constrained Sclera Recognition Method based on Stem-and-leaf Branches Network
2021, Pattern Recognition LettersCitation Excerpt :These characteristics of sclera make it ideal for human biometrics. However, the existing research in the field of sclera recognition treats it as the complement for iris recognition, rather than as a separate topic [4,16], leading to the existing datasets fusing iris and sclera vasculature [15], and some methods relying on the iris prior-knowledge [1]. It underestimates the potential of the sclera recognition system.
Iris anti-spoofing through score-level fusion of handcrafted and data-driven features
2020, Applied Soft Computing JournalCitation Excerpt :A similar approach is demonstrated in [33], where the BSIF features are extracted from three diverse sclera regions then separate scores are acquired with SVM and pooled through the majority voting. The use of the sclera region was additionally explored by [4] for soft lens detection, considering that there always exists a ring-like structure, i.e. lens boundary that is discovered through pixel intensity analysis. This method incorporates iris segmentation from eye images.
A Survey on Iris Presentation Attack Detection
2024, Zidonghua Xuebao/Acta Automatica SinicaA review on contact lens inspection
2023, Indonesian Journal of Electrical Engineering and Computer ScienceAutomated Attendance Tracking and Wage Management System Using IRIS Detection
2023, 2023 7th International Conference On Computing, Communication, Control And Automation, ICCUBEA 2023
- ☆
This paper has been recommended for acceptance by Maria De Marsico and Maria Frucci.