IRIS TECHNOLOGY: A REVIEW ON IRIS BASED BIOMETRIC SYSTEMS FOR UNIQUE HUMAN IDENTIFICATION

Biometric features are widely used in real time applications for unique human identification. Iris is one of the physiological biometric features which are regarded as highly reliable in biometric identification systems. Often iris is combined with other biometric features for robust biometric systems. It is also observed that biometrics is combined with cryptography for stronger security mechanisms. Since iris is unique for all individuals across the globe, many researchers focused on using iris or along with other biometrics for security with great precision. Multimodal biometric systems came into existence for better accuracy in human authentication. However, iris is considered to be most discriminatory of facial biometrics. Study of iris based human identification in ideal and non-cooperative environments can provide great insights which can help researchers and organizations that depend on iris-based biometric systems. The technical knowhow of iris strengths and weaknesses can be great advantage. This is more important in the wake of widespread use of smart devices which are vulnerable to attacks. This paper throws light into various iris-based biometric systems, issues with iris in the context of texture comparison, cancellable biometrics, iris in multi-model biometric systems, iris localization issues, challenging scenarios pertaining to accurate iris recognition and so on.


Introduction
Biometrics is an automated approach which exploits measurable physiological, physical and behavioural traits of humans for identification and authentication. Physiological and behavioral are the two categories in biometrics. The former refers to hand and palm geometry, DNA, face, iris, scent signature, keystroke dynamics and fingerprints while the latter refers to voice, gait, and typing rhythm. The real world applications of biometrics include detection and border security, fraud prevention, crime tracking, security, payment systems, attendance recording, physical and logical access controls, and identification of parties or individuals in general [28], [29]. Biometrics is one of the best ways in which individuals can be identified uniquely across the globe. Biometrics can be used in cryptography to secure communications in the real world networks [1].
Biometric templates when compromised, the security will be lost. To overcome this problem, cancellable biometrics approach came into existence [3], [20], and [30]. Hamming Distance Classifier (HDC) for predicting false rejection rate (FRR) and false acceptance rate (FAR) based on the hamming distance threshold was proposed [4]. Ocular biometrics was given importance by Simona and Arun [5].Studies were made on iris and face as biometric features to protect communications in mobile devices [6] It was noted from the literature that studies were made on binary iris code for reconstruction of original iris image [7]. Investigations on iris and fingerprints together for human identification were also carried out [8]. It is focused on the UID project in India named "Aaadhar". The investigations dealt the issues with biometric systems in the wake of security attacks on multimodel biometric systems [9, 10, and 19]. Iris localization is very important activity in commercial iris recognition systems. However, they could not perform well with ideal data as they work for controlled data. Many iris localization experiments were performed [11][12][13][14][15]. Lee et al. [16] made sensitivity analysis on biometric systems in the wake of attacks on such systems that help in finding the robustness of biometric system. Combination of combined error correction codes and finger prints [17]; multi-model biometric system using face and iris combination [18], SVM and feature selection techniques [21], Circular Hough Transform and K-Means algorithm [21], combination of different approaches [22] were used for iris recognition. Reverse bio-orthogonal wavelet transform technique was used for reliable iris recognition [24].
Iris hazards in the presence of noise were explored [25]. Pattern recognition and its importance in iris-based biometric system were presented by Unar et al. [26] while Zhu et al. [27] used iris based biometric system for random number generator. Iris based biometric system has become one of the most active search field sand it is driven by many applications towards authentication sand recognitions of an individual identity. From the above literature, it can be noted that limited studies were carried out on Iris Based Biometric Systems for Unique Human Identification.
In this paper, the concept and some of the important biometric systems which are Iris based are deliberated along with the security issues of the iris based human identification systems.
The remainder of the paper is structure as follows. Section II reviews iris templates and cancellable biometrics. Section III focuses on biometric binary strings. Section IV throws light on multi-model biometrics. Section V presents GA for iris reconstruction. Section VI focuses on security issues with biometric systems. Section VII and VIII discuss about iris localization and iris recognition systems. Section IX concludes the paper.

Iris Templates and Cancellable Biometrics
There is a study on iris-templates for crypto-biometric schemes as in [1]. This scheme helps users to get secret keys by using her biometric template. Fuzzy extractors are used to make the scheme robust. The scheme has both enrolment phase and verification phase. The enrolment ensures polynomial security and verification phase, also has polynomial complexity, and takes care of verifying the identity of users. Few authors focused on functional dual tree complex wavelet for biometric security and its applications include transient signal processing, image transmission, image compression and biometrics.
Biometric templates when compromised, the security will be lost. To overcome this problem, cancellable biometrics approach came into existence. This will take care of transformation functions in order to hide the original template. In this case the transformed biometric template when compromised, the original template can be used to make new transformation [30]. Towards cancellable biometrics as in [3] studied different fusion approaches in order to achieve cancellable recognition with multi-biometrics. They focused three cancellable transformations on two biometric modalities based on iris and voice. There is a methodology used for cancellable biometrics approach which is as shown in Figure 1. Two modalities are demonstrated with two biometric templates. Figure 1: Overview of methodology for cancellable biometrics [3] Void and iris datasets are used to make experiments. For transformations, three techniques are used namely convolution, interpolation, and bio-hashing. When compared, the interpolation has proved to be more accurate. Overall performance is improved when Sum or SVM techniques are used in all the cases analysed. With different fusion approaches, multi-biometric cancellable recognition was achieved. The results revealed that using multiple transformations can improve the robustness of the cancellable biometrics approach. In both the template cases, the cancellable transformations are generated and then multiple individual classifiers are generated. Finally multi-algorithm and multi-modal fusion is used for final transformation [3]. There is a study [20] on cancellable multi-biometrics based on adaptive bloom filters and iris codes.

Biometric Binary Strings
Hamming Distance Classifier (HDC) for predicting False Rejection Rate (FRR) and False Acceptance Rate (FAR) Based on the hamming distance threshold was proposed [4]. The proposed approach can be used in the real world biometric modalities such as face, signature, and iris and fingerprint texture. Moreover, they proposed a template protected biometric authentication system. Figure 2: Overview of template protected biometric verification system [4] As can be seen in Figure 2, it is evident that there are two phases such as enrolment and verification. In either case, feature extraction is made and real-valued classifier is built. There are two important modules such as secure bit extraction and secure key binding verification. The former module is used for transforming real-valued features into a binary string which is further used in secure key binding verification. The latter module is meant for verification of the protected target biometric string. Such string is bound with cryptographic key for highest level of security [4].

Multi-Model Biometrics
There is studied iris and face as biometric features to protect communications in mobile devices [6]. As the mobile devices are vulnerable to various attacks, authentication with iris and face could prevent them. They built a mobile management system using biometrics which is embedded in mobile devices. This solution can also be used in security-critical applications in the real world. Their system is named FIRME which has the architecture as presented in Figure  3. Figure 3: Combination of iris and face for recognition [6] As can be seen in Figure 3, it is evident that there are many phases for modelling face and iris and fusing them for recognition. The phases include capture, detection of iris, detection of face, segmentation of iris, segmentation of face, feature extraction of iris, anti-spoofing of face, feature extraction of face, template selection for iris and face, matching of iris and face and fusion. With the help of the two models and fusion, the system is able to recognize humans live.
There is a framework focused on biometric systems for mobiles using data mining techniques and ECG based identification [23]. There is studied [8] iris and fingerprints together for human identification. They focused on the UID project in India named "Aaadhar". The combination of iris and fingerprints make the system robust and can uniquely identify humans across the globe. From Figure 4, it can be assumed that the iris of human eye and fingerprints together form a basis for high security in human identification. The fusion of these two is challenging for feature selection. However, many real world systems are using the combination of both. For finding similarity in trained and testing samples two distance measures such as Mahalanobis distance and Euclidean distance are used. Thus the identity of a person can be established.

GA for Iris Reconstruction
There is a framework [7] used binary iris code for reconstruction of original iris image. Probabilistic approach was used along with genetic algorithms for iris image reconstruction from given binary templates. This solution was proved realistic and had potential to support iris as reliable biometric feature for human identification. This solution has three phases namely segmentation, normalization and occlusion mask and encoding. These three phases are as visualized in Figure 5. As can be seen in Figure 4, it is evident that the solution has three phases. In the first phase segmentation takes place. In the second phase normalization takes place for transforming iris segments into a rectangular image.
The encoding phase uses some sort of filtering that can for binary representation of iris image or iris code which is further used for human authentication.

Security Issues with Biometric Systems
Biometric systems that make use of multiple features of biometrics have been reported to face attacks. Though biometric technology captures what is being done and who is doing it, there are direct and indirect attacks directed towards face and iris fusion. Recent research revealed that multi-model biometric systems are vulnerable to spoofing attacks. There might be other software based attacks still unexplored in the real world. According to research as in [9] spoofing attacks are considered direct attacks that are made with synthetic biometric features or iris images that are forged. Indirect attacks are the attacks that are made on the inner modules of the biometric A research [10] presented a hypothesis "genetically undistinguishable irises have texture similarity that is not detected by iris biometrics". Genetically identical irises can be found with twins and both eye irises of same person. However, the similarity between genetically identical irises is not detected by iris biometrics. This provides more security as the biometrics is assumed to be highly secure. Some of the challenging queries with respect to left/right human irises are as presented in Figure 6. Figure 6: Challenging iris queries that causes incorrect responses [10] Though the iris pairs are matching, the responses were that they are unrelated. This is due to the hypothesis taken by the researcher which has been proved. Sometimes it is possible to depend on human experts when iris technology is unable answer correctly [10]. In similar lines a sin [19] made experiments on the hypothesis that "texture has effects on iris recognition". Their experiments proved that over a period of time iris recognition failure is attributed to the effects of texture and found the need for dealing with texture.

Iris Localization in Frontal Eye Images
Iris localization is very important activity in commercial iris recognition systems. However, the validation of method s is limited to laboratory data and not for realistic data. A research [11] proposed an algorithm that proved to be robust with not ideal data which is less constrained. It has operations like localizing outer and inner boundaries of iris, and the process of suppressing specular reflections. It also has regularization of circular boundaries. The results of this research Http://www.granthaalayah.com ©International Journal of Research -GRANTHAALAYAH [87] reveal that the algorithm is robust in presence of eyelids occlusions, eyelashes, hair, contact lens and glasses. The overview of the algorithm is as presented in Figure 7. Figure 7: Overview of the algorithm [11] As can be viewed in Figure 7, it is evident that the given eye image is subjected to preprocessing before applying two phases of iris localization and finally dealing with non-circular boundaries. The experiments conclude that specular reflections very useful in iris recognition, two phase strategy is robust, circular Hough transform can withstand and deal with broken contours, active contours and radial gradients can be used for regularizing inner and outer iris contours [11].
Research in similar lines focusing on gray level intensity [12], radial-gradient operator, and Hough transform for iris localization. In similar fashion, research [13] explored gray level statistics and image projection function for iris recognition. Yet in another experiment a s i n [14] used Hough transform, eccentricity and histogram-bisection for iris localization purposes. In another significant research activity as in [15] focused on non-ideal data for non-circular iris localization by proposing a new localization technique.

Other Approaches to Iris Recognition
Research as in [5] focused ocular biometrics including iris recognition. Ocular biometrics became popular as they are proved to be secure biometric features. They focused on the sclera texture and vasculature patterns for biometric authentication to form an ocular-based recognition system. Biometrics is the science of identifying people based on their behavioural or physical traits such as face, iris, fingerprints and voice. As in paper [17] combined error correction codes and finger prints in order to build an effective biometric system. There is a study [18] focused on a multi-model biometric system using face and iris combination. SVM and feature selection techniques were used in the recognition process. A research [21] proposed an algorithm for iris segmentation using Circular Hough Transform and K-Means algorithm. The experiments were made on iris recognition in unconstrained environments. Similar study was made as in [22] using combination of different approaches for iris recognition. The following Table presents various statistical comparisons of biometric techniques. The various technical issues involved in the recognition of iris can be subdivided into four parts. The first set of issues includes image acquisition. The second step includes segmentation of the iris from the iris image. The third part concerns with feature extraction from the segmented iris image. Finally the fourth part deals with the matching algorithms to match the iris pattern.

Conclusions and Future Work
In this paper, our focus is on iris as biometrics feature for secure authentication and identification of humans uniquely across the globe. Iris is one of the physiological biometric features which are regarded as highly reliable in biometric identification systems. It is used in multimodal biometrics and in combination with cryptography. It is also considered to be most inequitable of facial biometrics. However, it is found that iris localization in influenced by texture. When it is not interpreted properly, commercial iris-based biometric systems provide inaccurate results while identifying humans. Moreover, it is important that iris-based identification systems should work with both ideal and non-ideal iris images otherwise the security will be at stake. This study revealed that iris-based biometric systems tend to provide false results in non-cooperative environments. Another important insight is that iris can be used in mobile communications with smart devices. Cancellable biometrics is useful for robust security in the presence of attacks. There are direct and indirect attacks on multimodal biometrics that need to be overcome. Further research is required in order to see that such attacks cannot break security of systems which are based on biometrics. With these insights in mind, in future, we focus on ATM terminal design using iris recognition in banking domain.