Full Length Article
Optimal feature selection-based biometric key management for identity management system: Emotion oriented facial biometric system

https://doi.org/10.1016/j.jvcir.2020.103002Get rights and content

Highlights

  • Biometric key binding enhances security and reliability of authentication.

  • Key binding has been achieved using facial emotions.

  • Hybrid nature inspired algorithms yield improved performance.

  • Neural network based learning for enhanced recognition.

Abstract

Identity management systems with biometric key binding make digital transactions secure and reliable. A novel methodology is proposed to develop an intelligent key management system using facial emotions. Key binding with facial emotions makes use of an intrinsic user specific trait facilitating a more natural computer to human interaction. The proposed system utilizes metaheuristic swarm intelligence based optimization techniques to extract optimal features. The work demonstrates key binding by encrypting an image with a secret key bound to optimal features extracted from facial emotions. Efficiency and correctness of proposed key management is validated by successful decryption at receiving end with any one of the enrolled emotions given as input. Deer Hunting Optimization Algorithm and Chicken Swarm Optimization are merged to select optimal features from facial emotions. The derived algorithm is called Fitness Sorted Deer Hunting Optimization Algorithm with Rooster Update. Seven facial emotions — anger, disgust, fear, happiness, sadness, surprise and neutral are used to extract optimal features from Japanese Female Facial Expressions and Yale Facial datasets to train the neural network. Proposed work achieved better performance results over state-of-art optimization algorithms such as whale optimization algorithm, grey wolf optimization, chicken swarm optimization and deer hunting optimization algorithm. Accuracy of proposed model is 2.2% better than deer hunting optimization algorithm and 12.3% better than chicken swarm optimization for a key length 80.

Introduction

An identity management system (IMS) ensures security and reliability of digital transactions over a network. Digital services are provided to authenticated users who have valid credentials. Authorized levels of interaction amongst users, identity provider and service providers are to be regulated. Central administration, user self-service, role based access control and integrated user management are essential requirements for identity management systems [1]. Identity management increases efficiency and security of access control while decreasing complexity, cost and repetitive tasks. Integration of biometric traits with identity management systems brings in several advantages over username/password or token based authentication systems. Researchers have evinced a lot of interest in identifying novel physiological and behavioural traits for improving efficiency and security of identity management. Multi-modal biometrics is an area of active research to achieve increased efficiency. A significant challenge faced by researchers is in making biometric credential systems reliable and reproducible without sacrificing efficiency and efficacy of detection.

Biometrics has been used in broad range of applications such as e-commerce, physical and electronic access control, background verification and digital rights management [2]. Fusion of crytpographic keys with user specific traits addresses concerns of template security [3]. User specific traits are stored in smart card and utilized for matching features taken as input during validation [4]. Smart cards enable authentication with matching score within defined threshold value and are preferred over Session Initiation Protocol (SIP) servers [5]. The servers ignore user specific traits and depend on smart card for user verification [6], [7].

Conventional user name password or token based authentication systems have limitations. The major issue is that users are required to memorize usernames, passwords or maintain them securely making them susceptible to multiple attacks including dictionary attacks [8]. The solution to ensure secure storage of data involves substitution with cryptographic key [9], [10], [11]. Cryptographic models with smaller keys are easily broken while larger and complicated keys are difficult to remember. Larger keys are required to be stored securely and that in turn leads to threat vulnerability. Distribution of secret key is a challenging proposition in symmetric key cryptography [12], [13], [14].

Combination of biometrics and cryptography is a major area of research interest. Cryptographic keys are required to be precise to the last bit while user traits inherently show fuzzy nature due to intra-user variations. Biometric key release, key generation and key binding are approaches for combining cryptographic keys with user features [15]. Researchers are exploring newer traits with intelligent deep learning algorithms to perform user authentication. Keystroke dynamics and pattern of typing on mobile phones have been analysed with convolutional neural network to identify the user [16]. Deep learning techniques are employed for face recognition on mobile devices and for recognition systems based on finger veins [17], [18].

The major contribution of this work is:This work aims to develop a novel biometric key management system by considering facial emotions and key binding techniques. An environment with enhanced security is built by considering both facial and key features during encryption and decryption. Facial emotion features acquired by extracting Scale Invariant Feature Transform(SIFT) features are computationally complex for machine learning. A novel metaheuristic algorithm, namely, Fitness Sorted Deer Hunting Optimization Algorithm with Rooster Update (FDHOA-RU) has been developed for selecting optimal features from SIFT keypoints. The proposed FDHOA-RU ensures selection of optimal features, thereby, providing improved fitness resulting in higher efficiency during training. FDHOA-RU algorithm has been developed by integrating properties of Deer Hunting Optimization Algorithm (DHOA) and Chicken Swarm Optimization (CSO) algorithms. The FDHOA-RU algorithm circumvents problem of premature convergence by including updates from CSO algorithm. Key extraction from an input image is carried out by Double Random Phase Encoding (DRPE), Bose–Chaudhuri–Hocquenghem (BCH) encoding, shuffling and RSA encryption. The proposed approach has made key generation more specific and reduced the probability of information stealing. Facial emotions have been considered for neural network based training during encryption and decryption process along with the key, thus providing additional support to the biometric recognition system.

The paper is structured in following manner: An overview of recognition of facial emotions and challenges in binding cryptographic keys with user traits is provided in Section 2 with state-of-the-art works in key agreement protocol. Section 3 describes proposed key management for IMS with novel key binding procedure. Key encryption process is explained in Section 4. Selection of optimal features from facial emotions and training of neural network is described in Section 5. The process of image encryption and decryption with binding of user specific key is covered in Section 6. Section 7 evaluates the approach adopted and compares results obtained with other state of art metaheuristics algorithms. Finally, conclusions are drawn in Section 8.

Section snippets

Facial emotions

Facial emotions introduce natural aspects to conventional face recognition algorithms. Emotion feature set add additional dimension for enhanced versatility to detection accuracy [27]. Local Binary Pattern (LBP), Gabor filter, LBP with Three Orthogonal Planes, Graphics-processing based Active Shape Model (GASM) and Scale Invariant Feature Transform (SIFT) are some feature extraction methods explored to extract emotion feature set [28], [29], [30]. Research has been carried out in recent past

Proposed architecture

Multiple research have been carried out for binding cryptographic key with biometrics. Drawbacks of implemented methodologies of earlier studies include failure in secure transfer of encryption key and difficulty for user to store key in a secure manner. Diagrammatic representation of proposed key management is shown in Fig. 1, depicting both encryption and decryption process. The methodology introduces facial emotions based key binding system to encrypt an image with seven different emotions

BCH encoding and key shuffling

BCH coder encodes the encryption key. There exists a t-error correcting code with parameters t<2r1, s=2r1, swrt, tr2t+1 for any positive integer r. The linear cyclic code has ability to correct upto t random across (2r11) bit positions. BCH encoded key is split into blocks to perform shuffling operation. Each block of encryption key (BCH_EK) is designated by a number and the blocks are shuffled by user specific shuffling key Shuf_EK. The shuffled key is indicated by BCH_Shuf_EK.

SIFT-based feature extraction from facial emotions

SIFT identifies salient and stable feature points used for proposed key management. SIFT is image scaling and rotation invariant and moderately invariant to alterations in illumination allowing for alteration in occurrence of occlusion, noise, or clutter. SIFT algorithm is classified into four phases: Scale Space Extrema Detection, Key point Localization, Orientation Assignment and Key point Descriptor. The four phases are described briefly.

  • (a)

    Scale Space Extrema Detection: SIFT framework

Solution encoding and objective model

The proposed FDHOA-RU algorithm is used for both feature selection as well as training of neural network. Once SIFT features FeaneSIFT are given, optimal SIFT features FeaneSIFT are extracted. The weight function E˜lNN of neural network is optimized by FDHOA-RU for more accurate detection of authorized user. Encoding solution for optimal feature selection and classification is shown in Fig. 2.

Optimally generated features are represented as FeaneSIFT, in which ne=1,2,,Nfe and Nfe are

Experimental setup

The optimal feature selection-based biometric key management for IMS using facial emotions has been implemented in MATLAB 2018a installed on a PC with Windows 10 OS, 8 GB RAM and 64-bit operating system. Hardware platform for proposed work has not been implemented and has been left as future work. Facial biometric related to seven different emotions — normal, smile, sad, surprise, anger, fear, and disgust has been used. Following datasets have been used for experiments.

  • (a)

    Japanese Female Facial

Conclusion

A new IMS protocol has been demonstrated involving key extraction for encryption, feature extraction of different emotions, optimal feature selection, bit stream generation, and decryption process. The image is initially subjected to DRPE, wherein the key is created using CRPM and EFRT. Binary conversion, BCH encoding, shuffling and RSA encryption are performed on generated key, followed by extraction of features with different emotions of user facial biometric and optimal feature selection.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (43)

  • TorresJ. et al.

    A survey on identity management for the future network

    IEEE Commun. Surv. Tutor.

    (2013)
  • EastwoodS.C. et al.

    Biometric-enabled authentication machines: A survey of open-set real-world applications

    IEEE Trans. Hum.-Mach. Syst.

    (2016)
  • LiC. et al.

    A security-enhanced alignment-free fuzzy vault-based fingerprint cryptosystem using pair-polar minutiae structures

    IEEE Trans. Inf. Forensics Secur.

    (2016)
  • JainA.K. et al.

    Biometric authentication: System security and user privacy

    Computer

    (2012)
  • ShengW. et al.

    A biometric key generation method based on semisupervised data clustering

    IEEE Trans. Syst. Man Cybern.

    (2015)
  • LiC. et al.

    A new biocryptosystem-oriented security analysis framework and implementation of multibiometric cryptosystems based on decision level fusion

    IEEE Trans. Inf. Forensics Secur.

    (2015)
  • LiS. et al.

    Fingerprint combination for privacy protection

    IEEE Trans. Inf. Forensics Secur.

    (2013)
  • AndalibA.S. et al.

    A novel key generation scheme for biometric cryptosystems using fingerprint minutiae

  • NagarA. et al.

    Multibiometric cryptosystems based on feature-level fusion

    IEEE Trans. Inf. Forensics Secur.

    (2012)
  • NguyenT.H. et al.

    Performance and security-enhanced fuzzy vault scheme based on ridge features for distorted fingerprints

    IET Biom.

    (2015)
  • ZhangL. et al.

    Multiple-image encryption mechanism based on ghost imaging and public key cryptography

    IEEE Photonics J.

    (2019)
  • Cited by (1)

    This paper has been recommended for acceptance by Zicheng Liu.

    View full text