RELATED WORK: TEMPLATE PROTECTION

Here, in this section, we are discussing pioneer techniques in template protection, majorly focusing on template protection using CB. Furthermore, we are also illustrating a comparative analysis between templates transformed via deep learning- and non-deep learning-based methods.

Biometric Encryption

Sahai and Waters in the year 2005 [79] were the first to develop biometric-based encryption systems. Since then, several systems have been developed. In one of the work [34], face data are protected using the Shamir secret sharing key. Here, in the first step, binarised face features are obtained using tokenised pseudo-random numbers. This binarised representation is known as FaceHash. Later FaceHash is protected via the Shamir secret sharing key. In another notable work, Bansal [8] generates the key for the RSA algorithm using a matrix forged by fingerprints. Recently, a method is proposed by the name symmetric keyring encryption [47]. Here, in this method, biometric secret binding is carried out as fuzzy symmetric encryption.

Biometric Cryptosystems

BCs are majorly divided into two types: (i) key generation based and (ii) key binding based. In the case of key generation-based BCs during the enrollment phase, helper data are generated from biometric templates and these helper data are used to generate keys. Here, in this case, generated keys along with helper data are stored in the database. Here, matching is performed by comparing keys stored in the database.

A major difficulty in these technique is to generate keys having high entropy values. The following section mentions some of the notable work in this field.

Pioneer work in this field is carried out by Yevgeniy Dodis [20] in the year 2008. He has proposed a fuzzy extractor for securing fingerprint templates. In another notable work [101], iris features are extracted through 2-D Gabor filters, and further Reed-Solomon error-correcting code is used along with the hash function to generate cipher key. This cipher key is used for encrypting and decrypting the iris features. In Ref. [104], the authors have proposed a fingerprint authentication technique based on a delaunay triangle-based fuzzy extractor. The major advantage of this technique is that it exploits the distinctive properties of delaunay triangulation net to attain robust features and use them further to achieve registration free matching. In Ref. [102], the authors proposed a near equivalent dual-layer structure check (NeDLSC) algorithm based on the minutiae local structure for generating secure fingerprint templates. An online voting system is proposed by [89]. This scheme is based on a fuzzy extractor for providing biometric-based authentication, which is paired with a secret password to provide add-on security to the voter. In another notable work [63], a biometric authentication protocol is proposed based on Kerberos. Here, for the first time, the fuzzy extractor is embedded in the Kerberos scheme. This proposed protocol is resilient against several attacks like man-in- middle and reply attacks. In one of the latest work [64], the authors utilised the Chebyshev polynomial in combination with a fuzzy extractor to protect face datasets. Due to the chaotic properties of the Chebyshev polynomial, it serves as a good candidate for designing cryptosystems.

In the case of key binding-based systems, fuzzy commitment [41] and fuzzy vault [40] are two popular categories. The former one is used to secure biometric templates that can be used as a binary vector-like iris. Here, in the case of fuzzy commitment during the enrollment phase, the binary represented feature vector is XORed with the binary representation of the error-correcting code (obtained from the key) to obtain helper data. While during biometric authentication, feature is XORed with helper data to obtain the error-corrected code, which further generates the key. The latter one is used to secure point set-based biometric features like fingerprint minutiae. Here, during enrollment, the biometric feature point is embedded in a finite field and is evaluated on a polynomial, which is generated by a key. In order to add biometric security points, its polynomials are further mixed with random points. While during authentication, query biometric is used to generate actual polynomial from the stored representation. Table 2.3 depicts some of the popular key binding-based BCs.

Unimodal biometric templates often suffer from inter-class variations and nonuniversality. In such situations, multi-biometric templates come as a rescue measure. Here, two or more modalities of the same person are used to increase the efficacy of the system. A fuzzy vault-based template protection method for fusing fingerprints and palmprints was first proposed by Brindha and Natarajan [10]. Later, in the same year, Nagar [62] proposed a template protection technique that combines the advantages of both fuzzy vault and fuzzy commitment. Here, the biometric features are fused feature-wise. Recently, a multimodal biometric authentication system is proposed that fuses feature vectors from fingerprints and palmprints based on fuzzy vault [88]. The proposed scheme exhibits potential results.

TABLE 2.3

Key Binding-Based Biometric Cryptosystems