With the large development of multimedia applications over Internet and mobile networks, it has become a must to introduce more secure access methods to serve users. Passwords and Personal Identification Numbers (PINs) have been used as traditional access methods to the Internet and the other networks. However, the keys and passwords are exposed to being stolen or even forgotten [1]. The good alternative to avoid the limitations of keys and passwords is to use biometrics. Biometric-based access methods do not require to remember passwords or save keys, as biometrics are always accompanying persons [2]. The main problem with the utilization of biometrics as access tools is the need to create databases to store the user biometrics. These databases may be vulnerable to attacks. A possible solution to eliminate the effect of attacks is to use cancellable biometrics. The idea of cancellable biometrics is to use distorted or encrypted versions of the biometric templates rather than the original biometric templates in the recognition process. The intended distortion is created through non-invertible transforms. On the other hand, strong encryption algorithms can be used to generate encrypted biometric templates to be stored in the databases for the recognition task. A good property that should be achieved in cancellable biometric systems is the pattern diversity with unlinkability between patterns to use different patterns for different applications [3]. These properties can be guaranteed through changing the initial parameters or keys of the used transforms or encryption algorithms [4].

In another classification of cancellable biometric schemes, they can be categorized into helper-data-based schemes and cancellable biometric transformation schemes. In the helper-data-based schemes, some auxiliary data are integrated with the original templates. These data make it possible to restore or renew the key or initial parameters during the authentication process. In [5], both feature extraction and a fuzzy vault algorithm were used for cancellable fingerprint recognition. In [6], a cancellable fingerprint recognition scheme was built through the utilization of an Error Correction Code (ECC) [6]. The coding schemes give better representations of the obtained features that are difficult to be reversed.

In [7–9], the robustness of biometric encryption has been improved by applying a hybrid biometric cryptosystem, which combines two or more template protection techniques in order to enhance the privacy of users. Traditional encryption algorithms can be used for encrypting biometric templates, but they have some difficulties in implementation, and they require much time. On the other hand, chaotic-based algorithms are simple in implementation and may be implemented with permutation operations only [10]. That is why chaotic encryption is more recommended for biometric encryption.

In [11], both piecewise and logistic maps have been combined for biometric encryption, leading to high unpredictability in a highly-complicated and long-length chaotic pseudo-random number generator. In [12], the chaotic behavior of the logistic map is used for building a projection matrix of the individual biometric. This scheme is non-invertible, which fulfills the revocability, diversity, and robustness to attacks. The authors of [13,14] presented a hybrid cryptosystem based on a 3D chaotic map that improves security and efficiency in the presence of various attacks. In [15], chaotic encryption has been used to generate cancellable iris codes. A modification of the logistic map has been incorporated to enlarge the keyspace, and hence privacy is increased. In real-time applications, many digital cryptographic algorithms based on chaotic maps have been introduced with the corresponding hardware to protect the transferred multimedia data from rapid attacks [16]. Prototyping of an algorithm with FPGA implementation enhances the credibility that the algorithm will work practically [17]. FPGA prototypes are used to test certain functions in real time. Furthermore, with the high speed of FPGA prototypes, a larger amount of data can be used, potentially revealing bugs that are not found in simulation models. Hardware implementation is increasingly important due to the computational complexity of the algorithms and the high throughput requirements [18].

Field Programmable Gate Arrays (FPGAs) can be used in biometric applications, particularly with embedded applications in which latency and power are critical. However, running an encryption algorithm on FPGA mostly results in frustration as most encryption algorithms are designed for serial processors. In order to effectively optimize the implementation with FPGA, it is typically important to convert the algorithm to a new one compatible with hardware design at both operational and application levels. In [19], a modern encryption scheme for digital images based on a chaotic system was presented and implemented on FPGA. The resulting design is robust and effective.

Many researchers discussed the problem of cancellable biometric recognition systems [1–22], where all of them have drawbacks in investigating the attack analysis and a shortage in using more recent datasets. In addition, the round errors on the hardware systems give lower evaluation metrics. This paper introduces an improved hybrid encryption algorithm with its FPGA implementation to create an efficient, more protected, cancellable biometric recognition system that is immune to fraudsters. The main contributions of this work are as follows:

a) Proposal of an efficient cancellable biometric recognition system using a 3D chaos cryptosystem.

The symbols and notation utilized throughout the whole paper are summarized in Tab. 1. The rest of this work is arranged as follows. An overview of the cancellable biometric recognition system and its implementation with the FPGA is provided in Section 2. A description of the authentication quality assessment metrics used is given in Section 3. Section 4 presents the results of simulation and cryptanalysis with a detailed performance comparison. Finally, Section 5 summarizes the concluding remarks.

Fig. 1 shows the proposed hybrid cryptosystem based on 3D chaotic sequences. The framework of the suggested cryptosystem is represented in two cascaded stages: confusion and diffusion. In the confusion stage, the positions of pixels of the original biometric image are shuffled in row/column rotation. On the other hand, in the diffusion stage, the XOR operation is applied to change the values of the pixels.

FPGA prototyping of an algorithm increases the confidence that it will function properly in practice. For example, we use FPGA prototypes to implement functions and run simulation scenarios at high speed. In addition, because FPGA prototypes run faster, it is possible to use larger datasets, potentially exposing bugs that a simulation model would not reveal.

The suggested cryptosystem is developed with built-in MATLAB Simulink blocks as well as certain user-defined features. All data utilized are described as fixed point data, and each block should even provide a Hardware Description Language (HDL) coder. In the simulation, the discrete-no continuous state solver and fixed-step solution types are chosen.

The schematic diagram of the proposed cryptosystem model is depicted in Figs. 2–6. The model, which is made up of four main subsystems, is revealed. The schematic diagram for producing a 3D chaotic sequence is shown in Fig. 2. Fig. 3 shows the equalization of the histogram of the obtained chaotic sequence. Fig. 4 displays the stage of confusion. Fig. 5 depicts the diffusion stage.

The HDL workflow advisor tool generates the HDL code in the second step, as shown in Fig. 6. It facilitates the Register-Transfer Level (RTL) code and the test-bench generation from the proposed algorithm, performs synthesis tasks by invoking a supported third-party synthesis tool, and annotates critical path information back to the system. It also sets a particular workflow and guides the designer through the tasks necessary for full deployment. Each task performs a distinct step of the workflow. Feedback is adopted on the results of each task. If the task fails, the HDL workflow advisor tool provides information on how to modify the model to complete the task. The HDL code is then simulated to produce the RTL with the Xilinx generator, providing a well-suited hardware design system. The developed model is then synthesized using ISE software with Xilinx FPGA technology, as shown in Figs. 7 and 8.

Biometric authentication is the process that ensures whether a particular template is or is not the same as one of the templates stored. A new biometric template is compared during the verification phase to the existing ones to confirm matching with the stored templates. The effectiveness of the cancellable biometric recognition system is determined by correlation analysis and Receiver Operating Characteristic (ROC) curve. If the tested individual correlation score exceeds a predefined level, access is granted. Generally, the correlation score of an authorized individual should frequently be larger than that of an unauthorized individual [24,25].

The classification process of genuine and imposter users is based on a statistical concept. Two distributions are created for genuine and imposter correlation scores. The intersection point of both distributions reflects the efficiency of the system through the so-called Equal Error Rate (EER). These two distributions can be used to estimate the ROC curve of the system. The ideal ROC curve that reflects a success rate of 100% has a corner at the top-left point. The roll-off shape below the top left corner represents the deviation from ideality.

This paper investigated a modern biometric-based security application, namely cancellable biometric recognition. The 3D chaotic logistic map is exploited for this purpose. It is used to scramble biometric traits, including faces and fingerprints, to generate the cancellable biometric traits. Through the proposed cryptosystem, the user privacy is maintained. In addition, the distinguishability of biometric traits is guaranteed. The proposed cancellable biometric recognition has been evaluated statistically based on EER and area under ROC curves (AROC). Three datasets have been utilized in the performance evaluation of the proposed system. Simulation results revealed EER values of 6.2460 × 10⁻¹³ and average AROC values of 0.9998 for the proposed cancellable biometric recognition system.