The issue of cybersecurity is one of the important fields which is involved in different research trends. Biometric security is one of these trends which is involved in several applications such as access control systems and online identity verification. The protection of human biometrics can be performed using both bi-directional and unidirectional encryption. The unidirectional encryption is carried out based on cancelable biometric techniques. This paper proposes a cancelable biometric system based on image composition, deep dream, and hashing techniques. The objective of the proposed system is to generate visual and text cancelable biometrics. The visual cancelable templates are generated using image composition and deep dream, while the text templates are generated using SHA hashing techniques. The proposed system is validated by multi-biometric inputs including iris, palm, face, and fingerprint biometrics. In addition, it is evaluated in both visual and text forms. The simulation results reveal that the proposed system appears a superior performance among the works which handle this problem.
Research Article
Compression-Based Cancelable Multi-Biometric System
https://doi.org/10.21203/rs.3.rs-2241969/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Cancelable biometrics
Deep learning
Deep-Dream
Image composition
and DCT
Generally, security is a must for all organizations and institutions. Traditional security tools comprise passwords and Personal Identification Numbers (PINs) [1]. Unfortunately, these tools require that the user must remember large numbers or text strings. The limitations of these traditional tools to information system security are as follows [2]:
They recognize some of the characters linked with the individual but cannot figure out who invented them.
They are prone to be misplaced or stolen.
They are easily evaded and hackable.
They are incorrect.
In such a case, the system's security is jeopardized. Biometrics are very suitable for scenarios requiring dependability. They always accompany the person and do not necessitate any intervention from him. By more closely tying identity to the individual, biometrics can be helpful. It is a technology that makes an effort to identify someone based on personal traits and offers a high level of accuracy in identifying a person. [2, 3].
The authors of [4] describe biometrics as the different physical or logical features or attributes of the human body that are measured. The authors of [5] describe biometrics such as the fingerprint, iris, face, and DNA as methods of identifying individuals. Biometrics provides a rapid, discrete, and unobtrusive option for access control and identity verification. The authors of [6] talk about traditional and conventional biometric identification approaches.
Physical and behavioral biometrics are the two most often-used types of biometrics [7, 8]. In contrast to physical biometrics, which is body components, behavioral biometrics are activities that a person conducts. A person's unique fragrance, the shape of their ear, and even the way they talk are all unusual biometrics that could be used in the future. The primary physical biometrics that is measured are fingerprints, retina, hand engineering, iris scan, facial position, or recognition. The most frequent behavioral biometrics researched are voice recognition, signature, typing style, and walking. However, different biometrics are required for different purposes, therefore there is no superior or worse biometric technology [9].
The process of biometric acquisition requires a sensing process, interfacing, and signal processing unit for enhancement and feature extraction. Moreover, a data storage unit is required to save the features of authorized users. Finally, an output interface should exist in the system [10]. A common biometric identification method collects samples of a certain biometric and extracts and records the unique properties directly in the database. During the verification step, a newly extracted biometric characteristic is compared to previously stored features to determine whether or not it matches one of the previously saved templates. Whereas biometric authentication refers to the process of determining whether or not a specific individual matches the stored template.
The conventional biometric identification and verification process includes several security and privacy issues [11]. Original biometric templates, unlike passwords or encryption keys, are easily stolen; as a result, these templates cannot be canceled or amended, rendering biometrics unsafe. The authors of [12] raised an additional risk that can be addressed in systems that use traditional identity or authentication. This is known as cross-matching (diversity) and cross-application invariance, and any apps and systems that rely on users' biometrics can be easily targeted if biometrics database templates are made public, allowing for easy user tracking. The authors of [13] illustrated the importance of introducing a security system to unprotected templates, which expose authenticated users' information to theft and dissemination, to protect the stored template from previous challenges.
Cancelable biometrics is known as the intentional and recurrent distortion of a biometric signal based on a chosen transform [14]. Before storage, the original biometric or extracted features are modified by a one-way function in these systems. According to the above explanation, cancelable biometrics approaches improve diversity and unlinkability challenges. Different transforms can be used for different applications to get rid of cross-matching. Furthermore, the revocable cancelable biometrics approaches can be changed or canceled with new templates based on these biometric data [14, 15].
Non-invertible transform [15, 16], invertible transform [17, 18], and biometric salting [19, 20] are all examples of cancelable methods. A user-specific key is utilized for the generation of the cancelable template [19]. Hybrid biometric cryptosystems [20], on the other hand, combine more than two template protection techniques into a single biometric cryptosystem.
The majority of chaos-based biometric template encryption techniques rely on concepts called diffusion and confusion in chaotic encryption systems. The original biometric template is aesthetically cluttered and unrecognizable as a result of the confusing approach's shifting of the pixel positions. A diffusion technique modifies the grayscale values of pixels to change the statistical characteristics of a biometric template. The performance, durability, and security of encryption are improved when the confusion and dispersion stages are combined.
A variety of chaotic-based encryption algorithms have recently been proposed for the development of successful cancelable biometrics. Chaotic signals are suitable for encryption due to their sensitivity to the initial state, noise-like behavior, and ergodicity. Because they provide a good balance of processing power, high security, complexity, speed, and all of the above, chaos-based encryption algorithms are ideal for use in cancelable biometrics.
Position permutation and value transformation are two types of cancelable biometric-based encryption techniques [17–20]. The pixel position of the biometric template is relocated to a different location without altering the pixel value of the original biometric template in position permutation approaches. Value transformation techniques swap out one-pixel value for another without affecting the location. The XOR operation is one of the most widely used methods for transforming values and is used to achieve linear independence between two or more variables. The theory underlying XOR encryption is that without knowing the initial values of one of the two parameters, it is impossible to reverse the operation. To improve the security performance of the cancelable biometric-based encryption technology, the concept of shuffling the positions of pixels in the plain-biometric template and then changing the grey values of the shuffled pixels is used.
As a result, this paper introduces a practical implementation of a hybrid encryption technique based on composition and the deep-dream algorithm for the development of a reliable and robust cancelable biometrics identification system (face, iris, palmprint, and fingerprint). By combining numerous rotations and operations to generate the key to the encrypted biometrics templates, the proposed security-based cancelable biometrics system improves the robustness of biometrics recognition systems.
The remainder of this work is organized as shown below. The second section contains the related work. The third section presented the proposed cancelable biometric recognition system. The fourth section explains the authentication quality evaluation measures and provides simulation results; both visual and statistical evaluations. The fifth section contains the conclusion.
Generally, biometric systems depend on a single biometric trait. Although this strategy is simple, the need for a more secure system is a must, nowadays. The trend of multi-biometric systems is now emerging as it adds more security to the system.
Two or more biometric traits, such as the fingerprint, palm print, iris, or face, are combined in a multimodal biometric system. A single sensor's input may be routed through many algorithms, or a single biometric characteristic or several biometric attributes may be captured using many sensors. Multimodal systems, therefore, assist in accelerating the rate of picture recognition. Several multimodal biometric systems with two or more qualities have been introduced. Every suggested solution made use of a different feature extraction and fusion technique. Here, we'll discuss a few of them. Saha et al. [21] achieved 98.37% accuracy using feature-level integration of retina and iris recognition. Through the use of principal component analysis (PCA), they were able to significantly lessen the "curse of dimensionality" barrier that had previously plagued their line of work.
The authors of [22] proposed an approach for multi-biometric recognition using ECG, CNN, and multi-support vector machine (MSVM). It adopts a decision fusion strategy. Using techniques like Gabor, Radon, Ridgelet, and Radon-Gabor, features from the face and palm prints were recovered. These features were then combined and categorized using SVM and SSIM. [23]. The accuracy of the fused system increased from 92.5% for the face to 98.0% for the palm print.
Another strategy for biometric identification combines the ear and finger knuckle print because the ear biometric is stable from birth and thus unique [24]. The authors propose combining the physiological and behavioral aspects of the face, specifically the face and facial expression, to lessen the susceptibility of unimodal biometric authentication [25]. The technology uses facial expressions to identify an authorized user.
Bharadi et al. [26] described another technique that incorporates physiological and behavioral aspects. They depend on the iris and pressure signature. A KNN classifier has been utilized. It achieved a 90.18% according to a score-level fusion strategy. They worked on two subsets of the NIST-BSSR1 dataset [27, 28]. Hezil et al. [29] used feature-level fusion to fuse palm print and ear databases. On the IIT Delhi palm print and IIT Delhi-2 ear databases, they obtained 100% accuracy. With the use of fractional firefly (FFF) optimization, a new multimodal biometric system that combines images of the finger vein and knuckles through feature-level fusion has been created. It is feasible to achieve 96% higher accuracy [30].
The authors of [31] presented a breakthrough paradigm for multimodal biometric technology by merging a person's facial and iris traits. On huge datasets like SDUMLA-HMT, CASIA V1.0, MMU1, and FERET, experiments were conducted using a CNN-based architecture, a deep belief network (DBN), and a SoftMax classifier. The authors presented a database of more than 100 children's faces, fingerprints, and iris modalities (ages 18 months to 4 years), with photos taken in separate sessions [32].
Murakami et al. in [33] developed a cancelable biometric method based on hashing and Gaussian random vectors (GRV). The proposed architecture stores and authenticates using variations of the initial biometric rather than the initial biometric itself. Using face and palmprint biometrics, the study's findings on exhibitive, non-invertibility, and uniqueness were validated. As part of their contribution, Yang et al. developed A CBS based on a random slope approach [34]. According to this method, a secure, revocable system with irrevocable canceled templates is produced. The technique's two stages are random slope versions one (RS-V1) and two (RS-V2), which produce secure designs with dimensions cut in half. To validate the methodology, several biometric modalities were used, including finger veins, warm and visible faces, palm veins, and fingerprints.
The authors of [35] described a multi-biometric system based on finger vein and fingerprint modalities. The primary goal of this research was to meet the protection and revocability requirements for produced templates that include a fingerprint feature set based on minutiae and a finger-vein feature set based on an image. Following that, a fusion operation is performed on both sets, and the proposed approach is evaluated and analyzed before matching and security. After that, the proposed method was compared to the original Partial Discrete Fourier Transform (P-DFT). AK Yadav [36] a deep learning system for human recognition employing biometric modalities such as Iris, Fingerprint, and handwritten signature was proposed. Convolutional Neural Networks (CNNs) serves as the system's core, extracting features and categorizing images with a SoftMax classifier. The system was created by integrating three CNN models: one for each fingerprint recognition, iris recognition, and handwritten signature recognition. The CNN model was built upon the VGG-32 algorithm.
The multi-biometric strategy has been adopted in several studies. Bouzouina and Hamami [37] worked on both face and iris features. They used the SVM classifier for user verification. They achieved 98.8% accuracy. Hezil and Boukrouche [38] worked on ear and palm print features. In a different study, Veluchamy and Karl Marx [39] worked on finger vein and knuckle characteristics. Their system merged the traits at the feature level. They used the K-SVM algorithm to achieve 96% accuracy.
Chanukya et al. [40] recently used deep learning to create a multimodal biometric system capable of recognizing an individual based on fingerprint and ear images. To extract form features from traits, the system devised the modified region growth approach, and to extract texture features, it devised the local Gabor XOR pattern. The proposed method had a 97.33% accuracy rate. Furthermore, Ammour et al. [41] suggested a feature extraction method for multimodal biometric systems based on the face and iris. To extract iris characteristics, a multi-resolution 2D Log-Gabor filter was used. The facial characteristics were recovered using singular spectrum analysis and normal inverse Gaussian. The classification was done via fuzzy k-nearest neighbor (KNN). Score fusion and decision fusion were used to achieve feature fusion.
Abdellatef et al. [42] worked on different areas of the face to generate cancelable biometrics. They used a feature fusion strategy with a deep CNN to create the templates. Abdellatef et al. [43] another modification based on orientated rotated brief (ORB), histogram of oriented gradients (HOG), local binary patterns (LBPs), scale-invariant feature transform (SIFT), and accelerated robust features, are utilized to extract the hand-crafted features (SURF).
A deep learning system for human recognition based on finger vein, iris, and face biometric modalities was introduced by Alay N. et al. [44]. CNNs are the foundation of the system; they gather information from photos and categorize them using a SoftMax classifier. Three CNN models were combined to create the system which is the iris, face, and vein in the finger. The CNN model was built using VGG-16, the Adam optimization method, and a loss function based on categorical cross-entropy.
Based on picture composition and deep dream, this research offers a Multi-Biometric Cancelable System (MBCS). The proposed system's goal is to build a cancelable template that reflects an encrypted form of the input biometric photos, as shown in Fig. 1. This system receives images of the human iris, fingerprint, palm, and face as inputs. The inputs are fed into a composition module that employs DCT transformations and rotations. The created composite images are loaded into the deep-dream module, which generates the encrypted (cancelable) templates. The deep-dream module proposed is based on the inception method. This study also presents a text form of the cancelable templates. This form is constructed using SHA-224 and SHA-256 as hashing algorithms.
3.1 COMPRESSION USING DISCRETE COSINE TRANSFORM (DCT)
The suggested MBCS's main principle is to compress and encrypt four biometric images into a single biometric image. To propose-relate pixels, DCT transforms images from the time-space domain to the frequency domain. Keeping only, the first quartile of the DCT for each image aids in image segmentation. After rotating these quartiles, they are organized in a new DCT plane comprising information from all biometrics.
Figure 2 depicts real-time compression, with the first image representing the face, the second image representing the fingerprint, the third image representing the palmprint, and the fourth image representing the iris. The resulting new DCT plane is inverted to the time domain to obtain a diverse biometric image. The diversified image is then encrypted with deep-dream for increased security concerns.
3.2 DEEP-DREAM ALGORITHM
The second component of the proposed MBCS, as shown in Fig. 1 and detailed below, is the so-called deep-dream method [20], which reduces losses associated with the convolutional layers of our deep learning architecture.
This is because CNN performs digital filtering operations on input images, with each filter extracting features from the images. Because a CNN network is composed of numerous layers, each layer contains a specific number of filters chosen by a discovered design. As a result, a layer with, \({{N}}_{{l}}\) distinct filters would have,\({{N}}_{{l}}\), feature maps of size, \({{M}}_{{l}}\), where, \({{M}}_{{l}}\) is the height times the breadth of the feature map. Furthermore, the responses in layer l can be stored in a matrix, \({{F}}^{{l}}\) ∈ \({{R}}^{{{N}}_{{l}}\times {{M}}_{{l}}}\), where, \({{F}}_{{i}{j}}^{{l}}\) represents the activation of the \({{i}}^{{t}{h}}\)filter in layer l at position j.
Consequently, the new value of any pixel \({p_{new}}\) is estimated by taking the summation of all of the old surrounding pixels multiplied by the applied filter elements , which is formalized in (1).
$${p_{new}}{\text{=}}~\sum\limits_{{i \in S}} {{p_i}{w_i}}$$ (1)
If, \(\tilde {p}\) and \(\tilde {x}\) are the original and produced images, respectively, and, \({{P}}^{{l}}\) and \({{F}}^{{l}}\) are their respective feature representations in layer l, then the squared-error loss between both feature representations are defined as follows.
$${\text{L}}\left( {\tilde {p},\tilde {x},l} \right)=\frac{1}{2}~{\sum\limits_{{i,l}} {\left( {F_{{ij}}^{l} - P_{{ij}}^{l}} \right)} ^2}$$ (2)
then using the following formula, the differentiation of this loss w.r.t to layer l activations is calculated:
$$\frac{{\partial {\text{L}}}}{{\partial F_{{ij}}^{l}}}~{\text{=}}\left\{ \begin{gathered} F_{{ij}}^{l} - P_{{ij}}^{l}~~~~~~~~if~~F_{{ij}}^{l}>0~ \hfill \\ 0~~~~~~~~~~~~~~~~~~~if~~F_{{ij}}^{l}<0~ \hfill \\ \end{gathered} \right.$$ (3)
3.3 IMPLEMENTATION OF THE PROPOSED DEEP-DREAM ALGORITHM
Running the deep-dream algorithm (DDA) on a Convnet pre-trained model on ImageNet, as described in [20], necessitates access to a large number of Convnets, including VGG16, VGG19, Xception, ResNet 50, and others. While DDA can be done with any of them, the religious circle chosen usually changes impressions because different convnet models produce different academic highlights. In practice, Inception is known to produce attractive deep dreams [20]. As a result, we employ the Inception V3 model for the proposed MBCS scheme's DDA unit. Figure 3 depicts the schematics of the Inception V3 model.
The loss value is then computed, which is the quantity that must be maximized throughout the gradient-ascent process. The goal of filter visualization is to maximize the value of a specific filter in a specific layer. In this case, the activation of filters is maximized simultaneously in the number of layers. The method is intended to maximize a weighted sum of the L2 norms of a series of high-level layer activations. The precise layer collection used (as well as their contribution to the overall loss) has a significant impact on the images produced. As a result, these settings must be easily modifiable. Lower layers produce geometric patterns, and higher layers produce pictures that can be identified by ImageNet. It will begin with a fairly arbitrary arrangement of four layers, but it will undoubtedly need to investigate many different configurations afterward.
Furthermore, the maximization process is applied to the convnet layers' loss gradients, with three key parameters controlling the gradient ascent process. maximum loss, Lmax, gradient step S, and the number of iterations I. At iteration i, the gradient ascent X for loss gradients of layer l can be calculated using (4).
$${\text{X}}~{\text{=}}~\sum\limits_{{i,l}} {S \times {L_{i,l}}} ~~~~~~~~~if~X<{L_{\hbox{max} }}$$ (4)
Finally, for image processing, a set of scales (octaves) is defined; these octaves are generated by dream rein junctions at each layer. Using the rein junction process, the input image is upscaled at each layer and made more cancellable. Each consecutive scale is 1.40 times larger than the preceding one, therefore the image is predicted to be 40% larger than its current size. Starting with a small image and gradually scaling it up, as seen in Fig. 4.
Gradient ascent is conducted for each subsequent scale, from smallest to largest, to maximize the previously defined loss. The generated image is upscaled by 40% after each gradient ascent run. To prevent losing more image details with each scale-up (resulting in increasingly pixelated images or blurry). It was necessary to reinject the lost information back into the image after each scaleup, which was accomplished because it was known what the original image should look like at the bigger scale. Given two image sizes, IS and IL, the difference between their original images resized to size, IL, and IS, quantifies the information lost when transitioning from IS to IL.
This study proposes an MBCS that can generate both visual and text templates. The suggested system is based on image composition and deep-dream. The proposed composition strategy is based on the DCT technique, but the proposed deep-dream module is based on CNN technology from the start. Furthermore, the bio-hash templates proposed to use the SHA-512 hashing technique. Furthermore, fingerprints, palms, iris, and facial pictures are used in the proposed system. Twenty human biometric images are included in each biometric modality. Figures 5(a-d) show the biometric images, whereas Fig. 6 shows the cancelable patterns that resulted, which are used in this work.
4.1 VISUAL EVALUATION
The proposed MBCS is evaluated using a variety of assessment markers, including visual and statistical evaluation. The visual evaluation consists of the histogram, actual and impostor correlation scores, and the Receiver Operating Characteristics (ROC) curve.
4.1.1 HISTOGRAM ANALYSIS
Histograms are a simple way to visualize the intensity distribution across pixels in a photograph [45]. A good cancelable scheme should provide templates with the same distribution. Furthermore, such a plot has been discovered to be a symptom of the scheme's emptiness or noise permeation [46], which serves as validation of the scheme's ability to resist statistics-based attacks. This study presents a cancelable multi-biometric system based on deep fusion and style transfer. Figures 7(a-d) show histograms for fingerprint, palm, iris, and face biometrics, respectively, while Fig. 7(e) shows histograms for cancelable patterns.
4.1.2 CORRELATION ANALYSIS
The correlation coefficient is a different way of representing cancelable discombobulation systems. In correlation analysis, the correlation coefficient of neighboring pixels Rx,y is used to identify the relationship between a biometric picture input and its resulting cancelable template. The pristine biometric input's surrounding pixels should have values close to unity, whereas the cancelable templates should have values closer to zero [47].
We calculate Rxy using (5) and report plots of Rx,y for auto-correlation and cross-correlation in Figs. 8(a) and (b), respectively.
$${R_{xy}}=\frac{{\frac{1}{N}\sum\limits_{{i=1}}^{N} {({x_i} - \bar {x})({y_i} - \bar {y})} }}{{{\sigma _x}{\sigma _y}}}$$ (5)
where N refers to the number of pixels, x refers to the encrypted template stored in the database, y refers to the new subject encrypted template, \(\stackrel{-}{x}\) refers to the average of the values of the x-variable, \(\stackrel{-}{y}\) refers to the average of the values of the y-variable, \({x}_{i}\) refer to the values of x-variable in a sample, \({y}_{i}\) refer to the values of the y-variable in a sample, and \({\sigma }_{x}\) \({\sigma }_{y}\) is the multiplication of both x & y standard deviations. Figure 8(c) presents the score of correlation for original and generated images, while Fig. 8(d) illustrates the ROC curve.
4.2 STATISTICAL EVALUATION
This paper presents the qualitative, as well as quantitative evaluation indicators, which are included in this classification. The generated cancelable template must receive a quality rating from the qualitative assessment, and the generated template must receive a correlation, also the Signal-to-Noise Ratio (SNR) statistic from the quantitative evaluation.
4.2.1 QUANTITATIVE ANALYSIS
The suggested MBCS method is evaluated using three quantitative criteria: Number of Pixel Changes Rate (NPCR), Unified Average Changing Intensity (UACI), and Peak Signal-to-Noise Ratio (PSNR).
The NPCR, UACI, and PSNR for the given images I1 and I2 are calculated using (6), (7), and (8), respectively, M and N are the width and height of the image.
$$NPCR(\% )=\frac{1}{{M \times N \times 3}}\sum\limits_{{i=1}}^{M} {\sum\limits_{{j=1}}^{N} {\sum\limits_{{k=1}}^{3} S (i,j,k) \times 100} }$$ (6)
where,
where IMAX is the maximum value of the pixel and MSE refers to mean square error which is defined as:
$$MSE=\frac{1}{{M \times N}}\sum\limits_{{i=0}}^{{M - 1}} {\sum\limits_{{j=0}}^{{N - 1}} {\left[ {{I_2}(i,j) - {I_1}(i,j)} \right]} }$$ (10)
4.2.2 QUALITATIVE ANALYSIS
Qualitative evaluation includes using quality measures to analyze the plan's performance. To evaluate the developed cancelable template, the performance of the recommended MBCS technique is measured in terms of the Universal Image Quality Index (UIQ) and Spectral Distribution (SD).
4.2.2.1 SPECTRAL DISTORTION
SD provides a qualitative measure of the similarity of spectral information between two pictures [48]. SD values are widely regarded as indicating similarity between evaluated images [48]. SD is defined mathematically as:
$$SD=\frac{{\sum\limits_{{i=1}}^{M} {\sum\limits_{{j=1}}^{N} {\left| {x(i,j) - y(i,j)} \right|} } }}{{M \times N}}$$ (11)
where \(M \times N\) refers to the total number of pixels in an image, \(x(i,j)\) is the original image, and \(y(i,j)\) is the encrypted image.
4.2.2.2 UNIVERSAL IMAGE QUALITY INDEX
The UIQ is an additional metric for evaluating structural concordance between two images [49, 50]. UIQ, whose values range from − 1 to 1, is calculated mathematically using (12), with values closer to 1 indicating better congruence between pictures [47].
$$UQI(i,j)=\frac{{Co{v_{ij}}}}{{{\sigma _i}{\sigma _j}}} \cdot \frac{{2{\mu _i}{\mu _j}}}{{\mu _{i}^{2}+\mu _{j}^{2}}} \cdot \frac{{2{\sigma _i}{\sigma _j}}}{{\sigma _{i}^{2}+\sigma _{j}^{2}}}$$ (12)
Where \({\mu _i},{\mu _j}\) refer to the average of i and j, \({\sigma _i},{\sigma _j}\) refer to the variance i and j, and \(Co{v_{ij}}\) refers to the covariance of i and j.
By summarizing the results of the criteria, we set for the proposed MBCS system in Table 1; we bring the quantitative and qualitative evaluation to a close.
Table 1 displays the statistical analysis of the suggested MBCS. The metrics are determined by contrasting the original photos with the produced cancelable templates. The suggested MBCS performed effectively without the use of evaluation metrics. It has a UIQ of 0.912, an SD of 60.329, an NPCR of 99.225, a PSNR of 24.622, an SSIM of 0.068, and a UACI of 23.547. The proposed MBCS can be used privately for cyber security applications based on human biometrics, according to evaluation metrics.
Table 1: Statistical analysis of the proposed MBCS.
|
Image No. |
Quantitative Metrics |
Qualitative Metrics |
|||||||
|
Correlation |
NPCR (%) |
UACI |
PSNR (dB) |
SSIM |
UIQ |
SD |
|||
|
Diagonal |
Horizontal |
vertical |
|||||||
|
1 |
0.932 |
0.963 |
0.922 |
22.396 |
22.396 |
25.99 |
0.067 |
0.92 |
57.109 |
|
2 |
0.961 |
0.975 |
0.921 |
23.353 |
23.353 |
24.39 |
0.086 |
0.91 |
59.549 |
|
3 |
0.952 |
0.967 |
0.942 |
23.463 |
23.463 |
24.56 |
0.075 |
0.905 |
59.831 |
|
4 |
0.980 |
0.981 |
0.907 |
24.282 |
24.282 |
23.75 |
0.079 |
0.904 |
61.918 |
|
5 |
0.949 |
0.969 |
0.953 |
24.034 |
24.034 |
24.28 |
0.066 |
0.904 |
61.284 |
|
6 |
0.962 |
0.974 |
0.941 |
23.557 |
23.557 |
24.09 |
0.086 |
0.901 |
60.069 |
|
7 |
0.981 |
0.978 |
0.941 |
24.397 |
24.397 |
24.31 |
0.084 |
0.908 |
62.212 |
|
8 |
0.935 |
0.943 |
0.954 |
23.713 |
23.713 |
24.62 |
0.067 |
0.909 |
60.468 |
|
9 |
0.957 |
0.979 |
0.937 |
23.449 |
23.449 |
24.73 |
0.092 |
0.921 |
59.794 |
|
10 |
0.944 |
0.934 |
0.932 |
22.409 |
22.409 |
25.01 |
0.063 |
0.923 |
61.325 |
|
11 |
0.968 |
0.965 |
0.961 |
23.538 |
23.538 |
24.92 |
0.074 |
0.909 |
60.862 |
|
12 |
0.929 |
0.956 |
0.952 |
24.129 |
24.129 |
24.33 |
0.082 |
0.916 |
60.912 |
|
13 |
0.938 |
0.971 |
0.980 |
24.226 |
24.226 |
24.04 |
0.085 |
0.925 |
60.537 |
|
14 |
0.963 |
0.937 |
0.949 |
23.741 |
23.741 |
24.19 |
0.0906 |
0.918 |
61.121 |
|
15 |
0.985 |
0.982 |
0.962 |
22.998 |
22.998 |
25.02 |
0.073 |
0.906 |
58.647 |
|
16 |
0.946 |
0.947 |
0.981 |
24.306 |
24.306 |
25.56 |
0.019 |
0.908 |
59.778 |
|
17 |
0.933 |
0.951 |
0.935 |
24.561 |
24.561 |
24.69 |
0.026 |
0.911 |
60.625 |
|
18 |
0.965 |
0.961 |
0.957 |
22.965 |
22.965 |
24.93 |
0.035 |
0.921 |
61.109 |
|
19 |
0.958 |
0.983 |
0.944 |
23.078 |
23.078 |
24.87 |
0.048 |
0.914 |
60.216 |
|
20 |
0.981 |
0.972 |
0.968 |
22.346 |
22.346 |
24.16 |
0.077 |
0.924 |
59.226 |
|
Average |
99.225 |
23.547 |
24.622 |
0.068 |
0.912 |
60.329 |
|||
4.3 BIO-HASH TEMPLATES RESULTS
In this part, the simulation outcomes of bio-hash templates created using the SHA-512 hashing algorithm are explored. Each hashing algorithm aims to generate unique hash templates. Additionally, the resultant template is not common to use in cancelable systems, as it can be considered one of the security techniques, because the templates used frequently are encrypted images rather than text templates. We also look into the worst-case scenario, where a hacker predicts the hashing scenario and uses the same technique to compromise the system. As a result, the proposed bio-hash system is validated using an attacker template. We employed both the hamming distance and the correlation as pairwise distances to assess the gaps between both the generated templates as well as the attacker templates. Figures 9 and 10 depict the hamming distance curve and the correlation curve for the proposed SHA-224 and SHA-256 algorithms, respectively. The attacker template is far removed from the generated user templates, as evidenced by the hamming distance between each generated user template and the attacker template ranging between 0.9 and 1. Furthermore, the findings are grouped around 0 on the correlation curve, demonstrating a low correlation between the attacker template and the user templates.
The problem of cybersecurity has been discussed in this work. The proposed system is focused on generating cancelable templates in both visual and text forms. This paper proposes a cancelable biometric system based on image composition, deep dream, and hashing techniques. The objective of the proposed system is to generate visual and text cancelable biometrics. The visual cancelable templates are generated using image composition and deep-dream, while the text templates are generated using SHA hashing techniques. The proposed system is validated by multi-biometric inputs including iris, palm, face, and fingerprint biometrics. In addition, it is evaluated in both visual and text forms. The simulation results reveal that the proposed system appears a superior performance among the works which handle this problem.
Ethical Approval
Not applicable
Competing interests
Not applicable
Authors' contributions
The authors confirm contribution to the paper as follows:
Study conception and design: Emad A. Elshazly1, Nariman Abdel-Salam5.
Data collection: Emad A. Elshazly1, Ahmed Sedik3.
Analysis and interpretation of results: Fatma G. Hashad2, Ahmed Sedik3, Fathi E. Abd El-Samie4.
Draft manuscript preparation: Fatma G. Hashad2, Nariman Abdel-Salam5.
All authors reviewed the results and approved the final version of the manuscript.
Funding
Not applicable
Availability of data and materials
The data that support the findings of this paper is available from the corresponding author upon requsest.
- Sinha, A. and K. Singh, Image encryption using fractional Fourier transform and 3D Jigsaw transform. Optical Engineering, 2013.
- Rajpoot, Q.M. and C.D. Jensen. Security and Privacy in Video Surveillance: Requirements and Challenges. in ICT Systems Security and Privacy Protection. 2014. Berlin, Heidelberg: Springer Berlin Heidelberg.
- Enerstvedt, O.M., Analysis of Privacy and Data Protection Principles, in Aviation Security, Privacy, Data Protection and Other Human Rights: Technologies and Legal Principles, O.M. Enerstvedt, Editor. 2017, Springer International Publishing: Cham. p. 307-394.
- Zheng, X. The Application of Information Security Encryption Technology in Military Data System Management. in Man–Machine–Environment System Engineering. 2018. Singapore: Springer Singapore.
- Madaan, V., et al. Public Network Security by Bluffing the Intruders Through Encryption Over Encryption Using Public Key Cryptography Method. in Advanced Informatics for Computing Research. 2017. Singapore: Springer Singapore.
- Stergiou, C. and K.E. Psannis, Efficient and secure BIG data delivery in Cloud Computing. Multimedia Tools and Applications, 2017. 76(21): p. 22803-22822.
- Jain, A.K., A. Ross, and S. Prabhakar, An introduction to biometric recognition. IEEE Transactions on circuits and systems for video technology, 2004. 14(1): p. 4-20.
- Jegede, A., et al., Cancelable and hybrid biometric cryptosystems: current directions and open research issues. 2017.
- Rathgeb, C. and C. Busch, Multi-biometric template protection: Issues and challenges, in New trends and developments in biometrics. 2012, InTech.
- Nandakumar, K. and A.K. Jain, Biometric Template Protection: Bridging the performance gap between theory and practice. IEEE Signal Processing Magazine, 2015. 32(5): p. 88-100.
- Rachapalli, D.R. and H.K. Kalluri. A survey on biometric template protection using the cancelable biometric scheme. in 2017 Second International Conference on Electrical, Computer and Communication Technologies (ICECCT). 2017.
- Rane, S., Standardization of Biometric Template Protection. IEEE MultiMedia, 2014. 21(4): p. 94-99.
- Patel, V.M., N.K. Ratha, and R. Chellappa, Cancelable Biometrics: A review. IEEE Signal Processing Magazine, 2015. 32(5): p. 54-65.
- Moujahdi, C., et al. Spiral Cube for Biometric Template Protection. in Image and Signal Processing. 2012. Berlin, Heidelberg: Springer Berlin Heidelberg.
- Tarif, E.B., et al., A hybrid encryption/hiding method for secure transmission of biometric data in multimodal authentication system. Multimedia Tools and Applications, 2018. 77(2): p. 2485-2503.
- Sree, S.R.S. and N. Radha. Cancellable multimodal biometric user authentication system with fuzzy vault. in 2016 International Conference on Computer Communication and Informatics (ICCCI). 2016.
- Fatma G. Hashad, Osama Zahran, El-Sayed M. El-Rabaie, Ibrahim F. Elashry, Fathi E. Abd El-Samie. Fusion-based encryption scheme for cancelable fingerprint recognition. Multimedia Tools and Applications (2019) 78:27351–27381.
- Hossain, M. B., Rahman, M. T., Rahman, A. S., & Islam, S. (2014, May). A new approach of image encryption using 3d chaotic maps to enhance security of multimedia component. In 2014 International Conference on Informatics, Electronics & Vision (ICIEV) (pp. 1-6). IEEE.
- Zhang, W., Yu, H., Zhao, Y. L., & Zhu, Z. L. (2016). Image encryption based on three-dimensional bit matrix permutation. Signal Processing, 118, 36-50.
- Belazi, A., El-Latif, A. A. A., & Belghith, S. (2016). A novel image encryption scheme based on substitution-permutation network and chaos. Signal Processing, 128, 155-170.
- Saha, A., Saha, J., Sen, B.: (2019). An expert multi-modal person authentication system based on feature-level fusion of iris and retina recognition. In: 2019 International Conference on Electrical, Computer and Communication Engineering (ECCE).
- Hammad, M., Liu, Y., Wang, K.: Multimodal biometric authentication systems using convolution neural network based on different level fusion of ear and fingerprint. IEEE Access 7, 26527–26542 (2019).
- Rane, M., Deshpande, P., (2018) Multimodal biometric recognition system using feature level fusion. In: 2018 Fourth International Conference on Computing Communication Control and Automation (ICCUBEA)
- Rajalakshmi, B., Sumathi, S.: (2018) Survey of multimodal biometric using ear and finger knuckle image. In: 2018 International Conference on Communication, Computing, and Internet of Things (IC3IoT).
- Mei Yin, D., Mukhlas, A., Wan Chik, R., Talib Othman, A., Omar, S.: (2018) A proposed approach for biometric-based authentication using of face and facial expression recognition. In: 2018 IEEE 3rd International Conference on Communication and Information Systems (ICCIS)
- Bharadi, V., Pandva, B., Cosma, G.: (2018) Multi-modal biometric recognition using human iris and dynamic pressure variation of handwritten signatures. In: 2018 Fifth International Conference on Social Networks Analysis, Management and Security (SNAMS)
- Cheniti, M., Boukezzoula, N., Akhtar, Z.: Symmetric sum-based biometric score fusion. IET Biometrics 7, 391–395 (2018)
- Ammour, B., Bouden, T., Boubchir, L.: Face–iris multi-modal biometric system using multi-resolution Log-Gabor filter with spectral regression kernel discriminant analysis. IET Biometrics 7, 482–489 (2018)
- Hezil, N., Boukrouche, A.: Multimodal biometric recognition using human ear and palmprint. IET Biometrics 6, 351–359 (2017)
- Veluchamy, S., Karlmarx, L.: System for multimodal biometric recognition based on finger knuckle and finger vein using feature-level fusion and k-support vector machine classifier. IET Biometrics 6, 232–242 (2017)
- Al-Waisy, A., Qahwaji, R., Ipson, S., Al-Fahdawi, S.: (2017). A multimodal biometric system for personal identification based on deep learning approaches. In: 2017 Seventh International Conference on Emerging Security Technologies (EST)
- Basak, P., De, S., Agarwal, M., Malhotra, A., Vatsa, M., Singh, R.: (2017) Multimodal biometric recognition for toddlers and pre-school children. In: 2017 IEEE International Joint Conference on Biometrics (IJCB)
- Murakami, T., Ohki, T., Kaga, Y., Fujio, M., and Takahashi, K. (2019). Cancelable indexing based on low-rank approximation of correlation-invariant random filtering for fast and secure biometric identification. Pattern Recognition Letters, 126, 11-20.
- Yang, W., Wang, S., Hu, J., Zheng, G., and Valli, C. (2018). A fingerprint and finger-vein based cancelable multi-biometric system. Pattern Recognition, 78, 242-251.
- Kaur, H., and Khanna, P. (2015). Gaussian random projection based non-invertible cancelable biometric templates. Procedia Computer Science, 54, 661-670.
- Yadav AK. Deep Learning Approach for Multimodal Biometric Recognition System Based on Fusion of Iris, Fingerprint and Hand Written Signature Traits. Turkish Journal of Computer and Mathematics Education (TURCOMAT). 2021 May 10; 12(11):1627-40.
- E. Bouzouina, Y.; Hamami, L. Multimodal Biometric: Iris and face recognition based on feature selection of iris with GA and scores level fusion with SVM. In proceedings of 2017 2nd International Conference on Bio-Engineering for Smart Technologies, Paris, France, 30 August-1 September 2017; pp.1-7.
- F. Hezil, N, Boukrouche, A. Multimodal biometric recognition using human ear and palmprint. IET Biom. 2017, 6, 351-359.
- G. Veluchamy, S.; Karlmark, L.R. System
- H. Chanukya, P. S. V. V. N.; Thivakaran, T.K. Multimodel Biometric cryptosystem for human authentication using fingerprint and ear. Multimedia. Tools Appl. 2020, 79, 659-673.
- I. Ammour, B.; Boubchir, L.; Bouden, T.; Ramdani, M. Face-iris multimodal biometric identification system. Electronics 20202, 9, 85.
- Abdellatef E, Ismail NA, Abd Elrahman SE, Ismail KN, Rihan M, Abd El-Samie FE. Cancelable multi-biometric recognition system based on deep learning, The Visual Computer. 2020 Jun; 36(6):1097-109.
- Abdellatef E, Omran EM, Soliman RF, Ismail NA, Abd Elrahman SE, Ismail KN, Rihan M, Abd El-Samie FE, Eisa AA. Fusion of deep-learned and hand-crafted features for cancelable recognition systems. Soft Computing. 2020 Oct; 24(20):15189-208.
- Alay N, Al-Baity HH. Deep learning approach for multimodal biometric recognition system based on fusion of iris, face, and finger vein traits. Sensors. 2020 Jan; 20(19):5523.
- Abeer D. Algarni, Ghada El Banby, Sahar Ismail, Walid ElShafai, Fathi E. Abd El-Samie, and Naglaa F. Soliman, “Discrete Transforms and Matrix Rotation Based Cancelable Face and Fingerprint Recognition for Biometric Security Applications”, MDPI, Entropy 2020.
- Christian Rathgeb & Andreas Uhl, “A survey on biometric cryptosystems and cancelable biometrics”, EURASIP Journal on Information Security volume 2011, Article number: 3 (2011).
- Zahra Parvin, Hadi Seyedarabi & Mousa Shamsi, “A new secure and sensitive image encryption scheme based on new substitution with chaotic function”, An International Journal Multimedia Tools and Applications, Volume 70, Number 1, May 2014.
- Hilda Deborah, Noël Richard, and Jon Yngve Hardeberg, “A Comprehensive Evaluation of Spectral Distance Functions and Metrics for Hyperspectral Image Processing”, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, June 2015.
- Basma Abd El-Rahiem, Mohamed Amin, Ahmed Sedik, Fathi E. Abd El Samie, and Abdullah M. Iliyasu, “An efficient multi-biometric cancellable biometric scheme based on deep fusion and deep dream”, Journal of Ambient Intelligence and Humanized Computing, Springer, April 2022.
- Davide Piccini, Robin Demesmaeker, John Heerfordt, Jérôme Yerly, Lorenzo Di Sopra, Pier Giorgio Masci, Juerg Schwitter, Dimitri Van De Ville, Jonas Richiardi, Tobias Kober, Matthias Stuber, “Deep Learning to Automate Reference-Free Image Quality Assessment of Whole-Heart MR Images”, Radiology: Artificial Intelligence Volume 2: Number 3, 2020.
No competing interests reported.