Rabia Bakhteri

Biometric encryption based on a fuzzy vault scheme with a fast chaff generation algorithm

Fuzzy vault is a scheme that complements traditional cryptographic security systems by combining it with biometric authentication to overcome the security vulnerability inherent in cryptographic key storage. Biometric encryption systems based on fuzzy vault scheme are suitable for stand-alone security and authentication devices in the form of system-on-chip (SoC). However, the current fuzzy vault scheme has too many compute-intensive processes to make this feasible for SoC implementation. The most critical but compute-intensive function in the fuzzy vault scheme is the chaff generation which produces noise (chaff) points that hide the valid points inside the vault template. In this paper, we propose a new chaff generation algorithm which is computationally fast and viable for hardware acceleration by employing simple arithmetic operations. Complexity study shows that our algorithm has a complexity of O(n^2), which is a significant improvement over the existing method that exhibits O(n^3) complexity. Our experimental results show that, to generate 500 chaff points, the proposed algorithm gives a performance speed-up of over 140 times over existing Clancys algorithm. With the new chaff generation algorithm, it becomes much more amenable to implement the fuzzy vault scheme in the resource-constrained environment of system-on-chip.

Securing cryptographic key with fuzzy vault based on a new chaff generation method

A crucial issue in the design of a cryptographic system is the problem of key management. A state-of-the-art solution to this problem is to use bio-cryptosystems, in which cryptography is combined with biometrics. In this solution, the user biometrics is used to protect the cryptographic key. A popular approach to the design of such bio-cryptosystems is the application of a fuzzy vault scheme. This so-called vault is a secure storage in which the key is hidden within the biometric data mixed up with meaningless chaff points. The most critical operation in the fuzzy vault scheme is generation of these chaff points. Experiments will show that this module is the most compute-intensive part of the whole system. This paper introduces a new chaff generation algorithm for the fuzzy vault in a bio-cryptosystem. The proposed algorithm, which is based on a circle packing mathematical theorem, is computationally less intensive than existing methods. Experimental results show that the proposed algorithm is around 100 times faster than existing methods for 200, and above, number of chaff points, and therefore, is suitable for a real-time embedded system implementation.

Bounded activation functions for enhanced training stability of deep neural networks on visual pattern recognition problems

This paper focuses on the enhancement of the generalization ability and training stability of deep neural networks (DNNs). New activation functions that we call bounded rectified linear unit (ReLU), bounded leaky ReLU, and bounded bi-firing are proposed. These activation functions are defined based on the desired properties of the universal approximation theorem (UAT). An additional work on providing a new set of coefficient values for the scaled hyperbolic tangent function is also presented. These works result in improved classification performances and training stability in DNNs. Experimental works using the multilayer perceptron (MLP) and convolutional neural network (CNN) models have shown that the proposed activation functions outperforms their respective original forms in regards to the classification accuracies and numerical stability. Tests on MNIST, mnist-rot-bg-img handwritten digit, and AR Purdue face databases show that significant improvements of 17.31%, 9.19%, and 74.99% can be achieved in terms of the testing misclassification error rates (MCRs), applying both mean squared error (MSE) and cross-entropy (CE) loss functions This is done without sacrificing the computational efficiency. With the MNIST dataset, bounding the output of an activation function results in a 78.58% reduction in numerical instability, and with the mnist-rot-bg-img and AR Purdue databases the problem is completely eliminated. Thus, this work has demonstrated the significance of bounding an activation function in helping to alleviate the training instability problem when training a DNN model (particularly CNN).

An optimized second order stochastic learning algorithm for neural network training

This paper proposes an improved stochastic second order learning algorithm for supervised neural network training. The proposed algorithm, named bounded stochastic diagonal Levenberg-Marquardt (B-SDLM), utilizes both gradient and curvature information to achieve fast convergence while requiring only minimal computational overhead than the stochastic gradient descent (SGD) method. B-SDLM has only a single hyperparameter as opposed to most other learning algorithms that suffer from the hyperparameter overfitting problem due to having more hyperparameters to be tuned. Experiments using the multilayer perceptron (MLP) and convolutional neural network (CNN) models have shown that B-SDLM outperforms other learning algorithms with regard to the classification accuracies and computational efficiency (about 5.3% faster than SGD on the mnist-rot-bg-img database). It can classify all testing samples correctly on the face recognition case study based on AR Purdue database. In addition, experiments on handwritten digit classification case studies show that significant improvements of 19.6% on MNIST database and 17.5% on mnist-rot-bg-img database can be achieved in terms of the testing misclassification error rates (MCRs). The computationally expensive Hessian calculations are kept to a minimum by using just 0.05% of the training samples in its estimation or updating the learning rates once per two training epochs, while maintaining or even achieving lower testing MCRs. It is also shown that B-SDLM works well in the mini-batch learning mode, and we are able to achieve 3.32 × performance speedup when deploying the proposed algorithm in a distributed learning environment with a quad-core processor.

FPGA-based finger vein biometric system with adaptive illumination for better image acquisition

Finger vein imaging using near infrared methods have high sensitivity towards illumination and often suffer from bad image sharpness and loss of information. An embedded image acquisition subsystem with adaptive illumination control based on image quality assessment is introduced in this paper in order to acquire finger vein image with good quality for high accuracy authentication. The quality of the acquired finger vein image is assessed using two-dimensional (2D) entropy and the infrared illumination is adaptively adjusted based on the assessment result. 2D entropy for finger vein image quality assessment that is suitable for hardware implementation is proposed. Mathematically, the proposed 2D entropy significantly improves the performance as well as resource utilization of the quality assessment module compared to previous work. Besides that, buck converter is designed as the Light-Emitting Diode (LED) driver circuit to control the brightness level of the high power infrared LED array efficiently for finger vein image capture. The proposed subsystem is deployed in the FPGA-based finger vein biometric system operates on Nios2-Linux Real Time Operating System (RTOS). Experimental results show that finger vein images acquired through the proposed image acquisition subsystem contain more information as well as better image sharpness compared to finger vein images captured under fixed illumination.

Comparative study of electrocardiogram QRS complex detection algorithm on Field Programmable Gate Array platform

Nowadays, many people suffer from heart problems and hence the demand of inexpensive and efficient electrocardiogram (ECG) for frequent heart monitoring is becoming crucial. To make the ECG device portable, cost-effective and light-weight, an alternative of deploying an ECG system is on Field Programmable Gate Array (FPGA). It is also important to choose suitable algorithm that optimize in terms of feature extraction accuracy and computation time. This paper implements and compares two methods of ECG QRS detection in terms of Pan and Tompkins algorithm and Derivative-based Method on FPGA platform as embedded software computation. The Derivative-based Method is modified from fixed threshold to adaptive threshold to increase robustness in real time QRS detection. The input are 48 records of 30 minutes, total 24 hours ECG data obtained from MIT-BIH database as standard database for performance benchmarking. Both algorithm yield a different outcome in terms of accuracy and computation speed. Results show that though Pan and Tompkins shows a better accuracy (98.15%) on detecting the QRS complex compared to Derivative-based Method (96.73%), the latter consume less than half of computation time (a total 17.86 minutes to compute 24 hours ECG data). Compared to Pan and Tompkins algorithm (a total 45.2 minutes to compute 24 hours ECG data).

OpenCL-based hardware-software co-design methodology for image processing implementation on heterogeneous FPGA platform

Recently, the OpenCL hardware-software co-design methodology has gained traction in realizing effective parallel architecture designs in heterogeneous FPGA platforms. In fact, the portability of OpenCL on hardware ready platforms such as GPU or multicore CPU enables ease of design verification. This is true especially for parallel algorithms before implementing them using cumbersome HDL-based RTL design. In this paper we employed OpenCL programming platform based on Altera SDK for OpenCL (AOCL) to implement a Sobel filter algorithm as an image processing test case on a Cyclone V FPGA board. Using the portability of this platform, the performance of the kernel code is benchmarked against that of the GPU and multicore CPU implementations for different image and kernel sizes. Different optimization strategies are also applied for each platform. We found that increasing the Sobel filter kernel size from 3×3 to 5×5 results in only 11.3% increase in computation time for FPGA, while the effect was much more significant where the execution time was as high as 23.6% and 85.7% for CPU and GPU, respectively.

An Optimized Second Order Stochastic Learning Algorithm for Neural Network Training

The performance of a neural network depends critically on its model structure and the corresponding learning algorithm. This paper proposes bounded stochastic diagonal Levenberg-Marquardt (B-SDLM), an improved second order stochastic learning algorithm for supervised neural network training. The algorithm consists of a single hyperparameter only and requires negligible additional computations compared to conventional stochastic gradient descent (SGD) method while ensuring better learning stability. The experiments have shown very fast convergence and better generalization ability achieved by our proposed algorithm, outperforming several other learning algorithms.

Distributed B-SDLM: accelerating the training convergence of deep neural networks through parallelism

This paper proposes an efficient asynchronous stochastic second order learning algorithm for distributed learning of neural networks (NNs). The proposed algorithm, named distributed bounded stochastic diagonal Levenberg-Marquardt (distributed B-SDLM), is based on the B-SDLM algorithm that converges fast and requires only minimal computational overhead than the stochastic gradient descent (SGD) method. The proposed algorithm is implemented based on the parameter server thread model in the MPICH implementation. Experiments on the MNIST dataset have shown that training using the distributed B-SDLM on a 16-core CPU cluster allows the convolutional neural network (CNN) model to reach the convergence state very fast, with speedups of 6.03× and 12.28× to reach 0.01 training and 0.08 testing loss values, respectively. This also results in significantly less time taken to reach a certain classification accuracy ( 5.67× and 8.72× faster to reach 99% training and 98% testing accuracies on the MNIST dataset, respectively).

Convolutional Neural Networks with Fused Layers Applied to Face Recognition

In this paper, we propose an effective convolutional neural network (CNN) model to the problem of face recognition. The proposed CNN architecture applies fused convolution/subsampling layers that result in a simpler model with fewer network parameters; that is, a smaller number of neurons, trainable parameters, and connections. In addition, it does not require any complex or costly image preprocessing steps that are typical in existing face recognizer systems. In this work, we enhance the stochastic diagonal Levenberg–Marquardt algorithm, a second-order back-propagation algorithm to obtain faster network convergence and better generalization ability. Experimental work completed on the ORL database shows that a recognition accuracy of 100% is achieved, with the network converging within 15 epochs. The average processing time of the proposed CNN face recognition solution, executed on a 2.5 GHz Intel i5 quad-core processor, is 3 s per epoch, with a recognition speed of less than 0.003 s. These results show that the proposed CNN model is a computationally efficient architecture that exhibits faster processing and learning times, and also produces higher recognition accuracy, outperforming other existing work on face recognizers based on neural networks.