1 Introduction
Biometric is the science of verifying and establishing the identity of an individual through physiological features (fingerprint, hand vein, face and finger vein) or behavioral traits (handwriting, speech, and signature), for automatic recognition ((Singh et al.,2016; Malik et al. 2013). The law enforcement agencies was first abducted biometric systems in 1970s to investigate criminals through uses of fingerprint recognition (Jain and Kumar, 2012). However, in the current biometric technologies advancement with the growth of threats in information security, the biometric application systems have proliferated into the physical and logical access control domains (Bolle et al.,2013). There is encouragement of these kind of systems because of less cost of its biometric capture machines (Haar et al.,2013; Nixon et al., 2015). The application of these kinds can be applied to general or company domain at various levels, they can be integrated inside the single consumer’s products. For instance, Malaysia personality card called `MyKad’ is a general initiative organised by the government that integrates fingerprint biometric to enhance its function and effectiveness. In this case, an individual’s personality and citizenship can be varied based on his fingerprint, which also combat fraud and spoofing of national IDs. In addition, Malaysia’s Employees Provident Fund (EPF) is also used MyKad biometric authentication at the automatic kiosk and at the counter of other government agencies providing an authentication system to discrete account boxes. At the Office of Legislative Counsel in the United States, hand geometry recognition and iris scanning system has also been deployed for physical entrances to ensure the confidentiality (Haar et al.,2013). Some of biometric authentication, which are integrated inside discrete consumer products contain login into notebooks and smartphones that utilize low-cost curve fingerprint devices (Jain and Kumar, 2012).
In this present time, forensics are rapidly and increasingly evolving Biometric application technologies such in criminal identification and prison security. Also, Biometrics has the capability to remain extensively approved in a very broad range of national applications such as banking security, physical access control, information system security, customs and immigration, national ID systems, voter and driver registration. These technologies have been made possible by explosive advances in computing power and have been made necessary by the near general interconnection of computers around the world. Figure 1 shows typical of existing Biometric features.
Figure 1 A typical biometric features
However, every Biometrics authentication systems have its own shortcomings based on the qualities, capturing device, database and feature of that qualities. For instance, finger print happens to be a widespread feature for recognition but it can be easily spoof using dummy fingerprint, which is also sensitive to dirt, wet and age. Likewise, facial recognition is also vulnerable to the face expression and age (Agrawal. 2013) . Voice recognition is contingent upon the environmental condition and not secure from the recorded voice (Singh et al., 2012). Having consider the challenges in the current recognition system, there is need to design the robust unimodal recognition system to protect the privacy. Hence, in this recent age, Finger vein has been studied to be biometric verification of a current identification system (Yang et al. 2014). This system can be used for extensive variation of applications, which includes network authentication, employee time and attendance tracking, credit card authentication, and automobile security.
In line that hand holds plenty of information, which are easy to be retrieved, the biometric technologies of hand based of fingerprint and palm print are the most common in practice. It is known that fingerprint is the maximum recognized hand based biometric method where it has been used in various purposes for ages. Nevertheless, forgery is the problem of fingerprint based biometric system as it is visible to eyes. The sweat and dryness of finger surface can avert the clearance of finger pattern to be obtained, which can lower the performance of system. In order to solve the inadequacies of existing hand based biometric systems, the research into finger vein recognition comes to limelight. It has been verified that every finger has distinctive vein patterns which is useful in personal verification. Hence, it is confirmed that the advantages of finger vein based biometric system are more than that of any hands based biometric technologies. It is not easy for the finger vein pattern to be replicated as it is an internal feature. Also, the value of the vein pattern captured cannot be manipulated by the situation of the skin like that of palm vein or finger print. In addition, the magnitude of the machine of finger vein method can be made much smaller that of hand based biometrics (Kakkar et al., 2013).
1. Literature Review
In this age of increasing in information growth, the main social problem at hand are how to solve human identity recognition and protect information security. The conventional identity recognition contains two type methods, namely; contents based which are password, code and so on, and possessing based, which are smart card, licence, and so on. Nevertheless, with this development growth, many of these conventional identity recognition methods abuse begins to appear (WANG et al, 2011). Both code and password can be forgotten or stolen. Though passwords are still widely used, their lack of proper implementation has seen password gaining immense criticism (Bonneau and Preibusch, 2010). In case of certificate key and smart card, there is probability of been stolen, forged and lost. So, the identity recognition technology on biology features arises in order to solve the shortcomings of traditional identity recognition. Human in-built biology characteristics like face, vein and DNA and also, human conduct characteristics like voice, sign and gait are used to fish-out person identity in the case of optics, biosensor, inosculation of computer etc. Thus, this discovery become better in terms of accuracy, suitability, time, privacy, safety, and trustworthiness than uses of certificate, code and card. Now, finger print authentication scheme are recognition appears like mostly widely used mainly in area of information system security. However, the privacy of it cannot be granted as it can be spoofed and forged. In the area of security and application, the vein technology has many merits in biometrics such as generality and distinctiveness. That���s, as human age increases, the vein pattern does not change, which means is static in nature. Also, sickness, surgery, and epidermis does not change the body vein pattern such that there will be conflict over two people personal identification. Biometric traits are increasingly being used because it overcomes the problem of creating complex passwords and the user has one less thing to remember; their biometric traits will always be with them.
Table 1 describes the comparative evaluation of the major biometric technologies. Each of these techniques has its own merits and shortcomings. In all, fingerprint has the least expensive method and it is usually used in criminal���s identification. Nevertheless, there may be difficulty in image capturing of people due to their work and age which may make the finger print capturing incomplete and accurate. In addition to this, the uses or the application of fingerprints are limited as it can be replicated.
Table 1 Comparison of major biometrics techniques
Biometrics Accuracy Size of Template Cost Security Level Long-term Stability
Finger Vein High Medium High High High
Fingerprint Medium Small Low Low Low
Facial Recognition Low Large High Low Low
Iris Scan High Large High Medium Medium
Speaker Recognition Low Small Medium Low Low
Hand Geometry Low Large High Low Low
Finger vein contains vein patterns, which is the networks of blood vessel under the skin of finger. Individual has unique vein patterns as even twin has different vein. Interrelated traits of hand vein and palm vein have been deployed but they still do not have popular because of large space for database and high cost. Finger vein patterns have great advantages as compare to other popular biometrics traits with the following reasons (Kono et al,2015; Yang et al. 2014).
(1) The data used for authentication is obtained from the finger veins in a person���s body, and the difficulty of stealing this finger vein pattern makes identity fraud difficult.
(2) The technique has sufficient complexity and is highly accurate.
(3) Improvements to the algorithm provide fast authentication speed, and the technique is easy to use because it only requires the person to place their finger over a reader.
Compared to fingerprint recognition, another widely used biometric authentication technique, finger vein authentication has the following advantages.
(1) A high degree of repeatability between recording the initial pattern and subsequent checking because the condition of the finger surface (moisture or dryness, etc.) has little influence.
(2) Unlike people���s fingerprints, which can be replicated from marks they leave behind, counterfeiting someone���s finger vein pattern is difficult.
(3) Authentication can be performed without touching the sensor unit, meaning that the technique is not significantly influenced by dirt or scratching on the reader sensor.
Table 2: Related works and their Techniques
Ref Authors Topic Paper Technique Features Limitation/Comments
1 Yu, et al 2009 Finger-vein image recognition combining modified hausdorff distance with minutiae feature matching Modified Hausdoriff Distance Algorithm Minutiae techniques utilize vein features from the segmented blood vessel network. Segmentation errors may occur during the feature extraction process due to the low quality of finger vein images.
Improperly segmented networks may degrade the recognition accuracy significantly
2 Song, et al 2011 A finger-vein verification system using mean curvature. Matched Pixel Ratio (MPR) Mean Curvature The same as above
3 Huang, et al 2010 Finger-vein authentication based
on wide line detector and pattern normalization Wide line detector The same as above.
Introduction of Texture-Based Method based on Binary Patterns is a solution
4 Lee, et al 2009 Finger vein recognition using minutia-based alignment and local binary pattern-based feature extraction Local Binary Pattern (LBP) Minutiae points
5 Lee, Jung, & Kim 2011 New finger biometric method using near infrared imaging. Modified Gaussian high-pass filter (MGF)
Local binary Pattern + Local Derivative Pattern (LBP + LDP) Finger Code (Vein + geometry) Increasing in the feature extraction complexity
6 Rosdi, et al 2011 Finger vein recognition using local line
binary pattern. Local Line Binary Pattern (LLBP) Minutiae *LLBP computation is faster and reduces the feature extraction complexity BUT
*The computational time is greater
*Binary pattern operators usually use zero as the threshold, which makes the description of vein patterns in the sub-region sensitive to noise.
7 Mahri, et al 2010 Finger vein recognition algorithm
using phase only correlation Band Limited Phase Only Correlation (BLPOC) Mean Curvature Vein recognition is known to entail the problem that the thickness of a vein changes according to blood flow which varies according to weather and the health condition of the user
8 Kang, et al 2011 Multimodal biometric method that
combines veins, prints, and shape of a finger. Score-level fusion Finger vein,
Fingerprint, &
Finger geometry *Improvement of Mahri, et al 2010 work.
*Their geometry feature which based on finger thickness is easily affected by the rotation and translation of the finger
*The finger thickness or width is sensitive to segmentation errors
9 Mohd, et al 2014 Fusion of Band Limited Phase Only Correlation and Width Centroid Contour Distance for finger based biometrics Score-Level Fusion (Finger Vein Recognition based on BLPOC + finger Geometry Recognition based on WCCD) width (W) + Centroid
Contour Distance (CCD) = Width-Centroid Contour
Distance (WCCD) Propose technique for solving distortion problem is a further study
Also, propose fusion technique for better result is required
2. Problem Statement
Li, in his own case offered a modality-based bi-finger vein verification system (Yang et al,2014). In this case, the distinct sensor was used to take image of the finger vein and its shape. This system made uses of different finger vein network extraction algorithm and the coordinate system made uses of intersection of the forefinger and middle finger. The method determines the region of interest, the finger vein and shape features extraction and its corresponding fusion verification but the pattern identification of distorted images is not identified.
In the case of Lin and other researchers, they presented an algorithm that can segment the dorsal hand vein image and extract the vein skeleton (Yang et al., 2010). From their result, gravy and size was normalized, Gaussian low pass filter and median filter was used in removing the spot noise and the parallel stripe skimming noise separately. Afterward, an improved NiBlack algorithm segments the vein pattern and an area thresholding algorithm was used to clear the noise parts from the vein pattern. The system made used of dorsal hand vein and not finger vein, hence, there work does not focus on this on-going research work.
Liu and Song in year 2012 suggested a real-time embedded finger-vein recognition system for authentication on mobile devices (Liu and Song, 2012). This system was applied on a DSP platform and equipped through a novel finger-vein recognition algorithm. It took the system about 0.8 seconds to validate one input finger-vein sample and achieved an equal error rate of 0.07 percent on a database of 100 subjects. This system results showed that the finger-vein recognition system is capable for authentication on mobile devices. However, the system was not implemented on PC system to know its accuracy and throughput. The 100 subjects are also low compare to technology growth rate in the present age.
Lu (Yang et al., 2014) developed three frame works to conduct the combination of the width measurement and finger vein pattern ,i.e., the fusion frame work, the filter frame work and the hybrid framework but the stability of soft biometric trait extracted directly from images cannot be explored further, as the measurement of soft biometric trait depends on the scale of the image, which is related to image acquisition device. The width of phalangeal joint used also depends on the scale of the image.
Therefore, the problem statements are as follows:
1. Considering researches that have done on human verification system using finger vein unimodal biometric, it become very clear that the accuracy of human verification can be degraded by many factors such as the condition of a sensor, the health condition of a human, illumination variation, image distortion and so on (Yang et al. 2014; Peng et al, 2014 ; Asaari et al 2014). Multimodal method in integrating finger vein and finger geometry recognition is required to solve this problem.
2. Sometimes the shaded region hinders the both top and bottom areas of the captured finger images, which results into dim. When this happened, it is difficult to extract the finger edges and its vein pattern. If the extraction is not properly performed, the verification may degrade when the segmented geometry and vein region are mixed with inaccurate information.
3. Since the capturing device does not have any pegs or guiding bars, the finger alignment is varied in each collection. This situation will result in the finger image being affected by rotational issues, which can hamper accurate verification. Hence, the solution is to use geometry solution approach for rotational alignment.
In addition, according to (Dukes et al, 2013), it is an established fact that, in order to improve recognition performance, there is the need to combine several sources of, or types of, information either at the feature extraction stage or at the classification level, or even at both stages. Also, it was found that the areas of feature combination and the use of multi-classifier systems for Human verification system amounts need to be explored more (Ahmad and Mahmoud, 2013). Therefore, a deep learning method as Convolutional neural networks CNN (Kim, Yoon 2014 . Simonyan et al, 2014 . LeCun et al.2014) could improve the performance of Human verification system.
4 Research Objectives
The objectives of this research are:
1. To extract new effective texture features from both spatial and/or frequency domains to represent the finger vein image characteristics.
2. Design and implement a new reliable finger vein verification system with Convolutional neural networks technique that improves the accuracy of the Human verification system using finger vein images.
5 Research Scope
The scope of this research is bounded by the following:
��� This research is focused on the literal of Human verification system using finger vein images processing.
��� SDUMLA-HTML Finger vein public database (3,816 images) is used to make this work comparable with previous research.
The study methods will be conducted according to the workflow process as illustrated in Figure 2.
Stage 1: Literature Review
This stage involves a literature review on the current status of Human verification system using finger vein images.
Stage 2: Feature extraction
Feature extraction by Discrete Wavelet Transform (DWT), Discrete slantlet Transform (DST), The coefficient of level 0, 1, and 2 are concatenated in a single feature matrix.
Stage 3: implementation of Convolutional neural networks
Deep Learning Convolutional neural networks method is used to perform both identification and verification
Stage 4: Evaluation and Benchmarking
Evaluate the performance using average Equal Error Rate (EER) and Comparison with state of the art in the field.
Stage 5: Report Writing and Submission
It is estimated that the project paper will comprise of the following
Figure 2 : Research Framework
Convolutional Neural Networks
Recently, deep learning approach has shown a decent performance in pattern recognition and computer vision tasks. It shows that the methods of Deep Learning are successfully used in several biomedical image analysis challenges, such as brain image segmentation, and mitosis detection [58]. Classical image analysis involves a series of steps including preprocessing, image segmentation, and careful selection of features, learning, and classification. Deep Learning performance of these methods is strongly relying on the carefully chosen features and the accuracy of the previous steps such as feature dimension and fully connected approach(LeCun et al.2014).
Convolutional Neural Network (CNN) model is considering as one of the most common modern types of the Deep Learning methods used in image processing. CNN is considered as one type of feed-forward artificial neural network. It is a learning network that allows multiple levels of representation and abstraction. CNN is a multi-layer perceptron that consists of input layer, convolution layer, down sampling layer, and output layer. In CNN architecture, the convolution and down sampling layers might consist of multiple layers(LeCun et al.2014).
Convolutional Neural Networks Architecture
A CNN contain three main types of layers: convolution layers, Max-pooling layers, and a fully connected layer(LeCun et al.2014).
A. Convolutional layer
Convolutional layer convolves the result of previous layer with a set of learnable filters as shown in Figure 3, where the weights specify the convolution filter. Each filter is slid through the height and width of the input volume, creating a 2-dimensional activation map of that filter. The filters have the same depth as in the input. The size of the output can be controlled by three hyperparameters which are the depth, stride and zero-padding.
1. Depth: it is basically the number of filters that is applied to the input image. These filters detect structure such as edges, corners, blobs etc. as shown in Figure 4.
2. Stride: number of pixels the filter jumps while sliding over the image.
3. Zero-padding: padding zeros around the borders of the input to preserve its size.
Figure 3 shows the processing example with depth = 1, filter size 3 �� 3, stride=2, padding =1. Where If l = 1 input image oppositely the output of the previous layer is convolved by filters.
Figure 3: Example of a convolutional layer consists of four filters.
Figure 4: Sample learned filters, image adapted from
B. Pooling layer
Pooling layer reduces the size of their input and allows multi-scale analysis. Carpooling and average-pooling are the most popular pooling operators (Figure 7). These operators compute the maximum or the average value within a small spatial block . Pooling with filters size of 2 �� 2 with a stride of 2 are considered ideal. Figure (5), and (6) illustrates max pooling operation with 2 �� 2 filters (LeCun et al.2014).
Figure 5: Sample Convolutional layer processing
Figure 6: Example of max 2 �� 2 pooling layer that it is used to decrease the spatial size of the image to decrease the number of parameters Furthermore computation in the network
Figure 7: The average or the maximum pooling layer
C. Fully-connected layer
Fully-connected layer connects to all the neurons of the previous layer (Figure 8). Fully connected layers are typically used as last layer of the network and perform the classification. A sample of CNN is depicted in Figure 6 which shows all the three previously demonstrated layers (LeCun et al.2014).
Figure 8: A sample of CNN architecture
7 Finger vein Database
Finger vein recognition is a recently developed research hotspot. SDUMLA-HTML a finger vein database which, to the best of our knowledge, is the first open finger vein database. The device used to capture finger vein images is designed by Joint Lab for Intelligent Computing and Intelligent Systems of Wuhan University. The capture device is illustrated in Figure 9. In the capturing process, each subject was asked to provide images of his/her index finger, middle finger and ring finger of both hands, and the collection for each of the 6 fingers is repeated for 6 times to obtain 6 finger vein images. Some sample images are shown in Figure 10.
The finger vein database is composed of 6��6��106=3,816 images. Every image is stored in ���bmp��� format with 320��240 pixels in size. The total size of our finger vein database is around 0.85G Bytes.
Figure 9: The data capture device.
Figure 10. Finger vein images. The six images shows the original finger vein images,
8 EVALUATION FUNCATION
Evaluation function is more important to see the performance and weakness of the system. In fact, there are several functions which address to measure the performance and the accuracy of the system. However, in this research one evaluations function namely EER (equal error rate) have been adopted.
Equal error rate (EER) is a biometric security system algorithm used to predetermine the threshold values for its false acceptance rate and its false rejection rate. When the rates are equal, the common value is referred to as the equal error rate. The value indicates that the proportion of false acceptances is equal to the proportion of false rejections. The lower the equal error rate value, the higher the accuracy of the biometric system.
9 IMPORTANT OF RESEARCH
Many people have become accustomed to using their finger to gain access to some resource. This is because many organizations are incorporating fingerprint recognition devices for authentication purposes. However, using your finger for authentication doesn’t need to rely on your fingerprints for unique identification. Newly developed biometric called finger vein recognition can authenticate a user by using a vascular network inside the finger.
Vascular networks inside a finger forms a pattern which is unique for each individual even twins, unaffected by aging or external factors such as cuts or wear and tear. Finger vein recognition is not the only vascular recognition biometric. Palm, retina, and back of hand biometrics all rely on vascular networks for identification. In order to capture an image of the vascular network, a substance called Hemoglobin must be present in the blood stream. Hemoglobin is responsible for carrying oxygen from the lungs to the rest of the body, thus this substance is present in all living human beings. Fake or dead fingers don’t have Hemoglobin in them, thus they will fail to be authenticated. Together with Hemoglobin near infrared light in used to capture the vascular network inside the finger. Infrared light is shone through the finger, thus making the vascular network appear as shadows on the image. Although shining the light through the finger produces the best results, these devices are quite large, since the light source must be above the finger at all times. The device has a closed housing as well since the user must place their finger into the device.
10 The Strength and Significance of the Research
The strength of this research is the apply the Deep Learning Convolutional neural networks method for Human verification system using finger vein images processing, and propose new system that can be apply for produce new machine for finger vein Human verification. The aim of this research is to enhance the Human verification system using finger vein feature by proposing deep learning Convolutional neural networks method (CNN) , the geometrical based and the combined feature extraction techniques with the deep learning method technique that improve the accuracy of the Human verification system using finger vein images.
In the process, inevitably gaps will be identified and recommendations made to further improve and strengthen of Human verification system using finger vein which is the significance of the research.
���
11 Key Milestones of the Research
The progress of the research will be monitored through five key milestones, as follows:
Key Milestone 1 (November 2017)
Completion of Literature Review about the proposed research.
Key Milestone 2 (February 2018)
Completion of Feature extraction stage and finish all the preprocessing steps for all the images in the dataset.
Key Milestone 3 (May 2018)
Completion of the implementation of Deep Learning Convolutional neural networks and apply it for identification and verification of Human verification system using finger vein
Key Milestone 4 (May 2018)
Evaluation and Benchmarking
Key Milestone 5 (May 2018)
Report Writing and Submission
The research activities are shown in Appendix 2.
Milestone Chart
Main milestones/phases shown on higher chart, and sub-milestones for each phase on charts below
Research Activity 2017 2018
7 8 9 10 11 12 1 2 3 4 5 6
Literature Study X X X
Key Milestone 1 K1
preprocessing steps and Feature extraction stage X X X
Key Milestone 2 K2
implementation of CNN X X X
Key Milestone 3 K3
Evaluation and Benchmarking X
Key Milestone 4 K4
Report Writing and Submission X X
Key Milestone 5 K5
Key 1 Completion of Literature Review about the proposed research.
Key 2 Completion of Feature extraction stage and preprocessing steps.
Key 3 Completion of the implementation of Deep Learning Convolutional neural networks
Key 4 Evaluation and Benchmarking
Key 5 Report Writing and Submission
���
Reference
(Singh et al.,2016) Singh, Richa, Arun Ross, and Kevin W. Bowyer. “Special issue on information fusion in biometrics.” Information Fusion 32 (2016): 1-2.
(Malik et al. 2013) Malik, Iram, and Rohini Sharma. “Analysis of different techniques for finger-vein feature extraction.” Int. J. Comput. Trends Technol.(IJCTT) 4, no. 5 (2013).
(Jain and Kumar, 2012)Jain, Anil K., and Ajay Kumar. “Biometric recognition: an overview.” Second generation biometrics: The ethical, legal and social context. Springer Netherlands, 2012. 49-79.
(Bolle et al.,2013) Bolle, Ruud M., Jonathan H. Connell, Sharath Pankanti, Nalini K. Ratha, and Andrew W. Senior. Guide to biometrics. Springer Science & Business Media, 2013.
(Haar et al.,2013) van de Haar, Helen, Darelle van Greunen, and Dalenca Pottas. “The characteristics of a biometric.” In Information Security for South Africa, 2013, pp. 1-8. IEEE, 2013.
(Nixon et al., 2015) Nixon, Mark S., Paulo L. Correia, Kamal Nasrollahi, Thomas B. Moeslund, Abdenour Hadid, and Massimo Tistarelli. “On soft biometrics.” Pattern Recognition Letters 68 (2015): 218-230.
(Agrawal. 2013) Agrawal, Anshita. “User authentication mechanisms on Android.” PhD diss., Doctoral dissertation, Indian Institute of Technology, Bombay, 2013.
Singh, S., et al. “A validation study of the CEMACH recommended modified early obstetric warning system (MEOWS).” Anaesthesia 67.1 (2012): 12-18.
(Yang et al. 2014) Yang, Lu, Gongping Yang, Yilong Yin, and Xiaoming Xi. “Exploring soft biometric trait with finger vein recognition.” Neurocomputing135 (2014): 218-228.
Kakkar, Namita, Neha Mehan, and Kriti Jandrotia. “Finger Vein Verification with the Help of Gabor Filter and Repeated Line Tracking.” International Journal of Advanced Research in Computer Science 4.11(2013).
(WANG et al, 2011) WANG, Can, Ting-ting LIU, and Hui-fen WANG. “Study on Dynamic Password Authentication System in Enterprise Data Acquisition.” Manufacture Information Engineering of China 1 (2011): 014.
Bonneau and Preibusch, 2010 Bonneau, Joseph, and S��ren Preibusch. “The Password Thicket: Technical and Market Failures in Human Authentication on the Web.” WEIS. 2010.
(Kono et al,2015) Kono, Miyuki, Naoto Miura, Takao Fujii, Koichiro Ohmura, Hajime Yoshifuji, Naoichiro Yukawa, Yoshitaka Imura et al. “Personal Authentication Analysis Using Finger-Vein Patterns in Patients with Connective Tissue Diseases���Possible Association with Vascular Disease and Seasonal Change.” PloS one 10, no. 12 (2015): e0144952.
(Yang et al,2014) Yang, Jinfeng, and Yihua Shi. “Towards finger-vein image restoration and enhancement for finger-vein recognition.” Information Sciences 268 (2014): 33-52.
Yang, Jinfeng, and Xu Li. “Efficient finger vein localization and recognition.” Pattern Recognition (ICPR), 2010 20th International Conference on. IEEE, 2010.
Liu, Zhi, and Shangling Song. “An embedded real-time finger-vein recognition system for mobile devices.” IEEE Transactions on consumer Electronics 58.2 (2012).
(Unar et al, 2014) Unar, J. A., Woo Chaw Seng, and Almas Abbasi. “A review of biometric technology along with trends and prospects.” Pattern recognition 47, no. 8 (2014): 2673-2688.
(Peng et al, 2014) Peng, Jialiang, Ahmed A. Abd El-Latif, Qiong Li, and Xiamu Niu. “Multimodal biometric authentication based on score level fusion of finger biometrics.” Optik-International Journal for Light and Electron Optics 125, no. 23 (2014): 6891-6897.
(Yang et al, 2014) Yang, Wenming, Xiaola Huang, Fei Zhou, and Qingmin Liao. “Comparative competitive coding for personal identification by using finger vein and finger dorsal texture fusion.” Information sciences 268 (2014): 20-32.
Asaari, Mohd Shahrimie Mohd, Shahrel A. Suandi, and Bakhtiar Affendi Rosdi. “Fusion of band limited phase only correlation and width centroid contour distance for finger based biometrics.” Expert Systems with Applications 41.7 (2014): 3367-3382.
(Dukes et al, 2013) Dukes, Kais, Eric Atwell, and Nizar Habash. “Supervised collaboration for syntactic annotation of Quranic Arabic.” Language resources and evaluation 47, no. 1 (2013): 33-62.
Kim, Yoon. “Convolutional neural networks for sentence classification.” arXiv preprint arXiv:1408.5882 (2014).
Simonyan, Karen, and Andrew Zisserman. “Very deep convolutional networks for large-scale image recognition.” arXiv preprint arXiv:1409.1556(2014).
(LeCun et al.2014) LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. “Deep learning.” Nature 521, no. 7553 (2015): 436-444.