Finger vein identification system using capsule networks with hyperparameter tuning

IAES International Journal of Artificial Intelligence (IJ-AI)
Vol. 12, No. 4, December 2023, pp. 1636~1643
ISSN: 2252-8938, DOI: 10.11591/ijai.v12.i4.pp1636-1643  1636
Journal homepage: http://ijai.iaescore.com
Finger vein identification system using capsule networks with
hyperparameter tuning
Vandy Achmad Yulianto, Nazrul Effendy, Agus Arif
Intelligent and Embedded System Research Group, Department of Nuclear Engineering and Engineering Physics,
Faculty of Engineering, Universitas Gadjah Mada, Yogyakarta, Indonesia
Article Info ABSTRACT
Article history:
Received Apr 28, 2022
Revised Jan 10, 2023
Accepted Mar 10, 2023
Safety and security systems are essential for personnel who need to be
protected and valuables. The security and safety system can be supported
using a biometric system to identify and verify permitted users or owners.
Finger vein is one type of biometric system that has high-level security. The
finger vein biometrics system has two primary functions: identification and
verification. Safety and security technology development is often followed by
hackers' development of science and technology. Therefore, the science and
technology of safety and security need to be continuously developed. The
paper proposes finger vein identification using capsule networks with
hyperparameter tuning. The augmentation, convolution layer parameters, and
capsule layers are optimized. The experimental results show that the capsule
network with hyperparameter tuning successfully identifies the finger vein
images. The system achieves an accuracy of 91.25% using the Shandong
University machine learning and applications-homologous multimodal traits
(SDUMLA-HMT) dataset.
Keywords:
Biometric
Capsule networks
Finger vein
Identification
Security
This is an open access article under the CC BY-SA license.
Corresponding Author:
Nazrul Effendy
Intelligent and Embedded System Research Group, Department of Nuclear Engineering and Engineering
Physics, Faculty of Engineering, Universitas Gadjah Mada
Jl. Grafika 2, Yogyakarta, Indonesia
Email: nazrul@ugm.ac.id
1. INTRODUCTION
Personal identification is essential in many systems' safety and security, such as in a building or
financial deposit system. Technology developments by hackers often follow the development of security
systems. It gives its motivation so that security technology continues to be developed. One of the identification
systems still developing is the biometric-based identification system [1]. Some biometric systems capable of
being used in the identification process are fingerprint, face, iris, speech, and finger vein [2], [3].
The finger vein-based biometric method has some advantages in the security system. The process is
suitable for identifying the authentication system that needs high accuracy and security. This finger vein image
pattern will differ when a person is alive or has died, making it more challenging to fake. The finger vein
images were acquired with tools utilizing a near-infrared camera system. Image acquirement uses a different
contrast principle because of the deoxygenizing process on the vein's blood flow. The blood absorbs more
infrared radiation than the area around it. Therefore, the finger veins will be darker than in other areas. The
difference can be further processed and finally classified using artificial intelligence.
Several artificial intelligence methods have been implemented in real problems [4]–[8]. One artificial
intelligence application identifies and verifies finger vein images [9]–[13]. Several available algorithms can be
implemented from the deep learning method, such as convolutional neural networks (CNN) [14]–[16]. In

Int J Artif Intell ISSN: 2252-8938 
Finger vein identification system using capsule networks with hyperparameter… (Vandy Achmad Yulianto)
1637
feature recognition, CNN has an advantage in its performance and accuracy. However, CNN generally has a
disadvantage because it considers spatial information in the feature extraction process [17]. When several
images have the same value in a particular area, it can increase the difficulty level of pattern recognition. CNN
will assume that the images are identical, although different [18], [19]. This paper proposes a finger vein
identification system using capsule networks with hyperparameter tuning to overcome the problem. We
optimized the capsule architectures and routing iterations at the capsule layer to find the best capsule network
system to identify the finger vein images.
The rest of this paper is described as follows. First, the research method is described in detail in
section 2. Then, section 3 describes the experimental results, including the training and testing of the finger
vein identification system using capsule networks with hyperparameter tuning. Finally, in section 4, the
conclusion of this study is offered.
2. RESEARCH METHOD
This paper shows the finger vein image identification system using capsule networks with
hyperparameter tuning, as shown in Figure 1. The system consists of data preprocessing and capsule networks
with hyperparameter tuning. We used the capsule networks with hyperparameter tuning as the classifier.
Figure 1. Finger vein images identification system using capsule networks with hyperparameter tuning
2.1. Data collection and preprocessing
We used finger vein digital images from a dataset of the Shandong University machine learning and
applications-homologous multimodal traits (SDUMLA-HMT) [20], [21]. The data was divided into three parts:
training set, validation set, and test set. Each class has six image data. We split the data of each type into three
sections, i.e., four images for training, one for validation, and two for evaluation [19], [20]. This data
preparation was carried out to prevent overfitting [22]. This research increased the training set by varying the
data images by rotation and translation transformations.
The finger veins images of one finger are chosen for each subject. The number of images is 636,
with 106 classes. The finger vein images were preprocessed in two stages: the extraction of the region of
interest (ROI) and image enhancement using contrast limited adaptive histogram equalization (CLAHE)
method [23]–[25]. Figure 2 shows the ROI extraction process. ROI extraction was conducted by determining
the exact area where the finger vein images were taken. The site is obtained by determining the location of the
finger edge contour. Cutting the ROI area produces an image of 180×100 pixels. This ROI extraction is
conducted to reduce class variations caused by image capture errors.
Figure 2. ROI extraction process [21]
The CLAHE method is implemented using open-source computer vision (OpenCV) library. OpenCV
is a library of many tools provided for dynamic image processing by Intel [26], [27]. CLAHE improves the
adaptive histogram equalization by applying the clip limit on the histogram to decrease the possibility of

 ISSN: 2252-8938
Int J Artif Intell, Vol. 12, No. 4, December 2023: 1636-1643
1638
contrast. Figure 3 shows an example of the input and output of finger vein image contrast enhancement using
CLAHE.
Figure 3. Example of the input and output of finger vein image contrast enhancement using CLAHE [21]
2.2. Design of capsule networks with hyperparameter tuning
This research implements capsule networks as the finger vein identification system [28], [29]. We
used several libraries in the Python language, such as tensor flow, Numpy, Pandas, Keras, and Scikit-learn
[22], [30]–[33]. The hyperparameter variations in the model are the augmentation, the parameters of the capsule
layer, and the convolution layer. There are variations in capsule architectures and routing iterations at the
capsule layer. The convolution layer is used as an image feature extraction. We varied the number of
convolution layers and other parameters such as stride, kernel size, and input image size.
The model was trained using Google Colaboratory [34]. Training on each variation is carried out to
get the best variation in each layer. The better model architecture will be rearranged into a new model using
the best parameters for each variation. Each training stage is optimized using an Adam optimizer with a learning
rate of 0.001 and evaluated using a margin loss function.
3. RESULTS AND DISCUSSION
The finger vein identification system's results and performance are explained as follows. We present
the performance of the baseline model, the image processing with ROI Extraction and CLAHE, and the
augmentation effect. We also offer the model's performance with routing iterations, the variation of the capsule
layer architecture, and the convolution layer variation.
3.1. Baseline model
The baseline model used as a reference was based on what Sabour et al. [35] and Hinton et al. [36]
proposed. The accuracy obtained from the model training shows that the baseline model cannot recognize all
images correctly. It can be improved by giving the batch normalization function to the convolution layer in the
model [37]. Batch normalization is used to normalize the input in the activation values by normalization of the
mean and variance values. The baseline model performance after the batch normalization is shown in
Figure 4. The model using batch normalization increases the accuracy of the baseline model.
Figure 4. The accuracy of the baseline model using a batch normalization [21]

1639
3.2. Image processing with ROI extraction and CLAHE
In this research, the image processing conducted using ROI extraction and CLAHE increases the
model accuracy. Data with image processing has more apparent and more precise features. It makes it easier
for the model to recognize the features. The test is conducted on two images representing each class. The
system will choose the highest value as the predicted output of the system.
3.3. Augmentation effect
The augmentation of the training set is conducted to decrease overfitting because of the relatively
small number of the training set. The augmentation uses the principle that each image batch is augmented for
each epoch so that each trained data will have a new variation for each epoch. We augmented the finger vein
images with rotation and translation transformations. The process has been performed with an angle of 5
degrees. The translation transformation is undertaken with five percent of the image's horizontal and vertical
position differences. The augmentation effects on the training loss of the model are shown in Figure 5.
Although achieving minimal error was a little slow, capsule network were successfully trained even by using
data augmentation in the training phase.
Figure 5. Training loss of capsule network model of finger vein identification with augmentation variation
3.4. Routing iterations and capsule layer architecture
This paper optimizes the number of routing iterations and the capsule layer architecture. The term
routing in capsule networks can be found in several references [38]–[40]. In this paper, the identification system
uses three variations in the number of routing iterations at the capsule layer, i.e., three, four, and five. Figure 6
shows the effects of several routing iterations on the validation loss. The system achieved the highest accuracy
with four routing iterations. Another variation of the capsule layer is its architecture. The capsule architecture
in the baseline is eight capsules in the primary capsule layer and 16 capsules in the digitcaps layer. Table 1
shows the variation of the capsule architecture.
As for the digitcaps layer, the number of computational complexities is influenced by the number of
classification classes, vector dimensions, and input matrix channels of the primary capsule layer. The effect of
the capsule architecture variation on the validation accuracy of the model is shown in Figure 7. The accuracy
of the interpretation of the capsule architecture gets the same results for architectures 1 and 2. Still, the accuracy
convergence on the baseline architecture occurs faster, and accuracy with architecture 2 is more challenging to
achieve stability. Optimal accuracy is obtained from the model that uses capsule base dimensions, namely eight
capsule dimensions on the main capsule layer and 16 vector dimensions on digitcaps with three routing
iterations.
3.5. Convolution layer variation
The number of computational complexities in the convolution layer depends on several factors,
including Kernal size, the number of features, stride number, and image size. Variations in the convolution
layer focus more on the influence of the kernel, stride, and number of layers. Table 2 shows the convolution

 ISSN: 2252-8938
1640
model of the capsule network's finger vein identification system. Kernel and stride size may affect the feature
extraction. The smaller the number of strides, the better the feature extraction is. Thus, the number of strides
in the variation is chosen, limiting the number of computational complexities still capable of hardware
computing. The sizes of the kernel are 3×3, 5×5, and 9×9. The larger the kernel size, the higher the
computational parameters needed. The effect of the convolution layer variation on the validation accuracy is
shown in Figure 8. The convolution model 3 achieves an accuracy of 91.25%, the highest in the experiments.
Figure 6. The effect of the number of routing iterations on the validation loss [21]
Table 1. Architecture type of capsule layer of the finger vein identification system
No Architecture type #Primary capsule layer #Digitcaps layer
1 Baseline architecture 8 16
2 Architecture 1 8 8
Figure 7. The effect of capsule architecture on the validation accuracy of the model [21]

1641
Table 2. Convolution model of the finger vein identification system using capsule networks
Convolution layer model Convolution layer Specification
Baseline Model First Layer Nine kernels, one stride, ReLu
Convolution Model 1 First Layer 3 kernels, 2 strides, ReLu, Batch Normalization
Second Layer 3 kernels, 2 strides, ReLu, Batch Normalization
Third Layer 3 kernels, 2 strides, ReLu, Batch Normalization
Convolution Model 2 First Layer 9 kernels, 2 strides, ReLu, Batch Normalization
Second Layer 5 kernels, 2 strides, ReLu, Batch Normalization
Third Layer 3 kernels, 1 stride, ReLu, Batch Normalization
Convolution Model 3 First Layer 3 kernels, 1 stride, ReLu, Batch Normalization
Second Layer 3 kernels, 1 stride, ReLu, Batch Normalization
Third Layer 3 kernels, 2 strides, ReLu, Batch Normalization
Fourth Layer 3 kernels, 2 strides, ReLu, Batch Normalization
Fifth Layer 3 kernels, 2 strides, ReLu, Batch Normalization
Figure 8. The effect of the convolution model type on the validation accuracy [21]
4. CONCLUSION
This paper describes a finger vein identification system as a security system. Hyperparameter tuning
was carried out on the capsule networks, including variations in the capsule networks' architecture and the
convolution layer. The number of routing iterations and image preprocessing was also investigated. The
capsule network's finger vein identification system achieved an accuracy of 91.25% using the SDUMLA-HMT
dataset.
ACKNOWLEDGMENTS
We thank the Faculty of Engineering, Universitas Gadjah Mada, for providing facilities and financial
support for this research. This research was funded by the Faculty of Engineering, Universitas Gadjah Mada,
decree number 417/UN1/FTK/SK/HK/2020.
REFERENCES
[1] J. Oh, U. Lee, and K. Lee, “Usability evaluation model for biometric system considering privacy concern based on MCDM model,”
Security and Communication Networks, vol. 2019, pp. 1–14, Mar. 2019, doi: 10.1155/2019/8715264.
[2] S. Nafisah and N. Effendy, “Voice biometric system: The identification of the severity of cerebral palsy using mel-frequencies
stochastics approach,” International Journal of Integrated Engineering, vol. 11, no. 3, pp. 194–206, Sep. 2019,
doi: 10.30880/ijie.2019.11.03.020.
[3] T. Zhao, Y. Liu, G. Huo, and X. Zhu, “A deep learning iris recognition method based on capsule network architecture,” IEEE
Access, vol. 7, pp. 49691–49701, 2019, doi: 10.1109/ACCESS.2019.2911056.
[4] S. N. Sembodo, N. Effendy, K. Dwiantoro, and N. Muddin, “Radial basis network estimator of oxygen content in the flue gas of
debutanizer reboiler,” International Journal of Electrical and Computer Engineering (IJECE), vol. 12, no. 3, pp. 3044–3050,
doi: 10.11591/ijece.v12i3.pp3044-3050.

 ISSN: 2252-8938
1642
[5] N. Effendy, N. C. Wachidah, B. Achmad, P. Jiwandono, and M. Subekti, “Power estimation of G.A. siwabessy multi-purpose
reactor at start-up condition using artificial neural network with input variation,” in Proceedings - 2016 2nd International
Conference on Science and Technology-Computer, ICST 2016, Oct. 2017, pp. 133–138, doi: 10.1109/ICSTC.2016.7877362.
[6] D. E. P. Lebukan, A. N. I. Wardana, and N. Effendy, “Implementation of plant-wide PI-fuzzy controller in tennessee eastman
process,” in Proceedings - 2019 International Seminar on Application for Technology of Information and Communication: Industry
4.0: Retrospect, Prospect, and Challenges, iSemantic 2019, Sep. 2019, pp. 450–454, doi: 10.1109/ISEMANTIC.2019.8884301.
[7] R. Y. Galvani, N. Effendy, and A. Kusumawanto, “Evaluating weight priority on green building using fuzzy AHP,” in Proceedings
- 12th SEATUC Symposium, SEATUC 2018, Mar. 2018, pp. 1–6, doi: 10.1109/SEATUC.2018.8788887.
[8] E. D. Kurniawan, N. Effendy, A. Arif, K. Dwiantoro, and N. Muddin, “Soft sensor for the prediction of oxygen content in boiler
flue gas using neural networks and extreme gradient boosting,” Neural Computing and Applications, vol. 35, no. 1, pp. 345–352,
Sep. 2023, doi: 10.1007/s00521-022-07771-8.
[9] H. Qin and M. A. El-Yacoubi, “Deep representation-based feature extraction and recovering for finger-vein verification,” IEEE
Transactions on Information Forensics and Security, vol. 12, no. 8, pp. 1816–1829, Aug. 2017, doi: 10.1109/TIFS.2017.2689724.
[10] H. Qin and M. A. El-Yacoubi, “Finger-vein quality assessment by representation learning from binary images,” in Lecture Notes in
Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9489,
Springer International Publishing, 2015, pp. 421–431.
[11] M. V Madhusudhan, V. Udaya Rani, and C. Hegde, “Finger vein recognition model for biometric authentication using intelligent
deep learning,” International Journal of Image and Graphics, Dec. 2022, doi: 10.1142/S0219467822400046.
[12] K. Shaheed, H. Liu, G. Yang, I. Qureshi, J. Gou, and Y. Yin, “A systematic review of finger vein recognition techniques,”
Information (Switzerland), vol. 9, no. 9, p. 213, Aug. 2018, doi: 10.3390/info9090213.
[13] D. Gumusbas, T. Yildirim, M. Kocakulak, and N. Acir, “Capsule network for finger-vein-based biometric identification,”
in 2019 IEEE Symposium Series on Computational Intelligence, SSCI 2019, Dec. 2019, pp. 437–441,
doi: 10.1109/SSCI44817.2019.9003019.
[14] I. Boucherit, M. O. Zmirli, H. Hentabli, and B. A. Rosdi, “Finger vein identification using deeply-fused convolutional neural
network,” Journal of King Saud University - Computer and Information Sciences, vol. 34, no. 3, pp. 646–656, Mar. 2022,
doi: 10.1016/j.jksuci.2020.04.002.
[15] N. Ahmed, A. Yigit, Z. Isik, and A. Alpkocak, “Identification of leukemia subtypes from microscopic images using convolutional
neural network,” Diagnostics, vol. 9, no. 3, p. 104, Aug. 2019, doi: 10.3390/diagnostics9030104.
[16] Kusrini, M. R. A. Yudianto, and H. Al Fatta, “The effect of Gaussian filter and data preprocessing on the classification of Punakawan
puppet images with the convolutional neural network algorithm,” International Journal of Electrical and Computer Engineering,
vol. 12, no. 4, pp. 3752–3761, Aug. 2022, doi: 10.11591/ijece.v12i4.pp3752-3761.
[17] T. H. Obaida, A. S. Jamil, and N. F. Hassan, “Real-time face detection in digital video-based on Viola-Jones supported by
convolutional neural networks,” International Journal of Electrical and Computer Engineering, vol. 12, no. 3, pp. 3083–3091, Jun.
2022, doi: 10.11591/ijece.v12i3.pp3083-3091.
[18] N. Effendy, D. Ruhyadi, R. Pratama, D. F. Rabba, A. F. Aulia, and A. Y. Atmadja, “Forest quality assessment based on bird sound
recognition using convolutional neural networks,” International Journal of Electrical and Computer Engineering (IJECE), vol. 12,
no. 4, pp. 4235–4242, doi: 10.11591/ijece.v12i4.pp4235-4242.
[19] R. Das, E. Piciucco, E. Maiorana, and P. Campisi, “Convolutional neural network for finger-vein-based biometric identification,”
IEEE Transactions on Information Forensics and Security, vol. 14, no. 2, pp. 360–373, Feb. 2018,
doi: 10.1109/TIFS.2018.2850320.
[20] Y. Yin, L. Liu, and X. Sun, “SDUMLA-HMT: A multimodal biometric database,” in Lecture Notes in Computer Science (including
subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 7098 LNCS, Springer Berlin
Heidelberg, 2011, pp. 260–268.
[21] V. A. Yuliantoro, “Finger vein biometric identification using capsule networks method (In Indonesia: Identifikasi biometrik vena
jari dengan metode capsule networks),” Universitas Gadjah Mada, 2020.
[22] L. Cunha, “Deep learning with Python (2a
ed) - François Chollet - Manning, outubro 2021, 504 pp.,” Interações: Sociedade e as
novas modernidades, no. 42, pp. 113–115, Jun. 2022, doi: 10.31211/interacoes.n42.2022.r1.
[23] U. Kuran and E. C. Kuran, “Parameter selection for CLAHE using multi-objective cuckoo search algorithm for image contrast
enhancement,” Intelligent Systems with Applications, vol. 12, p. 200051, Nov. 2021, doi: 10.1016/j.iswa.2021.200051.
[24] S. S. M. Sheet, T. S. Tan, M. A. As’ari, W. H. W. Hitam, and J. S. Y. Sia, “Retinal disease identification using upgraded CLAHE
filter and transfer convolution neural network,” ICT Express, vol. 8, no. 1, pp. 142–150, Mar. 2022, doi: 10.1016/j.icte.2021.05.002.
[25] R. A. Manju, G. Koshy, and P. Simon, “Improved method for enhancing dark images based on CLAHE and morphological
reconstruction,” Procedia Computer Science, vol. 165, pp. 391–398, 2019, doi: 10.1016/j.procs.2020.01.033.
[26] Z. Zhu and Y. Cheng, “Application of attitude tracking algorithm for face recognition based on OpenCV in the intelligent door
lock,” Computer Communications, vol. 154, pp. 390–397, Mar. 2020, doi: 10.1016/j.comcom.2020.02.003.
[27] I. H. Chen, S. C. Ho, and M. Bin Su, “Computer vision application programming for settlement monitoring in a drainage tunnel,”
Automation in Construction, vol. 110, p. 103011, Feb. 2020, doi: 10.1016/j.autcon.2019.103011.
[28] M. Kwabena Patrick, A. Felix Adekoya, A. Abra Mighty, and B. Y. Edward, “Capsule networks – a survey,” Journal of King Saud
University - Computer and Information Sciences, vol. 34, no. 1, pp. 1295–1310, Jan. 2022, doi: 10.1016/j.jksuci.2019.09.014.
[29] Y. Wang et al., “Capsule networks showed excellent performance in the classification of hERG blockers/nonblockers,” Frontiers
in Pharmacology, vol. 10, Jan. 2020, doi: 10.3389/fphar.2019.01631.
[30] M. J. J. Douglass, “Book review: Hands-on machine learning with scikit-learn, keras, and tensorflow, 2nd edition by aurélien géron,”
Physical and Engineering Sciences in Medicine, vol. 43, no. 3, pp. 1135–1136, Aug. 2020, doi: 10.1007/s13246-020-00913-z.
[31] F. Pedregosa FABIANPEDREGOSA et al., “Scikit-learn: Machine Learning in Python,” Journal of Machine Learning Research,
vol. 12, no. 85, pp. 2825–2830, 2011, [Online]. Available: http://jmlr.org/papers/v12/pedregosa11a.html.
[32] V. H. Nhu et al., “Effectiveness assessment of Keras based deep learning with different robust optimization algorithms for shallow
landslide susceptibility mapping at tropical area,” Catena, vol. 188, p. 104458, May 2020, doi: 10.1016/j.catena.2020.104458.
[33] S. Osah, A. A. Acheampong, C. Fosu, and I. Dadzie, “Deep learning model for predicting daily IGS zenith tropospheric delays in
West Africa using TensorFlow and Keras,” Advances in Space Research, vol. 68, no. 3, pp. 1243–1262, May 2021,
doi: 10.1016/j.asr.2021.04.039.
[34] B. Prashanth, M. Mendu, and R. Thallapalli, “Cloud based machine learning with advanced predictive analytics using Google
Colaboratory,” Materials Today: Proceedings, Mar. 2021, doi: 10.1016/j.matpr.2021.01.800.
[35] S. Sabour, N. Frosst, and G. E. Hinton, “Dynamic routing between capsules,” Oct. 2017, [Online]. Available:
http://arxiv.org/abs/1710.09829.

1643
[36] G. Hinton, S. Sabour, and N. Frosst, “Matrix capsules with EM routing.” 2018, [Online]. Available:
https://research.google/pubs/pub46653/.
[37] F. Schilling, “The effect of batch normalization on deep convolutional neural networks,” 2016.
[38] K. Ahmed and L. Torresani, “STAR-Caps: Capsule networks with straight-through attentive routing,” Advances in Neural
Information Processing Systems, vol. 32, 2019.
[39] V. Mazzia, F. Salvetti, and M. Chiaberge, “Efficient-CapsNet: capsule network with self-attention routing,” Scientific Reports,
vol. 11, no. 1, Jul. 2021, doi: 10.1038/s41598-021-93977-0.
[40] T. Hahn, M. Pyeon, and G. Kim, “Self-routing capsule networks,” Advances in Neural Information Processing Systems, vol. 32,
2019, [Online]. Available: http://vision.snu.ac.kr/projects/self-routing.
BIOGRAPHIES OF AUTHORS
Vandy Achmad Yulianto received a B.Eng. degree in Engineering Physics from the
Department of Nuclear Engineering and Engineering Physics, Faculty of Engineering, Universitas
Gadjah Mada, in 2020. He is a research assistant at the Intelligent and Embedded System Research
Group, Department of Nuclear Engineering and Engineering Physics, Faculty of Engineering,
Universitas Gadjah Mada. Currently, he is interested in researching machine learning and its
application in engineering. He can be contacted at email: vandy.achmad25@gmail.com.
Nazrul Effendy received a B.Eng. degree in Instrumentation Technology of Nuclear
Engineering and an M.Eng. degree in Electrical Engineering from Universitas Gadjah Mada in
1998 and 2001. He received a Ph.D. degree in Electrical Engineering from Chulalongkorn
University in 2009. He was a research fellow at the Department of Control and Computer
Engineering, the Polytechnic University of Turin, in 2010 and 2011 and a visiting researcher in
Shinoda Lab (Pattern Recognition & Its Applications to Real World), Tokyo Institute of
Technology in 2009. He is an Associate Professor and the coordinator of the Intelligent and
Embedded System Research Group in the Department of Nuclear Engineering and Engineering
Physics, Faculty of Engineering, Universitas Gadjah Mada. He is a member of the Indonesian
Association of Pattern Recognition, the Indonesian Society for Soft Computing, the Indonesian
Artificial Intelligence Society, and the International Association for Pattern Recognition. He can
be contacted at email: nazrul@ugm.ac.id.
Agus Arif received a B.Eng. degree in nuclear engineering from Universitas Gadjah
Mada in 1991 and an M.Eng. degree in Engineering Physics from Institut Teknologi Bandung in
2000. He was a research assistant in the Department of Electrical & Electronic Engineering,
Universiti Teknologi PETRONAS, Malaysia, from 2009 to 2011. He is currently an assistant
professor and a member of the Intelligent and Embedded System Research Group in the
Department of Nuclear Engineering and Engineering Physics, Faculty of Engineering, Universitas
Gadjah Mada. His research interests are instrumentation, control, machine learning, and
applications. He can be contacted at email: agusarif@ugm.ac.id.

Finger vein identification system using capsule networks with hyperparameter tuning

More Related Content

Similar to Finger vein identification system using capsule networks with hyperparameter tuning (20)

More from IAESIJAI (20)

Recently uploaded (20)

Finger vein identification system using capsule networks with hyperparameter tuning