# A Sparsified Densely Connected Network with Separable Convolution for Finger-Vein Recognition

^{*}

## Abstract

**:**

## 1. Introduction

- First, we proposed a low-memory overhead and sufficiently lightweight network model based on a group of sparsified dense connection (SDC) blocks. In each SDC block, the number of output feature channels is compressed as much as possible under the premise of retaining efficient information of the aggregated features.
- Second, to facilitate a smaller model size and faster convergence rate, we substituted the standard convolutional kernels with separable convolutional kernels, namely, a sequence assembly of $1\times 1$ point-wise convolution and depth-wise convolution was adopted.
- Finally, considering intra-class variation and inter-class similarity, we introduced a more robust margin penalty to strengthen the discriminative power of a softmax loss framework, which was defined on the geodesic distance of angular space.

## 2. Related Works

#### 2.1. Skip Connection and Sparsification

#### 2.2. Separable Convolution

## 3. Methodology

#### 3.1. Framework of SC-SDCN Model

#### 3.2. Details of SDC Blocks

#### 3.3. Loss Function and Training Strategy

#### 3.4. Evaluation Criteria

## 4. Experimental Results and Discussion

#### 4.1. Finger Vein Datasets

#### 4.2. Experimental Settings

#### 4.2.1. Splitting of Training/Test Set

#### 4.2.2. Network Initialization and Optimization

#### 4.3. Analysis of Different Splitting Ratios of Training/Test Set

#### 4.4. Analysis of Computational Cost

#### 4.5. Analysis of Loss Function

#### 4.6. Comparison with State-of-the-Art

## 5. Conclusions and Future Research

## Author Contributions

## Funding

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Khan, S.A.; Naaz, S. Comparative Analysis of Finger Vein, Iris and Human Body Odor as Biometric Approach in Cyber Security System. In Proceedings of the 2020 2nd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA), Bangalore, India, 5–7 March 2020; pp. 525–530. [Google Scholar]
- Shaheed, K.; Liu, H.; Yang, G.; Qureshi, I.; Gou, J.; Yin, Y. A Systematic Review of Finger Vein Recognition Techniques. Information
**2018**, 9, 213. [Google Scholar] [CrossRef] [Green Version] - Kumar, R.; Bharti, V. A Critical Review of Finger Vein Recognition Techniques for Human Identification. In Proceedings of the 2021 Third International Conference on Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, 2–4 September 2021; pp. 1–9. [Google Scholar]
- Shaheed, K.; Mao, A.; Qureshi, I.; Kumar, M.; Hussain, S.; Zhang, X. Recent advancements in finger vein recognition technology: Methodology, challenges and opportunities. Inf. Fusion
**2022**, 79, 84–109. [Google Scholar] [CrossRef] - Yao, Q.; Song, D.; Xu, X.; Zou, K. A Novel Finger Vein Recognition Method Based on Aggregation of Radon-Like Features. Sensors
**2021**, 21, 1885. [Google Scholar] [CrossRef] - Liu, F.; Yang, G.; Yin, Y.; Wang, S. Singular value decomposition based minutiae matching method for finger vein recognition. Neurocomputing
**2014**, 145, 75–89. [Google Scholar] [CrossRef] - Meng, X.; Zheng, J.; Xi, X.; Zhang, Q.; Yin, Y. Finger vein recognition based on zone-based minutia matching. Neurocomputing
**2021**, 423, 110–123. [Google Scholar] [CrossRef] - Miura, N.; Nagasaka, A.; Miyatake, T. Feature extraction of finger-vein patterns based on repeated line tracking and its application to personal identification. Mach. Vis. Appl.
**2004**, 15, 194–203. [Google Scholar] [CrossRef] - Huang, B.; Dai, Y.; Li, R.; Tang, D.; Li, W. Finger-Vein Authentication Based on Wide Line Detector and Pattern Normalization. In Proceedings of the 2010 20th International Conference on Pattern Recognition, Istanbul, Turkey, 23–26 August 2010; pp. 1269–1272. [Google Scholar]
- Miura, N.; Nagasaka, A.; Miyatake, T. Extraction Of Finger-vein Patterns Using Maximum Curvature Points In Image Profiles. Ieice Trans. Inf. Syst.
**2007**, e90-d, 1185–1194. [Google Scholar] [CrossRef] [Green Version] - Song, W.; Kim, T.; Kim, H.C.; Choi, J.H.; Kong, H.J.; Lee, S.R. A finger-vein verification system using mean curvature. Pattern Recognit. Lett.
**2011**, 32, 1541–1547. [Google Scholar] [CrossRef] - Syarif, M.A.; Ong, T.S.; Teoh, A.B.J.; Tee, C. Enhanced maximum curvature descriptors for finger vein verification. Multimed. Tools Appl.
**2017**, 76, 6859–6887. [Google Scholar] [CrossRef] - Yang, L.; Yang, G.; Xi, X.; Meng, X.; Zhang, C.; Yin, Y. Tri-Branch Vein Structure Assisted Finger Vein Recognition. IEEE Access
**2017**, 5, 21020–21028. [Google Scholar] [CrossRef] - Yang, L.; Yang, G.; Yin, Y.; Xi, X. Finger Vein Recognition With Anatomy Structure Analysis. IEEE Trans. Circuits Syst. Video Technol.
**2018**, 28, 1892–1905. [Google Scholar] [CrossRef] - Krishnan, A.; Thomas, T.; Mishra, D. Finger Vein Pulsation-Based Biometric Recognition. IEEE Trans. Inf. Forensics Secur.
**2021**, 16, 5034–5044. [Google Scholar] [CrossRef] - Huang, D.; Tang, Y.; Wang, Y.; Chen, L.; Wang, Y. Hand-Dorsa Vein Recognition by Matching Local Features of Multisource Keypoints. IEEE Trans. Cybern.
**2015**, 45, 1823–1837. [Google Scholar] [CrossRef] - Meng, X.; Yang, G.; Yin, Y.; Xiao, R. Finger Vein Recognition Based on Local Directional Code. Sensors
**2012**, 12, 14937–14952. [Google Scholar] [CrossRef] - Liu, H.; Yang, L.; Yang, G.; Yin, Y. Discriminative Binary Descriptor for Finger Vein Recognition. IEEE Access
**2018**, 6, 5795–5804. [Google Scholar] [CrossRef] - Hu, N.; Ma, H.; Zhan, T. Finger vein biometric verification using block multi-scale uniform local binary pattern features and block two-directional two-dimension principal component analysis. Optik
**2020**, 208, 163664. [Google Scholar] [CrossRef] - Wu, J.D.; Liu, C.T. Finger-vein pattern identification using principal component analysis and the neural network technique. Expert Syst. Appl.
**2011**, 38, 5423–5427. [Google Scholar] [CrossRef] - Zhang, L.; Sun, L.; Li, W.; Zhang, J.; Cai, W.; Cheng, C.; Ning, X. A Joint Bayesian Framework Based on Partial Least Squares Discriminant Analysis for Finger Vein Recognition. IEEE Sens. J.
**2022**, 22, 785–794. [Google Scholar] [CrossRef] - Kumar, A.; Zhou, Y. Human Identification Using Finger Images. IEEE Trans. Image Process.
**2012**, 21, 2228–2244. [Google Scholar] [CrossRef] - Yin, Y.; Liu, L.; Sun, X. SDUMLA-HMT: A Multimodal Biometric Database. In Biometric Recognition; Springer: Berlin/Heidelberg, Germany, 2011; pp. 260–268. [Google Scholar]
- Ye, Y.; Ni, L.; Zheng, H.; Liu, S.; Zhu, Y.; Zhang, D.; Xiang, W.; Li, W. FVRC2016: The 2nd Finger Vein Recognition Competition. In Proceedings of the 2016 International Conference on Biometrics (ICB), Halmstad, Sweden, 13–16 June 2016; pp. 1–6. [Google Scholar]
- Lu, Y.; Xie, S.; Yoon, S.; Yang, J.; Park, D. Robust Finger Vein ROI Localization Based on Flexible Segmentation. Sensors
**2013**, 13, 14339–14366. [Google Scholar] [CrossRef] - Asaari, M.S.M.; Suandi, S.A.; Rosdi, B.A. Fusion of band limited phase Only correlation and width centroid contour distance for finger based biometrics. Expert Syst. Appl.
**2014**, 41, 3367–3382. [Google Scholar] [CrossRef] - Xie, S.J.; Yang, J.; Yoon, S.; Yu, L.; Park, D.S. Guided Gabor Filter for Finger Vein Pattern Extraction. In Proceedings of the 2012 Eighth International Conference on Signal Image Technology and Internet Based Systems, Naples, Italy, 25–29 November 2012; pp. 118–123. [Google Scholar]
- Yang, G.; Xi, X.; Yin, Y. Finger Vein Recognition Based on (2D)
^{2}PCA and Metric Learning. J. Biomed. Biotechnol.**2012**, 2012, 324249. [Google Scholar] [CrossRef] [Green Version] - Lu, Y.; Wu, S.; Fang, Z.; Xiong, N.; Yoon, S.; Park, D.S. Exploring finger vein based personal authentication for secure IoT. Future Gener. Comput. Syst.
**2017**, 77, 149–160. [Google Scholar] [CrossRef] - Lu, Y.; Yoon, S.; Xie, S.J.; Yang, J.; Wang, Z.; Park, D.S. Efficient descriptor of histogram of salient edge orientation map for finger vein recognition. Appl. Opt.
**2014**, 53, 4585–4593. [Google Scholar] [CrossRef] - Huang, H.; Liu, S.; Zheng, H.; Ni, L.; Zhang, Y.; Li, W. DeepVein: Novel finger vein verification methods based on Deep Convolutional Neural Networks. In Proceedings of the 2017 IEEE International Conference on Identity, Security and Behavior Analysis (ISBA), New Delhi, India, 22–24 February 2017; pp. 1–8. [Google Scholar]
- Fairuz, S.; Habaebi, M.H.; Elsheikh, E.M.A. Finger Vein Identification Based On Transfer Learning of AlexNet. In Proceedings of the 2018 7th International Conference on Computer and Communication Engineering (ICCCE), Kuala Lumpur, Malaysia, 19–20 September 2018; pp. 465–469. [Google Scholar]
- Gumusbas, D.; Yildirim, T.; Kocakulak, M.; Acir, N. Capsule Network for Finger-Vein-based Biometric Identification. In Proceedings of the 2019 IEEE Symposium Series on Computational Intelligence (SSCI), Xiamen, China, 6–9 December 2019; pp. 437–441. [Google Scholar]
- Hou, B.; Yan, R. Convolutional Autoencoder Model for Finger-Vein Verification. IEEE Trans. Instrum. Meas.
**2020**, 69, 2067–2074. [Google Scholar] [CrossRef] - Zeng, J.; Wang, F.; Deng, J.; Qin, C.; Zhai, Y.; Gan, J.; Piuri, V. Finger Vein Verification Algorithm Based on Fully Convolutional Neural Network and Conditional Random Field. IEEE Access
**2020**, 8, 65402–65419. [Google Scholar] [CrossRef] - Choi, J.; Noh, K.J.; Cho, S.W.; Nam, S.H.; Owais, M.; Park, K.R. Modified Conditional Generative Adversarial Network-Based Optical Blur Restoration for Finger-Vein Recognition. IEEE Access
**2020**, 8, 16281–16301. [Google Scholar] [CrossRef] - Kuzu, R.S.; Piciucco, E.; Maiorana, E.; Campisi, P. On-the-Fly Finger-Vein-Based Biometric Recognition Using Deep Neural Networks. IEEE Trans. Inf. Forensics Secur.
**2020**, 15, 2641–2654. [Google Scholar] [CrossRef] - Ren, H.; Sun, L.; Guo, J.; Han, C.; Cao, Y. A high compatibility finger vein image quality assessment system based on deep learning. Expert Syst. Appl.
**2022**, 196, 116603.1–116603.12. [Google Scholar] [CrossRef] - Kuzu, R.S.; Maiorana, E.; Campisi, P. Vein-based Biometric Verification using Transfer Learning. In Proceedings of the 2020 43rd International Conference on Telecommunications and Signal Processing (TSP), Milan, Italy, 7–9 July 2020; pp. 403–409. [Google Scholar]
- Ou, W.F.; Po, L.M.; Zhou, C.; Rehman, Y.A.U.; Xian, P.F.; Zhang, Y.J. Fusion loss and inter-class data augmentation for deep finger vein feature learning. Expert Syst. Appl.
**2021**, 171, 114584. [Google Scholar] [CrossRef] - Zhang, Z.; Tang, Z.; Wang, Y.; Zhang, Z.; Yan, S.; Wang, M. Compressed DenseNet for Lightweight Character Recognition. arXiv
**2019**, arXiv:1912.07016. [Google Scholar] - He, Y. A New Lightweight DenseNet Based on Mix-Structure Convolution. IOP Conf. Ser. Mater. Sci. Eng.
**2020**, 790, 012113. [Google Scholar] [CrossRef] [Green Version] - Fang, Y.; Wu, Q.; Kang, W. A novel finger vein verification system based on two-stream convolutional network learning. Neurocomputing
**2018**, 290, 100–107. [Google Scholar] [CrossRef] - Zhao, D.; Ma, H.; Yang, Z.; Li, J.; Tian, W. Finger vein recognition based on lightweight CNN combining center loss and dynamic regularization. Infrared Phys. Technol.
**2020**, 105, 103221. [Google Scholar] [CrossRef] - Shen, J.; Liu, N.; Xu, C.; Sun, H.; Xiao, Y.; Li, D.; Zhang, Y. Finger Vein Recognition Algorithm Based on Lightweight Deep Convolutional Neural Network. IEEE Trans. Instrum. Meas.
**2022**, 71, 1–13. [Google Scholar] [CrossRef] - He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
- Huang, G.; Liu, Z.; Van Der Maaten, L.; Weinberger, K.Q. Densely Connected Convolutional Networks. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017; pp. 2261–2269. [Google Scholar]
- Wan, K.; Min, S.J.; Ryoung, P.K. Multimodal Biometric Recognition Based on Convolutional Neural Network by the Fusion of Finger-Vein and Finger Shape Using Near-Infrared (NIR) Camera Sensor. Sensors
**2018**, 18, 2296. [Google Scholar] - Tang, S.; Zhou, S.; Kang, W.; Wu, Q.; Deng, F. Finger vein verification using a Siamese CNN. IET Biom.
**2019**, 8, 306–315. [Google Scholar] [CrossRef] - Shaheed, K.; Mao, A.; Qureshi, I.; Kumar, M.; Hussain, S.; Ullah, I.; Zhang, X. DS-CNN: A pre-trained Xception model based on depth-wise separable convolutional neural network for finger vein recognition. Expert Syst. Appl.
**2022**, 191, 116288. [Google Scholar] [CrossRef] - Song, J.M.; Kim, W.; Park, K.R. Finger-Vein Recognition Based on Deep DenseNet Using Composite Image. IEEE Access
**2019**, 7, 66845–66863. [Google Scholar] [CrossRef] - Noh, K.J.; Choi, J.; Hong, J.S.; Park, K.R. Finger-Vein Recognition Based on Densely Connected Convolutional Network Using Score-Level Fusion With Shape and Texture Images. IEEE Access
**2020**, 8, 96748–96766. [Google Scholar] [CrossRef] - Kuzu, R.S.; Maiorana, E.; Campisi, P. Vein-Based Biometric Verification Using Densely-Connected Convolutional Autoencoder. IEEE Signal Process. Lett.
**2020**, 27, 1869–1873. [Google Scholar] [CrossRef] - Ahmad Radzi, S.; Khalil-Hani, M.; Bakhteri, R. Finger-vein biometric identification using convolutional neural network. Turk. J. Electr. Eng. Comput. Sci.
**2016**, 24, 1863–1878. [Google Scholar] [CrossRef] - Lee, Y.H.; Khalil-Hani, M.; Bakhteri, R. FPGA-based finger vein biometric system with adaptive illumination for better image acquisition. In Proceedings of the 2012 International Symposium on Computer Applications and Industrial Electronics (ISCAIE), Kota Kinabalu, Malaysia, 3–4 December 2012; pp. 107–112. [Google Scholar]
- Hong, H.G.; Lee, M.B.; Park, K.R. Convolutional Neural Network-Based Finger-Vein Recognition Using NIR Image Sensors. Sensors
**2017**, 17, 1297. [Google Scholar] [CrossRef] [Green Version] - Hu, H.; Kang, W.; Lu, Y.; Fang, Y.; Liu, H.; Zhao, J.; Deng, F. FV-Net: Learning a finger-vein feature representation based on a CNN. In Proceedings of the 2018 24th International Conference on Pattern Recognition (ICPR), Beijing, China, 20–24 August 2018; pp. 3489–3494. [Google Scholar]
- Lu, Y.; Xie, S.; Wu, S. Exploring Competitive Features Using Deep Convolutional Neural Network for Finger Vein Recognition. IEEE Access
**2019**, 7, 35113–35123. [Google Scholar] [CrossRef] - Ton, B.T.; Veldhuis, R.N.J. A high quality finger vascular pattern dataset collected using a custom designed capturing device. In Proceedings of the 2013 International Conference on Biometrics (ICB), Madrid, Spain, 4–7 June 2013; pp. 1–5. [Google Scholar]
- Jalilian, E.; Uhl, A. Finger-vein recognition using deep fully convolutional neural semantic segmentation networks: The impact of training data. In Proceedings of the 2018 IEEE International Workshop on Information Forensics and Security (WIFS), Hong Kong, China, 11–13 December 2018; pp. 1–8. [Google Scholar]
- Yang, W.; Hui, C.; Chen, Z.; Xue, J.; Liao, Q. FV-GAN: Finger Vein Representation Using Generative Adversarial Networks. IEEE Trans. Inf. Forensics Secur.
**2019**, 14, 2512–2524. [Google Scholar] [CrossRef] [Green Version] - Hou, B.; Yan, R. Triplet-Classifier GAN for Finger-Vein Verification. IEEE Trans. Instrum. Meas.
**2022**, 71, 1–12. [Google Scholar] [CrossRef] - Das, R.; Piciucco, E.; Maiorana, E.; Campisi, P. Convolutional Neural Network for Finger-Vein-Based Biometric Identification. IEEE Trans. Inf. Forensics Secur.
**2019**, 14, 360–373. [Google Scholar] [CrossRef] [Green Version] - Xie, C.; Kumar, A. Finger vein identification using Convolutional Neural Network and supervised discrete hashing. Pattern Recognit. Lett.
**2019**, 119, 148–156. [Google Scholar] [CrossRef] - Hou, B.; Yan, R. ArcVein-Arccosine Center Loss for Finger Vein Verification. IEEE Trans. Instrum. Meas.
**2021**, 70, 1–11. [Google Scholar] [CrossRef] - Yeh, J.; Chan, H.T.; Hsia, C.H. ResNeXt with Cutout for Finger Vein Analysis. In Proceedings of the 2021 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS), Hualien, Taiwan, 16–19 November 2021; pp. 1–2. [Google Scholar]
- Yang, W.; Huang, X.; Zhou, F.; Liao, Q. Comparative competitive coding for personal identification by using finger vein and finger dorsal texture fusion. Inf. Sci.
**2014**, 268, 20–32. [Google Scholar] [CrossRef] - Srivastava, R.K.; Greff, K.; Schmidhuber, J. Highway Networks. arXiv
**2015**, arXiv:1505.00387. [Google Scholar] - Huang, G.; Sun, Y.; Liu, Z.; Sedra, D.; Weinberger, K.Q. Deep Networks with Stochastic Depth. In Proceedings of the Computer Vision—ECCV 2016, Amsterdam, The Netherlands, 11–14 October 2016; pp. 646–661. [Google Scholar]
- Hu, H.; Dey, D.; Giorno, A.D.; Hebert, M.; Bagnell, J.A. Log-DenseNet: How to Sparsify a DenseNet. arXiv
**2017**, arXiv:1711.00002. [Google Scholar] - Zhu, L.; Deng, R.; Maire, M.; Deng, Z.; Mori, G.; Tan, P. Sparsely Aggregated Convolutional Networks. In Proceedings of the Computer Vision—ECCV 2018, Munich, Germany, 8–14 September 2018; pp. 192–208. [Google Scholar]
- Chao, P.; Kao, C.Y.; Ruan, Y.; Huang, C.H.; Lin, Y.L. HarDNet: A Low Memory Traffic Network. In Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Republic of Korea, 27 October–2 November 2019; pp. 3551–3560. [Google Scholar]
- Szegedy, C.; Liu, W.; Jia, Y.; Sermanet, P.; Reed, S.; Anguelov, D.; Erhan, D.; Vanhoucke, V.; Rabinovich, A. Going deeper with convolutions. In Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA, 7–12 June 2015; pp. 1–9. [Google Scholar]
- Howard, A.G.; Zhu, M.; Chen, B.; Kalenichenko, D.; Wang, W.; Weyand, T.; Andreetto, M.; Adam, H. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. arXiv
**2017**, arXiv:1704.04861. [Google Scholar] - Chollet, F. Xception: Deep Learning with Depthwise Separable Convolutions. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017; pp. 1800–1807. [Google Scholar]
- Deng, J.; Guo, J.; Yang, J.; Xue, N.; Kotsia, I.; Zafeiriou, S. ArcFace: Additive Angular Margin Loss for Deep Face Recognition. IEEE Trans. Pattern Anal. Mach. Intell.
**2022**, 44, 5962–5979. [Google Scholar] - Qiu, S.; Liu, Y.; Zhou, Y.; Huang, J.; Nie, Y. Finger-vein recognition based on dual-sliding window localization and pseudo-elliptical transformer. Expert Syst. Appl.
**2016**, 64, 618–632. [Google Scholar] [CrossRef] - Qin, H.; El-Yacoubi, M.A. Deep Representation-Based Feature Extraction and Recovering for Finger-Vein Verification. IEEE Trans. Inf. Forensics Secur.
**2017**, 12, 1816–1829. [Google Scholar] [CrossRef] - Lu, X.; Tao, M.; Fu, X.; Gui, G.; Ohtsuki, T.; Sari, H. Lightweight Network Design Based on ResNet Structure for Modulation Recognition. In Proceedings of the 2021 IEEE 94th Vehicular Technology Conference (VTC2021-Fall), Norman, OK, USA, 27–30 September 2021; pp. 1–5. [Google Scholar]
- Li, X.; Duan, C.; Yin, P.; Wang, N. Pedestrian Re-identity Based on ResNet Lightweight Network. J. Phys. Conf. Ser.
**2021**, 2083, 032087. [Google Scholar] [CrossRef]

**Figure 4.**Illustration of the respective pair of sample images (original acquired image and corresponding ROI image) in both finger vein datasets.

**Figure 5.**DET curves of different splitting ratios of training/test set on two-finger vein datasets.

Feature | Method (Ref) | Dataset | Performance Criteria | ||
---|---|---|---|---|---|

Type | EER (%) | ACC (%) | |||

Vein-Level | Points | SVD-based Minutiae Matching [6] | HKPU [22] ^{a} | 5.01 | 95.71 |

SDUMLA [23] ^{b} | 2.46 | 98.63 | |||

Zone-based Minutia Matching [7] | HKPU [22] | 0.36 | 99.67 | ||

SDUMLA [23] | 2.61 | 96.27 | |||

Lines | Repeated Lines [8] | own ^{g} | 0.145 | – | |

Wide Line Detector (WLD) [9] | own | 0.87 | – | ||

Curvatures | Max Curvature [10] | own | 0.0009 | – | |

Mean Curvature [11] | own | 0.25 | – | ||

Enhanced Maximum Curvature [12] | SDUMLA [23] | 0.14 | – | ||

PKU [24] ^{c} | 0.33 | – | |||

Radon-like Features (RLF) [5] | HKPU [22] | 5.47 | – | ||

MMCBNU [25] ^{d} | 3.33 | – | |||

FV-USM [26] ^{e} | 0.93 | – | |||

Anatomies | Anatomy Structure Analysis based | HKPU [22] | 0.38 | – | |

Vein Extraction (ASAVE) [14] | SDUMLA [23] | 1.39 | – | ||

Pulsation of Veins [15] | Video data [15] ^{f} | 0.8 | – | ||

Image-Level | Local Pattern | Local Directional Code (LDC) [17] | MMCBNU [25] | 1.03 | – |

Discriminative Binary Descriptor [18] | HKPU [22] | 0.69 | – | ||

SDUMLA [23] | 1.89 | – | |||

Filtering | Guided Gabor Filter [27] | SDUMLA [23] | 2.24 | – | |

Gabor Filters and Morphological [22] | HKPU [22] | 0.65 | – | ||

Statistical | Principal Component Analysis (PCA) [20] | own | – | 99.0 | |

${\left(2D\right)}^{2}$PCA [28] | own | – | 99.17 | ||

Histogram of Competitive | MMCBNU [25] | 0.36 | – | ||

Orientations and Magnitudes (HCOM) [29] | |||||

Histogram of Salient Edge | MMCBNU [25] | 0.9 | – | ||

Orientation Map (HSEOM) [30] | |||||

HKPU [22] | 1.17 | – | |||

Partial Least Squares | SDUMLA [23] | 2.15 | 97.52 | ||

Discriminant Analysis (PLS-DA) [21] | MMCBNU [25] | 0.63 | – | ||

FV-USM [26] | 0.15 | 99.86 |

^{a}Finger Image Database from Hong Kong Polytechnic University (HKPU). http://www4.comp.polyu.edu.hk/~csajaykr/fvdatabase.htm, (accessed on 1 January 2021).

^{b}Homologous Multi-modal Traits Database (SDUMLA). http://mla.sdu.edu.cn/sdumla-hmt.html, (accessed on 1 January 2021).

^{c}Finger Vein Database from Peking University (PKU). http://rate.pku.edu.cn/, (accessed on 1 March 2019).

^{d}Finger Vein Database from Multimedia Lab of Chonbuk National University (MMCBNU). http://multilab.jbnu.ac.kr/MMCBNU_6000, (accessed on 1 January 2021).

^{e}Finger Vein Database from Universiti Sains Malaysia (FV-USM). http://drfendi.com/fv_usm_database/, (accessed on 1 January 2021).

^{f}Finger Vein Video Database (FV_IIITMK_VideoData). https://duk.ac.in/crictr/arya/sampledatapage.html, (accessed on 1 June 2022).

^{g}Some Self Built Finger Vein Databases.

Category | Method (Ref) | Dataset | Performance Criteria | ||
---|---|---|---|---|---|

EER (%) | ACC (%) | ||||

LeNet-5 CNN [54] | UTM [55] ^{a} | – | 99.0 | ||

DeepVein (VGGNet-16) [31] | FVRC2016 [24] ^{b} | 2.14 | – | ||

VGGNet-16 [56] | SDUMLA [23] | 0.804 | – | ||

SDUMLA [23] | 1.20 | – | |||

FV-Net (VGGFace-Net) [57] | MMCBNU [25] | 0.30 | – | ||

FV-USM [26] | 0.76 | – | |||

AlexNet [32] | Unknown | – | 91.67 | ||

CNN Competitive Order (CNN-CO) [58] | SDUMLA [23] | 2.37 | – | ||

MMCBNU [25] | 0.74 | – | |||

Capsule Network [33] | HKPU [22] | – | 88.0 | ||

SDUMLA [23] | – | 100.0 | |||

MMCBNU [25] | – | 100.0 | |||

UTFVP [59] ^{c} | – | 94.0 | |||

Classical | Convolutional Auto-Encoder (CAE) [34] | SDUMLA [23] | 0.21 | 99.78 | |

FV-USM [26] | 0.12 | 99.95 | |||

Fully Convolutional Network (FCN) [60] | SDUMLA [23] | 3.88 | – | ||

UTFVP [59] | 1.80 | – | |||

Without | Fully Convolutional Network+ | HKPU [22] | 2.37 | – | |

Skip | Conditional Random Field (FCN+CRF) [35] | SDUMLA [23] | 5.83 | – | |

Connection | MMCBNU [25] | 0.36 | – | ||

Generative Adversarial Networks | SDUMLA [23] | 0.94 | – | ||

(FV-GAN) [61] | |||||

Conditional GAN [36] | HKPU [22] | 1.81 | – | ||

SDUMLA [23] | 3.934 | – | |||

HKPU [22] | 0.40 | – | |||

Triplet-classifier GAN [62] | SDUMLA [23] | 1.33 | – | ||

FV-USM [26] | 0.14 | – | |||

Long Short-Term Memory Network | own | – | 99.10 | ||

(CNN-LSTM) [37] | |||||

HKPU [22] | – | 95.32 | |||

SDUMLA [23] | – | 97.48 | |||

Self Build | Das‘s Model [63] | FV-USM [26] | – | 97.53 | |

UTFVP [59] | – | 95.56 | |||

Two-Stream CNN [43] | SDUMLA [23] | 0.47 | – | ||

MMCBNU [25] | 0.10 | – | |||

Light CNN (LCNN) [64] | HKPU [22] | 0.13 | – | ||

Lightweight | Lightweight CNN Combining | MMCBNU [25] | 0.503 | 99.05 | |

Center Loss and Dynamic Regularization [44] | FV-USM [26] | 1.07 | 97.95 | ||

Lightweight Deep CNN [45] | SDUMAL [23] | 1.13 | 99.30 | ||

PKU [24] | 0.67 | 99.60 | |||

Multimodal with ResNet-101 [48] | HKPU [22] | 0.83 | – | ||

SDUMLA [23] | 2.43 | – | |||

SDUMLA [23] | 0.66 | – | |||

ResNet+Siamese CNN [49] | MMCBNU [25] | 0.12 | – | ||

FV-USM [26] | 0.30 | – | |||

ResNet | Efficient Channel Attention | HKPU [22] | 1.82 | 99.01 | |

Residual Network (ECA-Resnet) [65] | SDUMLA [23] | 2.14 | 98.91 | ||

With | FV-USM [26] | 0.89 | 99.42 | ||

Skip | ResNeXt-101 [66] | FV-USM [26] | – | 98.10 | |

Connection | Xception Model with | SDUMLA [23] | – | 98.50 | |

Depth-wise Separable CNN [50] | THU-FVFDT2 [67] ^{d} | – | 90.0 | ||

DenseNet | DenseNet-161+Composite Image [51] | HKPU [22] | 0.33 | – | |

SDUMLA [23] | 2.35 | – | |||

DenseNet-161+Score-level Fusion [52] | HKPU [22] | 0.05 | – | ||

SDUMLA [23] | 1.65 | – | |||

Densely-Connected | HKPU [22] | 0.228 | 99.67 | ||

Convolutional Autoencoder [53] | SDUMLA [23] | 0.025 | 99.98 |

^{a}Finger Vein Database from VeCAD Laboratory, University Technology Malaysia (UTM).

^{b}2nd Competition on Finger Vein Recognition Competition (FVRC2016). http://rate.pku.edu.cn, (accessed on 1 January 2019).

^{c}University of Twente Finger Vascular Pattern (UTFVP). http://www.sas.el.utwente.nl/home/datasets, (accessed on 1 January 2019).

^{d}Tsinghua University Finger Vein and Finger Dorsal Texture Database (THU-FVFDT). https://www.sigs.tsinghua.edu.cn/labs/vipl/thu-fvfdt.html, (accessed on 1 January 2021).

Layer | Kernel Size | Stride | Input Size | Output Size |
---|---|---|---|---|

Conv | $3\times 3$ | 2 | $224\times 224\times 3$ | $112\times 112\times 24$ |

Conv | $1\times 1$ | 1 | $112\times 112\times 24$ | $112\times 112\times 48$ |

DW-Conv | $3\times 3$ | 2 | $112\times 112\times 48$ | $56\times 56\times 48$ |

SDC block | $\left(\begin{array}{cc}1\times 1& Conv\\ \\ 3\times 3& DW-Conv\end{array}\right)$$\times 4$ | 1 | $56\times 56\times 48$ | $56\times 56\times 72$ |

Conv | $1\times 1$ | 1 | $56\times 56\times 72$ | $56\times 56\times 96$ |

DW-Conv | $3\times 3$ | 2 | $56\times 56\times 96$ | $28\times 28\times 96$ |

SDC block | $\left(\begin{array}{cc}1\times 1& Conv\\ \\ 3\times 3& DW-Conv\end{array}\right)$$\times 16$ | 1 | $28\times 28\times 96$ | $28\times 28\times 292$ |

Conv | $1\times 1$ | 1 | $28\times 28\times 292$ | $28\times 28\times 320$ |

DW-Conv | $3\times 3$ | 2 | $28\times 28\times 320$ | $14\times 14\times 320$ |

AvgPool | $14\times 14\times 320$ | $1\times 1\times 320$ | ||

Flatten | $1\times 1\times 320$ | 320 | ||

Dropout | 320 | 320 | ||

Linear | 320 | 1000 |

**Table 4.**Parameters of the first SDC block, which includes 4 combined layers, and each combined layer contains a point-wise convolution layer followed by a depth-wise separable convolution layer. Here, the output channel size of the corresponding combined layer is taking an integer result of $\mathit{k}\times {\mathit{m}}^{n}$, and ⊗ in the ‘Linked Layers’ column represents channel-level concatenation.

Layers | Linked Layers | Input Size | Kernels | Output Size | n | m | k |
---|---|---|---|---|---|---|---|

0 (Input) | – | – | – | $56\times 56\times 48$ | – | – | – |

1 | 0 | $56\times 56\times 48$ | PW: $16\times 1\times 1\times 48$ | $56\times 56\times 16$ | 0 | 1.6 | 16 |

DW: $3\times 3$ | |||||||

2 | 1⊗0 | $56\times 56\times 64$ | PW: $26\times 1\times 1\times 64$ | $56\times 56\times 26$ | 1 | ||

DW: $3\times 3$ | |||||||

3 | 2 | $56\times 56\times 26$ | PW: $16\times 1\times 1\times 26$ | $56\times 56\times 16$ | 0 | ||

DW: $3\times 3$ | |||||||

4 | 3⊗2⊗0 | $56\times 56\times 90$ | PW: $40\times 1\times 1\times 90$ | $56\times 56\times 40$ | 2 | ||

DW: $3\times 3$ | |||||||

5 (Output) | 4⊗3⊗1 | $56\times 56\times 72$ | – | – | – | – | – |

**Table 5.**Parameters of the second SDC blocks, which have 16 combined layers, and each combined layer contains a point-wise convolution layer followed by a depth-wise separable convolution layer. Here the output channel size of the corresponding combined layer is taking an integer result of $\mathit{k}\times {\mathit{m}}^{n}$.

Layers | Linked Layers | Input Size | Kernels | Output Size | n | m | k |
---|---|---|---|---|---|---|---|

0 (Input) | – | – | – | $28\times 28\times 96$ | – | – | – |

1 | 0 | $28\times 28\times 96$ | PW: $20\times 1\times 1\times 96$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

2 | 1⊗0 | $28\times 28\times 116$ | PW: $32\times 1\times 1\times 116$ | $28\times 28\times 32$ | 1 | ||

DW: $3\times 3$ | |||||||

3 | 2 | $28\times 28\times 32$ | PW: $20\times 1\times 1\times 32$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

4 | 3⊗2⊗0 | $28\times 28\times 148$ | PW: $52\times 1\times 1\times 148$ | $28\times 28\times 52$ | 2 | ||

DW: $3\times 3$ | |||||||

5 | 4 | $28\times 28\times 52$ | PW: $20\times 1\times 1\times 52$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

6 | 5⊗4 | $28\times 28\times 72$ | PW: $32\times 1\times 1\times 72$ | $28\times 28\times 32$ | 1 | ||

DW: $3\times 3$ | |||||||

7 | 6 | $28\times 28\times 32$ | PW: $20\times 1\times 1\times 32$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

8 | 7⊗6⊗4⊗0 | $28\times 28\times 200$ | PW: $82\times 1\times 1\times 200$ | $28\times 28\times 82$ | 3 | 1.6 | 20 |

DW: $3\times 3$ | |||||||

9 | 8 | $28\times 28\times 82$ | PW: $20\times 1\times 1\times 82$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

10 | 9⊗8 | $28\times 28\times 102$ | PW: $32\times 1\times 1\times 102$ | $28\times 28\times 32$ | 1 | ||

DW: $3\times 3$ | |||||||

11 | 10 | $20\times 28\times 32$ | PW: $20\times 1\times 1\times 32$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

12 | 11⊗10⊗8 | $28\times 28\times 134$ | PW: $52\times 1\times 1\times 134$ | $28\times 28\times 52$ | 2 | ||

DW: $3\times 3$ | |||||||

13 | 12 | $28\times 28\times 52$ | PW: $20\times 1\times 1\times 52$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

14 | 13⊗12 | $28\times 28\times 72$ | PW: $32\times 1\times 1\times 72$ | $28\times 28\times 32$ | 1 | ||

DW: $3\times 3$ | |||||||

15 | 14 | $28\times 28\times 32$ | PW: $20\times 1\times 1\times 32$ | $28\times 28\times 20$ | 0 | ||

DW: $3\times 3$ | |||||||

16 | 15⊗14⊗ | $28\times 28\times 282$ | PW: $132\times 1\times 1\times 282$ | $28\times 28\times 132$ | 4 | ||

12⊗8⊗0 | DW: $3\times 3$ | ||||||

16⊗15⊗13 | $28\times 28\times 292$ | – | – | – | – | – | |

17 (Output) | ⊗11⊗9⊗7 | ||||||

⊗5⊗3⊗1 |

Name | Subjects | Fingers | Classes | Samples/Class | Total Samples | Sessions | Orientations |
---|---|---|---|---|---|---|---|

index | |||||||

MMCBNU | 100 | middle | 600 | 10 | 6000 | 1 | right |

ring | |||||||

FV-USM | 123 | index middle | 492 | 12 | 5904 | 2 | down |

**Table 7.**ACC and EER results on the MMCBNU dataset with a different number of SDC blocks in the SC-SDCN architecture, in which ‘SL’ denotes a fully self-training procedure, and ‘TL’ denotes the pre-training on the ImageNet dataset and then transfers the pre-trained parameters to the target FV dataset for fine-tuning.

Architectures | Criteria | Splitting Ratio of Training/Test Set | Model | ||||
---|---|---|---|---|---|---|---|

9:1 | 8:2 | 7:3 | 6:4 | 5:5 | Size (M) | ||

2 SDC+SL | ACC | 0.9974 | 0.9968 | 0.9967 | 0.9953 | 0.9926 | 4.22 |

EER | 0.0034 | 0.0059 | 0.0059 | 0.0094 | 0.0155 | ||

2 SDC+TL | ACC | 0.9998 | 0.9982 | 0.9983 | 0.9981 | 0.9968 | |

EER | 0.0001 | 0.0032 | 0.0038 | 0.0031 | 0.0058 | ||

3 SDC+SL | ACC | 0.9985 | 0.9961 | 0.9961 | 0.9937 | 0.9932 | 12.1 |

EER | 0.0016 | 0.0066 | 0.0076 | 0.0113 | 0.0137 | ||

3 SDC+TL | ACC | 0.9992 | 0.9982 | 0.9981 | 0.9963 | 0.9971 | |

EER | 0.0009 | 0.0035 | 0.0036 | 0.0067 | 0.0061 | ||

4 SDC+SL | ACC | 0.9934 | 0.9926 | 0.9931 | 0.9886 | 0.9912 | 30.7 |

EER | 0.0115 | 0.0125 | 0.0135 | 0.0206 | 0.0149 | ||

4 SDC+TL | ACC | 0.9942 | 0.9951 | 0.9929 | 0.9896 | 0.9901 | |

EER | 0.0095 | 0.0121 | 0.0126 | 0.0194 | 0.0184 |

**Table 8.**ACC and EER results on the FV-USM dataset with a different number of SDC blocks in the SC-SDCN architecture.

Architectures | Criteria | Splitting Ratio of Training/Test Set | Model | ||||
---|---|---|---|---|---|---|---|

9:1 | 8:2 | 7:3 | 6:4 | 5:5 | Size (M) | ||

2 SDC+SL | ACC | 0.9938 | 0.9945 | 0.9924 | 0.9902 | 0.9932 | 4.22 |

EER | 0.0110 | 0.0112 | 0.0165 | 0.0234 | 0.0167 | ||

2 SDC+TL | ACC | 0.9974 | 0.9972 | 0.9963 | 0.9959 | 0.9962 | |

EER | 0.0045 | 0.0048 | 0.0069 | 0.0086 | 0.0082 | ||

3 SDC+SL | ACC | 0.9880 | 0.9894 | 0.9892 | 0.9883 | 0.9894 | 12.1 |

EER | 0.0201 | 0.0229 | 0.0217 | 0.0285 | 0.0269 | ||

3 SDC+TL | ACC | 0.9968 | 0.9956 | 0.9923 | 0.9946 | 0.9951 | |

EER | 0.0045 | 0.0093 | 0.0108 | 0.0122 | 0.0092 | ||

4 SDC+SL | ACC | 0.9885 | 0.9881 | 0.9876 | 0.9855 | 0.9857 | 30.7 |

EER | 0.0203 | 0.0231 | 0.0225 | 0.0282 | 0.0279 | ||

4 SDC+TL | ACC | 0.9840 | 0.9907 | 0.9898 | 0.9912 | 0.9895 | |

EER | 0.0225 | 0.0162 | 0.0142 | 0.0171 | 0.0181 |

Architectures | Model Size | Training | Feature Extraction |
---|---|---|---|

(M) | (s/epoch) | (ms/image) | |

2 SDC+TL | 4.22 | 4.4077 | 0.0651 |

3 SDC+TL | 12.1 | 5.8126 | 0.0959 |

4 SDC+TL | 30.7 | 6.6173 | 0.1122 |

Architectures | Model Size | Training | Feature Extraction |
---|---|---|---|

(M) | (s/epoch) | (ms/image) | |

2 SDC+TL | 4.22 | 2.0135 | 0.0656 |

3 SDC+TL | 12.1 | 2.6914 | 0.0958 |

4 SDC+TL | 30.7 | 3.0614 | 0.1118 |

**Table 11.**AAMP loss function compared with traditional softmax loss on the MMCBNU dataset. The ACC and EER are obtained after 200 epochs.

AAMP | Traditional Softmax | |||
---|---|---|---|---|

TL | SL | TL | SL | |

ACC | 0.9998 | 0.9974 | 0.9867 | 0.9828 |

EER | 0.0001 | 0.0034 | 0.0180 | 0.0254 |

**Table 12.**AAMP loss function compared with traditional softmax loss on the FV-USM dataset. The ACC and EER are obtained after 200 epochs.

AAMP | Traditional Softmax | |||
---|---|---|---|---|

TL | SL | TL | SL | |

ACC | 0.9974 | 0.9938 | 0.9897 | 0.9826 |

EER | 0.0045 | 0.0110 | 0.0165 | 0.0296 |

Category | Method | MMCBNU | FV-USM |
---|---|---|---|

Handcrafted | Local Directional Code (LDC) [17] | 1.03 | – |

Histogram of Salient Edge | 0.9 | – | |

Orientation Map (HSEOM) [30] | |||

2D-PCA [77] | – | 2.32 | |

Histogram of Competitive | 0.36 | – | |

Orientations and Magnitudes (HCOM) [29] | |||

Radon-like Features (RLF) [5] | 3.33 | 0.93 | |

Partial Least Squares | 0.63 | 0.15 | |

Discriminant Analysis (PLS-DA) [21] | |||

Deep learning | FCN+Segmentation [78] | – | 1.42 |

Two-stream CNN [43] | 0.1 | – | |

CNN Competitive Order (CNN-CO) [58] | 0.74 | – | |

Convolutional Auto-Encoder (CAE) [34] | – | 0.12 | |

Lightweight CNN Combining | 0.503 | 1.07 | |

Center Loss and Dynamic Regularization [44] | |||

Our Proposed SC-SDCN | 0.01 | 0.45 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Yao, Q.; Xu, X.; Li, W.
A Sparsified Densely Connected Network with Separable Convolution for Finger-Vein Recognition. *Symmetry* **2022**, *14*, 2686.
https://doi.org/10.3390/sym14122686

**AMA Style**

Yao Q, Xu X, Li W.
A Sparsified Densely Connected Network with Separable Convolution for Finger-Vein Recognition. *Symmetry*. 2022; 14(12):2686.
https://doi.org/10.3390/sym14122686

**Chicago/Turabian Style**

Yao, Qiong, Xiang Xu, and Wensheng Li.
2022. "A Sparsified Densely Connected Network with Separable Convolution for Finger-Vein Recognition" *Symmetry* 14, no. 12: 2686.
https://doi.org/10.3390/sym14122686