臺灣博碩士論文加值系統

當今社會的生物辨識技術已經有諸多的應用市場，比如機場的通關系統、Face ID、公家機關、公司機構的門禁系統等等，皆可以利用每個人與生俱來不同的生理特徵來做為屬於每個人自己的安全鎖，進而保障個人隱私財務。
其中，靜脈辨識是在生物辨識的領域中，一個較具有隱密性且特徵獨特性的一個分支，其原因在於靜脈屬於人類內部的生理特徵，且必須經由特殊波段的光線照射才能取得較為完整的靜脈影像，且每個人的靜脈紋路都不盡相同，做為辨識的主要依據足夠穩定。
近年來，神經網路技術迅速竄起，原因在於目前電腦技術先進，大數據(Big Data)取得以及處理相對容易，透過大數據建立的資料集，我們可以利用神經網路迅速地幫我們分析數據庫內含有的特徵，並作為我們分類的依據。而在生理辨識的議題上，使用者們通常不會希望自己的個人隱私上傳到大數據的雲端庫，這也造就了邊緣運算(Edge Computing)的議題被大量關注。
延伸以上這些概念，本論文提出一個低複雜度之卷積神經網路用於手指靜脈的辨識系統，Top-1 accuracy 可達95%。此一神經網路系統可以在用戶端獨立運作，透過樹莓派嵌入式開發板(Raspberry Pi Embedded Board)上安裝的近紅外線(near-infrared, NIR)的攝像頭取得使用者的手指影像後，可以在用戶端迅速的提取靜脈特徵並回傳使用者的身分為何。本系統的特點在於，透過設計卷積神經網路的矽智財(Silicon Intellectual Property)，跟650MHz工作頻率下的ARM Cortex-A9 dual core 在Linux環境下運行的CPU相比，縮短了在神經網路前向運算120倍的時間，達到了邊緣運算這個概念的實現。
模擬與驗證分別採用了山東大學機器學習與數據實驗室(Machine Learning and Data Mining Laboratory of Shandong University)的SDUMLA-HMT Database [27] 以及本實驗室自己開發的近紅外線靜脈影像截取裝置，兩者在本論文提出的低複雜度神經網路準確度可達95%，並實現10位受測者的手指靜脈辨識，由於複雜度低，神經網路前向運算可以達到即時運算，使整套系統離商品化更邁進一步。

Vein identification is a vital branch with more invisibility and unique features among biometric field. The reason is that veins belong to the physiological characteristics of human beings, and must be irradiated by specific bands of light to obtain a complete vein image. Also, taking vein image for identification has strong reliability for the reason that each person’s vein has its unique pattern.
Thanks to the advanced computer technology, neural network has rapidly become main trend of classification method for image classification. Through establishing big database, we can utilize neural network for analyzing features, thus become the classification basis of database. What’s more, users usually won’t want their personal information be uploaded to clouds. Therefore, edge computing has become a vital issue for the sake of protecting user’s privacy.
Inspired by these concepts, we proposed a low complexity convolutional neural network for finger vein recognition with top-1 accuracy of 95%. This neural network system can operate independently in client mode. After fetching the finger vein image of the user through the near-infrared camera mounted on Raspberry Pi embedded board, the vein feature can be efficiently extracted by vein curving algorithms and the identity of the user can quickly be returned.
In order to implement the concept of edge computing, our proposed system is characterized by designing a silicon intellectual property (SIP) for the purpose of shortening the inference time of neural network. Compared to the ARM Cortex-A9 dual core CPU running in Linux environment working at 650 MHz, 120x acceleration can be obtained.
Simulation and verification used the SDUMLA-HMT finger vein database [28] provided by Machine Learning and Data Mining Laboratory of Shandong University and our laboratory’s own database. Both dataset can reach 95% of accuracy for 10 users’ identity identification while inferenced by our low complexity neural network. Moreover, with low complexity, real-time identification can be achieved at the neural network inference side, and further close to commercialization.

誌謝 i
中文摘要 ii
Abstract iii
Contents v
List of Tables vii
List of Figures viii
Chapter 1 Introduction 1
1.1 Background 1
1.2 Motivation 3
1.3 Contributions 4
1.4 Thesis Organization 4
Chapter 2 Literature Review 5
2.1 Overview 5
2.2 Convolutional Neural Network 5
2.3 Optimizers 9
2.3.1 Stochastic Gradient Descent (SGD) 9
2.3.2 Momentum 9
2.3.3 RMSprop 10
2.3.4 Adaptive Moment Estimation (Adam) 10
2.4 CNN Frameworks 11
2.4 Vein Curving Methods 14
2.4.1 Repeated Line Tracking 14
2.4.2 Maximum Curvature 16
Chapter 3 System Architecture 18
3.1 Overview 18
3.2 Training Stage 19
3.2.1 Training Dataset 20
3.2.2 Image Pre-processing 21
3.2.3 Data Augmentation 22
3.2.4 Neural Network Training 24
3.3 Inference Stage 25
3.3.1 IP structure 27
3.3.2 Data Streaming Structure 28
3.3.3 General Matrix Multiplications (GEMMs) 29
3.3.4 Stream Convolution Slide Window 30
3.3.5 Stream Matrix Multiplication 31
3.3.6 Maxpool 32
3.3.7 Fully Connected 32
3.3.8 Jupyter Notebook 33
Chapter 4 Experimental Evaluation 35
4.1 Experimental Setting 35
4.1.1 Experimental Data 37
4.1.2 Experimental Design 38
4.2 Experimental Result 39
4.2.1 Laboratory Own Dataset 39
4.2.1.1 Impact of Input Image Size and Enhancement 39
4.2.1.2 Impact of Convolutional Filter Size and Layers 41
4.2.1.3 Impact of Batch Normalization 43
4.2.1.4 Impact of Optimizers 44
4.2.1.5 Impact of FPGA Acceleration 45
4.2.2 SDUMLA-HMT Database 46
4.3 Throughput Estimation 47
Chapter 5 Conclusion and Future Work 48
References 49

[1]C. W. Chang, “Palm-Vein Identification System using Ripple-Freeze Training on Deep Convolutional Network”, Master thesis, EE Department, National Chung Hsing University, 2019.
[2]Z. Zhou, E. Y. Du, N. L. Thomas, and E. J. Delp, “A new human identification method: sclera recognition,” IEEE Trans. Syst., Man, Cybern. A, vol. 42, no. 3, pp. 571–583, May 2012.
[3]A. Wahab, S. H. Chin and E. C. Tan, “Novel approach to automated fingerprint recognition," in IEE Proceedings - Vision, Image and Signal Processing, vol. 145, no. 3, pp. 160-166, June 1998.
[4]B. A. Rosdi, , C. W. Shing, and S. A. Suandi. “Finger vein recognition using local line binary pattern. ” Sensors, 2011, 11, 11357-11371.
[5]S. A. Radzi, M. Khalil-Hani, and R. Bakhteri. “Finger-vein biometric identification using convolutional neural network. ” Turkish Journal of Electrical Engineering & Computer Sciences. 2014.
[6]M. A. Turk and A. P. Pentland, “Face recognition using eigenfaces,” Proceedings. 1991 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Maui, HI, USA, 1991, pp. 586-591.
[7]K. Chellapilla, S. Puri, and P. Simard, “High performance convolutional neural networks for document processing,” in IWFHR’06.
[8]A. Krizhevsky, I. Sutskever, and G. Hinton. “Imagenet classification with deep convolutional neural networks.” In NIPS, 2012.
[9]K. Simonyan, and A. Zisserman. “Very deep convolutional networks for large-scale image recognition.” arXiv:1409.1556, 2014.
[10]K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” arXiv:1512.03385, 2015.
[11]G. Huang, Z. Liu, K. Q. Weinberger, and L. Maaten. “Densely connected convolutional networks.” In CVPR, 2017.
[12]M. Abadi, A. Agarwal et al., “Tensorflow: Large-scale machine learning on heterogeneous distributed systems,” arXiv:1603.04467, 2016.
[13]G. Hinton, N. Srivastava, and K. Swersky. Lecture 6a overview of mini–batch gradient descent. http://www.cs.toronto.edu/˜tijmen/csc321/slides/lecture_slides_lec6.pdf.
[14] D. Kingma, and J. Ba, “Adam: A method for stochastic optimization.” arXiv:1412.6980, 2014.
[15]N. Qian, “On the momentum term in gradient descent learning algorithms,” Neural Netw., vol. 12, no. 1, pp. 145–151, 1999.
[16]Y. Jia, et al., “Caffe: Convolutional architecture for fast feature embedding,” arXiv:1408.5093, 2014.
[17]M. Abadi, et al., “Tensorflow: A system for large-scale machine learning.” Tech. rep., Google Brain, 2016.
[18]J. Bergstra et al., “Theano: A CPU and GPU math compiler in python,”in Proc. 9th Python Sci. Comput. Conf. (SciPy), 2010, pp. 1–7.
[19]N. Miura, A. Nagasaka, and T. Miyatake, “Feature extraction of fingervein patterns based on repeated line tracking and its application to personal identification,” Mach. Vis. Appl., vol. 15, no. 4, pp. 194–203, Oct. 2004.
[20]N. Miura, A. Nagasaka, and T. Miyatake, “Extraction of finger-vein patterns using maximum curvature points in image profiles,” in Proc. IAPR Conf. Mach. Vis. Appl., Tsukuba Science City, Japan, May 2005, pp. 347–350.
[21]J. Bottleson, S. Kim, J. Andrews, P. Bindu, D. N. Murthy, and J. Jin,“clcaffe: Opencl accelerated caffe for convolutional neural networks,” in 2016 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW). IEEE, 2016, pp. 50–57.
[22]Intel FPGA. The Intel® FPGA SDK for Open Computing Language (OpenCL), 2016.
[23]S. Winograd. “Arithmetic complexity of computations,” vol. 33. Siam, 1980.
[24]Jong Hwan Ko, B. Mudassar, T. Na and S. Mukhopadhyay, "Design of an energy-efficient accelerator for training of convolutional neural networks using frequency-domain computation," 2017 54th ACM/EDAC/IEEE Design Automation Conference (DAC), Austin, TX, 2017, pp. 1-6.
[25]T. Kluyver et al., ‘‘Jupyter notebooks—A publishing format for reproducible computational workflows,’’ in Proc. ELPUB, 2016, pp. 87–90.
[26]Xilinx. Axi reference guide. 2011.
[27]C. He, Z. Li, L. Chen and J. Peng, "Identification of finger vein using neural network recognition research based on PCA," 2017 IEEE 16th International Conference on Cognitive Informatics & Cognitive Computing (ICCI*CC), Oxford, 2017, pp. 456-460.
[28]Y. Yin, L. Liu, and X. Sun, “SDUMLA-HMT: A multimodal biometric database,” in Proc. 6th Chin. Conf. Biometric Recognit., Beijing, China, Dec. 2011, pp. 260–268.
[29]B. Huang, Y. Dai, R. Li, D. Tang, and W. Li. Finger-vein authentication based on wide line detector and pattern normalization. In 20th International Conference on Pattern Recognition, pages 1269 –1272, 2010.
[30]Han, W.-Y. and Lee, J.-C. “Palm vein recognition using adaptive Gabor filter. Expert Systems with Applications,” vol. 39, no. 18, pp. 13225–13234, 2012.
[31]Wu, J.-D, and Liu, C-T. "Finger-vein pattern identification using principal component analysis and the neural network technique." Expert Systems with Applications, vol. 38, no. 5, pp. 5423–5427, 2011.
[32]E. C. Lee, H. C. Lee, and K. R. Park, “Finger vein recognition using minutia-based alignment and local binary pattern-based feature extraction,” Int. J. Imag. Syst. Technol., vol. 19, no. 3, pp. 179–186, 2009.
[33]E. C. Lee, H. Jung, and D. Kim. New finger biometric method using near infrared imaging. Sensors, vol. 11, no. 3, pp. 2319-2333, 2011.
[34]H. Wang, “Palm Vein Recognition System with Modified LBP”, Master thesis, EE Department, National Chung Hsing University, 2017.
[35]G. Yang, R. Xiao, Y. Yin, and L. Yang, “Finger vein recognition based on personalized weight maps,” Sensors, vol. 13, no. 9, pp. 12093–12112, 2013.
[36]S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML, 2015.
[37]Jia, Yangqing. Learning Semantic Image Representations at a Large Scale. PhD thesis, EECS Department, University of California, Berkeley, May 2014. http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-93.html.
[38]S. I. Venieris and C. Bouganis, "fpgaConvNet: A Framework for Mapping Convolutional Neural Networks on FPGAs," 2016 IEEE 24th Annual International Symposium on Field-Programmable Custom Computing Machines (FCCM), Washington, DC, 2016, pp. 40-47.
[39]E. Wang and J. J. Davis and P. Y. K. Cheung, "A PYNQ-based Framework for Rapid CNN Prototyping," In IEEE Symposium on Field-programmable Custom Computing Machines (FCCM), 2018.