近年來，機器學習和深度學習受到了高度的關注，特別是在與使用深度學習相關的分類，例如：資料探勘，人臉和語音辨識等。其中性能的提升主要是由於複雜的演算法和架結構，部分原因則歸功於好的數據資料。本論文的主要動機是將卷積神經網路(CNN)用於人臉辨識，其目的是透過轉移學習 (Transfer Learning)，使用新數據訓練預先訓練模型(Pre-trained Model)的方法，進而獲得正確的預測和準確的分類結果，在三個分類的訓練資料庫中，各有兩百張圖片。這個訓練資料庫是用來進一步訓練改良的預先訓練模型(pre-trained model) ，進而可以在不同的情境下測試資料庫影像，並在各別的情境下得到準確的預測輸出

Machine learning and deep learning particularly have gained a lot of attention in recent years, especially for classification related tasks, such as text mining, face and speech, etc. The performance increase is mostly due to complex algorithm and architecture, and partly due to the use of good data sets. The main motivation of this thesis is to train a Convolutional Neural Network (CNN) based system for face recognition aiming at positive prediction and appreciative accuracy result. By way of transfer learning, a pre-trained model can be tailored for different applications with new data. The resulting output attains good accuracy and result in different cases. The objective is to differentiate 3 labeled categories, each with 200 images in the training dataset. The training data is provided to modify the pre-trained model, which is further classified with the test images in different scenarios, where the prediction results achieve high accuracy for each individual case.

TABLE OF CONTENT
1. INTRODUCTION ........................................................................................................ 1
1.1 Objective ............................................................................................................... 2
1.2 Problem Description ............................................................................................... 2
1.3 Related Work .......................................................................................................... 3
1.4 Convolutional Neural Network (CNN) ................................................................. 3
1.5 Transfer learning ..................................................................................................... 5
2. LITERATURE REVIEW ................................................................................................. 6
2.1 Local receptive fields .............................................................................................. 8
2.2 Shared weights and biases ...................................................................................... 9
2.3 Activation and pooling ............................................................................................ 10
2.4 NEURAL NETWORK: ........................................................................................ 11
2.4.1 Forward Propagation ................................................................................. 12
2.4.2 Activation Function ................................................................................... 13
2.4.3 Back Propagation ....................................................................................... 14
2.5 Layer Type .............................................................................................................. 14
2.5.1 Convolutional Layer .................................................................................. 15
2.5.2 Max Pooling............................................................................................... 16
2.5.3 Fully Connected layer ................................................................................ 16
2.5.4 Overfitting .................................................................................................. 18
2.5.5 Dropout ...................................................................................................... 18
2.5.6 L2 Regularization ...................................................................................... 18
3. AlexNet Model: ............................................................................................................ 19
3.1 Convolutional layers ............................................................................................... 20
3.2 Max pooling ............................................................................................................ 24
3.3 Fully Connected Layer ............................................................................................ 25
3.4 ReLU (Rectified Linear Unit) layers ...................................................................... 26
3.5 Dropout ................................................................................................................... 27
3.6 Cross Channel Normalization ................................................................................. 27
3.7 Soft max and Classification layers .......................................................................... 28
3.8 Transfer learning ..................................................................................................... 28
3.9 Data Set ................................................................................................................... 30
3.9.1 Training data .............................................................................................. 30
3.9.2. Test data .................................................................................................... 30
3.10 Training Approach ................................................................................................ 32
4. SIMULATION RESULTS ............................................................................................. 39
4.1 Masked Face ........................................................................................................... 39
4.2 Face Cover .............................................................................................................. 44
4.3 Front face ................................................................................................................ 49
4.4 Side face .................................................................................................................. 52
5. CONCLUSION .............................................................................................................. 57
REFERENCES ............................................................................................................... 59
LIST OF FIGURES
Figure 1. Illustration of the BP process of training neural networks ............................................ 7
Figure 2. The performance of the trained neural networks with BP depends on the
initialization point. ........................................................................................................................... 8
Figure 3. Representation of a local receptive field ................................................................................... 9
Figure 4. The performance graph .................................................................................................................. 10
Figure 5. Architecture of a three-layer neural network .......................................................................... 12
Figure 6. Rectified Linear Unit (ReLU), i.e. g(net) =max (0, net). .................................................13
Figure 7. An illustration of a convolutional layer. .................................................................................. 15
Figure 8. Illustration of input (left) and output (right) of the max-pooling function,
With kw = kh = 2 and sw = sh = 2. ............................................................................................. 16
Figure 9. A fully connected layer, where every input neuron is connected to every
Output neuron. ................................................................................................................................... 17
Figure 10. Architecture of Alexnet ......................................................................................................... 19
Figure 11. Features obtained of convolutional layer 1 ............................................................................ 21
Figure 12. Features obtained on convolutional layer 2 ........................................................................... 22
Figure 13. Features obtained on convolutional layer 3 ........................................................................... 22
Figure 14. Features obtained on convolutional layer 4 ...................................................................... 23
Figure 15. Features obtained on convolutional layer 5 ........................................................................... 23
Figure 16. Images generated from the final fully connected layer 1
Correspond to the image classes ........................................................................................ 25
Figure 17. Fully connected layer 2 filters the image and gives the output ..................................... 26
Figure 18. Fully connected layer 3 used to store the detail of the image
that actually set-in dataset ............................................................................................................ 26
Figure 19. Pre-trained model Flowchart ....................................................................................................... 29
Figure 20. Modified layer in the pre-trained model ................................................................................. 29
Figure 21. Training face data ............................................................................................................................ 30
Figure 22. Masked image test data ....................................................................................... 31
Figure 23. Face cover test data ............................................................................................. 31
Figure 24. Front face test data .............................................................................................. 32
Figure 25. Side face test data ............................................................................................... 32
Figure 26. Accessing the model layer case 1 .............................................................................................. 39
Figure 27. Inputting the training data case 1 ............................................................................................... 40
Figure 28. Modification of the layer case 1 ................................................................................................. 40
Figure 29. Training the input data case 1 ..................................................................................................... 41
Figure 30. Training graph case 1 (Test 1) .................................................................................................. 41
Figure 31. Output result of case 1 (Test 1) ................................................................................................ 42
Figure 32. Training graph case 1 (Test 2) .................................................................................................. 42
Figure 33. Output result of case 1 (Test 2) ................................................................................................ 43
Figure 34. Training graph case 1 (Test 3) .................................................................................................. 43
Figure 35. Output result of case 1 (Test 3) ................................................................................................ 44
Figure 36. Accessing the model layer case 2 .............................................................................................. 44
9
Figure 37. Inputting the training data case 2 ............................................................................................... 45
Figure 38. Modification of the layer case 2 ................................................................................................. 45
Figure 39. Training the input data case 2 ..................................................................................................... 46
Figure 40. Training graph case 2 (Test 1) .................................................................................................. 46
Figure 41. Output result of case 2 (Test 1) ................................................................................................ 47
Figure 42. Training graph case 2 (Test 2) .................................................................................................. 47
Figure 43. Output result of case 2 (Test 2) ................................................................................................ 48
Figure 44. Training graph case 2 (Test 3) .................................................................................................. 48
Figure 45. Output result of case 2 (Test 3) ................................................................................................ 49
Figure 46. Accessing the model layer case 3 .............................................................................................. 49
Figure 47. Inputting the training data case 3 ............................................................................................... 50
Figure 48. Modification of the layer case 3 ................................................................................................. 50
Figure 49. Training the input data case 3 ..................................................................................................... 50
Figure 50. Training graph case 3 (Test 1) .................................................................................................. 51
Figure 51. Output result of case 3 (Test 1) ................................................................................................ 51
Figure 52. Accessing the model layer case 4 .............................................................................................. 52
Figure 53. Inputting the training data case 4 ............................................................................................... 52
Figure 54. Modification of the layer case 4 ................................................................................................. 53
Figure 55. Training the input data case 4 ..................................................................................................... 53
Figure 56. Training graph case 4 (Test 1) .................................................................................................. 54
Figure 57. Output result of case 4 (Test 1) ................................................................................................ 54

List of Tables
Table I. The visualize layers used to visualize the extracted feature ................................................ 20
Table II. Modified layer that depicts the visualize layers used to visualize
the extracted feature ............................................................................................................................. 29
Table III. Post Training output result information .................................................................................... 38
Table IV. Simulation results with their output label and accuracy .................................................... 55
Table V. Other approaches with their output and conclusion .............................................................. 56

[1] S.P. Bahurupi and D.S. Chaudhari, “Principal Component Analysis for Face Recognition,” International Journal of Engineering and Advanced Technology (IJEAT), pp.2249-8958, 2012. [2] H. Kong, X. Li, J.G. Wang, and C. Kambhamettu, “Ensemble LDA for face recognition,” In International Conference on Biometrics, Springer, Berlin, Heidelberg, pp.166-172, 2006. [3] A. Krizhevsky, I. Sutskever, and G.E. Hinton, “ImageNet classification with deep convolutional neural networks,” In Advances in neural information processing systems, pp. 1097-1105, 2012.
[4] H.B. Björgvinsdottir and R. Seibold, “Face recognition based on embedded systems,” Master's Thesis in Mathematical Sciences, Lund university, Sweden, 2016. [5] J. Schmidhuber, “Deep Learning in Neural Networks: An overview.” Neural networks 61, pp. 85-117, 2015. [6] J. Gu, Z. Wang, J. Kuen, L. Ma, A. Shahroudy, B. Shuai, T. Liu, X. Wang, L. Wang, G. Wang and J. Cai, “Recent advances in convolutional neural networks.” Pattern Recognition, 2017. [7] Z. Huang, Z. Pan, and B. Lei, “Transfer Learning with Deep Convolutional Neural Network for SAR Target Classification with Limited Labeled Data,” International journal of Remote Sensing, 9(9), pp. 907, 2017. [8] B.M. Lake, T.D. Ullman, J. B. Tenenbaum, and S.J. Gershman, “Building Machines That
Learn and Think Like People,” Behavioral and Brain Sciences 40, 2017. [9] A. Krizhevsky and G. Hinton, “Learning multiple layers of features from tiny images,” Vol. 1, No. 4, p. 7, University of Toronto, Canada, Apr 8, 2009. [10] P.J. Werbos, “Consistency of HDP applied to a simple reinforcement learning problem,” Neural networks, 3(2), pp. 179-189, 1990. [11] D.E. Rumelhart, G. E. Hinton, and R.J. Williams, “Learning Representations by Back-Propagating Errors,” Nature, 323(6088), pp. 533, 1986. [12] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, 86(11), pp. 2278-2324, 1989. [13] R.O. Duda, P.E. Hart, and D.G. Stork, Pattern Classification, John Wiley & Sons, 2012. [14] G.B. Huang, Z. Bai, L.L.C. Kasun, and C.M. Vong, “Local receptive fields based extreme learning machine,” IEEE Computational Intelligence Magazine, 10(2), pp. 18-29, 2015. [15] D. de Ridder, “Shared weights neural networks in image analysis, Master's thesis in Applied physics Department,” Delft University of Technology, Netherlands, March 1996. [16] C. Gulcehre, K. Cho, R. Pascanu, and Y. Bengio, “Learned-norm pooling for deep feedforward and recurrent neural networks,” In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Springer, Berlin, Heidelberg, pp. 530-546, 2014. [17] F. Manessi and A. Rozza, “Learning Combinations of Activation Functions,” arXiv preprint arXiv:1801.09403, 2018. [18] D. Chen, X. Cao, F. Wen, and J. Sun, “Blessing of dimensionality: High-dimensional
feature and its efficient compression for face verification,” In Computer Vision and Pattern Recognition (CVPR), 2013 IEEE Conference on, pp. 3025-3032, IEEE, 2013. [19] G.E. Hinton, S. Osindero, and Y.W. The, “A Fast Learning Algorithm for Deep Belief Nets,” Neural computation, 18(7), pp. 1527-1554, 2006. [20] V. Dumoulin, and F. Visin, “A guide to convolution arithmetic for deep learning,” arXiv preprint arXiv:1603.07285, 2016. [21] W. Zhou, S. Newsam, C. Li, and Z. Shao, “Learning Low Dimensional Convolutional Neural Networks for High-Resolution Remote Sensing Image Retrieval,” International journal of Remote Sensing, 9(5), pp. 489, 2017. [22] S. Han, J. Pool, J. Tran, and W. Dally, “Learning both weights and connections for efficient neural network,” In Advances in neural information processing systems, pp. 1135-1143, 2015. [23] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M.Bernstein and A.C. Berg, “Image Net Large Scale Visual Recognition Challenge,” International Journal of Computer Vision, 115(3), pp. 211-252, 2015. [24] Math Works, Alex Net, [online], Available: https://www.mathworks.com/help/nnet/ref/alexnet.html#bvn44n6, July 6, 2018. [25] K. Murphy, “Machine learning: a probabilistic approach.” Massachusetts Institute of Technology, pp. 1-21, 2012. [26] D.P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014.
[27] Math works, Training options [online], Available: https://www.mathworks.com/help/nnet/ref/trainingoptions.html , July 6, 2018.
[28] W. Wang, J. Yang, J. Xiao, S. Li, and D. Zhou, “Face recognition based on deep learning,” In International Conference on Human Centered Computing, Springer, Cham, pp. 812-820, 2014.
[29] U. Aiman and V. P. Vishwakarma, “Face recognition using modified deep learning neural network,” In Computing, Communication and Networking Technologies (ICCCNT), 2017 8th International Conference on, IEEE, pp. 1-5, 2017.
[30] H. Khalajzadeh, M. Mansouri, and M. Teshnehlab, “Face recognition using convolutional neural network and simple logistic classifier,” In Soft Computing in Industrial Applications, Springer, Cham, pp. 197-207,2014. [31] W. Wang, “Deep learning in object recognition, detection, and segmentation,” Foundations and Trends® in Signal Processing, 8(4), pp. 217-382, 2016.
[32] MathWorks, Visualize features of a convolutional neural network, [online], Available: https://www.mathworks.com/help/nnet/examples/visualize-features-of-a-convolutional-neural-network.html, July 6, 2018. [33] W. Ma and J. Lu, “An Equivalence of Fully Connected Layer and Convolutional Layer,” arXiv preprint arXiv:1712.01252, 2017.