本文是以深度正規化卷積式神經網路(DNCNN)為基礎的偽裝人臉辨識系統，此系統需要訓練兩組DNCNN辨識網路，第一組辨識網路的功能為辨識輸入人臉圖像的偽裝分類，該網路將人臉偽裝輸入圖像區分為三類，分別是無偽裝、上半臉偽裝及下半臉偽裝，在辨認完成之後，系統會將圖像中有偽裝的上半臉或下半臉部分移除後，再將無偽裝的半張臉圖像輸入到第二組辨識網路，而第二組辨識網路的功能就是辨識該輸入圖像的人員身分。在執行上述兩組DNCNN辨識網路的訓練與辨識之前，需先將人像的原始圖像樣本進行前處理。前處理採用Viola-Jones 臉部偵測演算法，先將人臉位置的區塊尋找出來，再擷取全臉方形圖像並切割成半張人臉圖像當作訓練圖像資料樣本，然而為了減少因為訓練樣本不足，而造成的過度訓練的問題，此系統以人臉圖像中心旋轉方式，產生更多的人臉圖像來強化辨識網路的訓練。在前處理完成後，將不同人臉圖像分類、收集與切割後，即可開始進行訓練與測試網路。由本篇實驗結果顯示，本系統對於偽裝人臉圖像辨識，與參考文獻的辨識率結果相近。

In this paper, we propose a disguised face recognition system based on Deep Normalization and Convolution Neural Network (DNCNN), this system include two trained DNCNN identification Network. The function of first trained identification network is to identify the type of disguised of the input face image. This network classifies human face disguised input images into three categories, No disguised, Upper half face disguised and Lower half face disguised. After the classification is completed, the system will remove the upper half disguised or the lower half disguised of the face image, and remaining the non-disguised half face images, then input it into the second recognition network. The function of the second recognition network is to recognize the identification of the input half face image.
To reduce the over-fitting caused by imbalanced and insufficient training samples. Before performing the training and identification of the above two   DNCNN recognition networks, we need to perform the pre-process on the original image samples first. The image pre-process is used the Viola-Jones face detection algorithm. The algorithm first finds out the block of the face position of original images, then the pre-process rotates and captures the face block image or half face images for the training and testing of recognition networks. After the preprocessing is completed, we can perform the training and testing of DNCNN recognition networks. The experimental results show that the system achieved a similar recognition rates as the reference.

目錄
中文摘要	I
英文摘要	III
目錄	V
圖目錄	VIII
表目錄	XI
第一章 緒論	1
1.1 研究背景	1
1.2 研究動機與目的	2
1.3 論文架構	3
第二章 卷積式神經網路	5
2.1. 神經網路	5
2.2. 卷積式神經網路的運作	8
2.2.1 卷積（Convolution）	10
2.2.2 線性整流單元 (ReLU / Rectified Linear Unit )	14
2.2.3 池化（Pooling）	16
2.2.4 深度學習(Deep learning)	17
2.2.5 全連接層(Fully Connected Layers)	19
2.2.6 反向傳播（Backpropagation）	21
2.3 以深度正規化及卷積式神經網路辨識煙霧影像	23
2.3.1 深度正規化及卷積式神經網路(DNCNN)	23
2.3.2 DNCNN的網路架構	24
2.3.3 圖像資料強化技術	26
2.3.4 以DNCNN網路為基礎的應用	27
第三章 偽裝人臉辨識系統之架構	29
3.1. 網路架構	29
3.2 辨識系統流程	29
3.3 訓練圖像之前處理	31
3.3.1 Viola-Jones 臉部偵測演算法	31
3.3.2 前處理之流程	32
3.3.3 臉部圖像之截取	34
3.3.4 身份辨識訓練圖像的強化處理	36
3.4 網路訓練	38
3.5 辨識網路的測試	39
第四章 實驗結果	40
4.1 實驗環境	40
4.2 資料庫	40
4.3 訓練與測試之人臉圖像庫	42
4.4 偽裝分類網路實驗結果	46
4.5 偽裝身分辨識網路實驗結果	48
第五章 結論與未來展望	53
參考文獻	54

圖目錄
圖2.1 單一神經元運算模型 [24]	6
圖2.2 多層神經元網路運算模型 [24]	8
圖2.3 CNN從低階層到全連接層 [26]	9
圖2.4 笑臉符號圖像的特徵 [26]	10
圖2.5 相對應像素的乘積 [26]	12
圖2.6 不同位置的卷積計算 [26]	12
圖2.7 不同特徵的卷積圖 [26]	13
圖2. 8 斜坡函數	14
圖2. 9 線性整流將負數轉化為零 [26]	14
圖2.10不同特徵的卷積圖經線性整流後之結果 [26]	15
圖2.11 池化的運作方式 [26]	16
圖2.12不同特徵的卷積圖被池化後之結果 [26]	17
圖2.13深度學習的流程 [26]	18
圖2.14 高、低階層所能辨認出的特徵 [26]	19
圖2.15 在全連接層進行投票 [26]	21
圖2.16反向傳播 [26]	23
圖2.17 DNCNN的網路架構 [29]	25
圖2.18 圖像資料強化處理範例 [29]	27
圖3.1辨識系統建立與測試流程	31
圖3.2訓練圖像之前處理	33
圖3.3 臉部圖像截取範例-1	34
圖3.4 臉部圖像截取範例-2	35
圖3.5 訓練偽裝分類網路所使用的圖像	35
圖3.6含有黑色三角形的訓練圖像	37
圖3.7 旋轉後的訓練圖像截取	37
圖3.8 旋轉不同角度後之半張臉訓練圖像	38
圖3.9訓練人臉辨識之DNCNN網路 [29]	39
圖4.1 資料庫的人像照片-1	41
圖4.2 資料庫的人像照片-2	41
圖4.3 資料庫的人像照片-3	41
圖4.4 資料庫的人像照片-4	42
圖4.5 無偽裝圖像	43
圖4.6 上半臉偽裝圖像	43
圖4.7 下半臉偽裝圖像	44
圖4.8 上半臉身分訓練圖像	45
圖4.9 下半臉身分訓練圖像	45
圖4.10 全臉及半臉測試圖像產生過程	46
圖4.11 實驗結果之CMC曲線	52

表目錄
表4.1 偽裝分類的辨識率-1	47
表4.2 偽裝分類的辨識率-2	48
表4.3 偽裝身分的辨識率	49
表4.4 上半臉、下半臉單獨訓練的辨識率	50
表4.5 訓練時降低學習率的辨識結果	51

參考文獻
[1] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," Proc. Int. Conf. Neural Inf. Process. Syst., p. 1097–1105, 2012. 
[2] A. Sharif Razavian , H. Azizpour, J. Sullivan, and S. Carlsson, "CNN features off-the-shelf: An astounding baseline for recognition," Proc. IEEE Conf. Comput. Vis. Pattern Recognit. Workshops, p. 512–519., 2014. 
[3] A. Babenko, A. Slesarev, A. Chigorin, and V. Lempitsky, “Neural codes for image retrieval,” Proc. Eur. Conf. Comput. Vis., pp. 584-599, 2014. 
[4] J. Ng, F. Yang, and L. Davis, "Exploiting local features from deep networks for image retrieval," Proc. IEEE Conf. Comput. Vis. Pattern Recog. Workshops, p. 53–61, 2015. 
[5] G. Tolias, R. Sicre, and H. J_egou, "Particular object retrieval with integral max-pooling of CNN activations," Proc. Int. Conf. Learn. Representations, 2016. 
[6] Liang Zheng, Yi Yang, and Qi Tian, "SIFT Meets CNN: A Decade Survey of Instance Retrieval," IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, Vols. 40, NO. 5, May 2018. 
[7] "AlphaGo," DeepMind, [Online]. Available: https://deepmind.com/research/alphago/.
[8] Alvin, "人機大戰再敗 李世乭：AlphaGo讓我開始挑戰對圍棋的傳統想法," The News Lens 關鍵評論, 15 3 2016. [Online]. Available: https://www.thenewslens.com/article/38191.
[9] 張樵蘇, "柯潔決戰阿爾法狗勝者獎金150萬美元," 新華社 新華網, 10 4 2017. [Online]. Available: http://www.xinhuanet.com/sports/2017-04/10/c_1120783734.htm.
[10] B. B. (Intel), "Archived - Code Sample: Facial Recognition Using Intel® RealSense™ SDK," Intel, 16 March 2015. [Online]. Available: https://software.intel.com/en-us/articles/code-sample-facial-recognition-using-intel-realsense-sdk.
[11] Xiaoqing Zhang, Dong Ren, “A Survey of Approaches to Feature Extraction in Face Recognition,” International Conference on Computer Science and Network Technology, 2012. 
[12] Billy Y.L. Li, Ajmal S. Mian, "Using Kinect for Face Recognition Under Varying Poses, Expressions, Illumination and Disguise," Applications of Computer Vision (WACV), vol. 7, no. 5, pp. 1498-1510, 2013. 
[13] Sheng G., Kai W. (Intel), "Archived - Intel® RealSense™ Archived - Intel® RealSense SDK-Based Real-Time Face Tracking and Animation," Intel, 14 11 2016. [Online]. Available: https://software.intel.com/en-us/articles/intel-realsense-sdk-based-real-time-face-tracking-and-animation.
[14] MoneyDJ, “臉部辨識將成智慧手機標配？2020 年估達 10 億支導入,” TechNews 科技新報 , 9 2 2018. [線上]. Available: https://technews.tw/2018/02/09/more-than-one-billion-smartphones-to-feature-facial-recognition-in-2020/.
[15] 蘋果仁, "除了 iPhone X，其他廠牌也有臉部辨識，但你知道差在哪裡嗎？," 蘋果仁, 20 10 2017. [Online]. Available: https://applealmond.com/posts/15291.
[16] T. News, "iPhoneX臉部辨識破解？CNN找雙胞胎測試," TVBS News, 1 11 2017. [Online]. Available: https://news.tvbs.com.tw/fun/803572.
[17] 劉哲君、吳青駿, "2對雙胞胎實測iPhoneX臉部辨識 一秒解鎖…傻眼了," SETN 三立新聞網, 3 11 2017. [Online]. Available: http://www.setn.com/News.aspx?NewsID=311345.
[18] 雷鋒網, "人臉辨識準確率黑人比白人差很多，主要受限於資料庫與光影呈現問題," TechNews 科技新報, 12 2 2018. [Online]. Available: https://technews.tw/2018/02/12/facial-recognition-is-accurate-if-youre-a-white-guy/.
[19] 蘋果仁, "Face ID 戴安全帽也能解鎖嗎？各種生活情境測試結果出爐！," 蘋果仁, 3 11 2017. [Online]. Available: https://applealmond.com/posts/16276.
[20] T. I. Dhamecha, A. Nigam, R. Singh, and M. Vatsa, "Disguise Detection and Face Recognition in Visible and Thermal Spectrums," Proceedings of International Conference on Biometrics, 2013. 
[21] T. I. Dhamecha, R. Singh, M. Vatsa, and A. Kumar, “Recognizing Disguised Faces: Human and Machine Evaluation,” PLoS ONE, 2014. 
[22] 黃彪鈺, “偽裝人臉辨識之預處理系統,” 於 淡江大學電機工程系碩士班（通訊與波組）碩士論文, 2016. 
[23] Mahmood Sharif, Sruti Bhagavatula, Lujo Bauer, Michael K. Reiter, "Adversarial Generative Nets: Neural Network Attacks on State-of-the-Art Face Recognition," arXiv:1801.00349v1, 31 Dec. 2017. 
[24] M. Nielsen, “Neural networks and deep learning,” 9 8 2017. [線上]. Available: https://github.com/hanifanm/myblogdocument/blob/master/Neural%20networks%20and%20deep%20learning.pdf.
[25] Warren S. McCulloch, Walter Pitts, "A logical calculus of the ideas immanent in nervous activity," 1943. [Online]. Available: https://link.springer.com/article/10.1007%2FBF02478259.
[26] B. Rohrer, "How do Convolutional Neural Networks work?," Data Science and Robots Blog, 18 August 2016. [Online]. Available: https://brohrer.github.io/how_convolutional_neural_networks_work.html.
[27] D. A. Hanuschkin, "Convolutional Neural Networks with Matlab, Caffe and TensorFlow," University of Freiburg, 2018. [Online]. Available: http://www.optophysiology.uni-freiburg.de/Research/research_DL/CNNsWithMatlabAndCaffe.
[28] B. Rohrer, "Deep Learning Demystified," Data Science and Robots Blog, 22 May 2016. [Online]. Available: https://brohrer.github.io/deep_learning_demystified.html.
[29] ZHIJIAN YIN1, BOYANG WAN1, FEINIU YUAN2, XUE XIA2, AND JINTING SHI , "A Deep Normalization and Convolutional Neural Network for Image Smoke Detection," Digital Object Identifier 10.1109/ IEEE ACCESS. 2017.2747399, 2017. 
[30] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: A simple way to prevent neural networks,” J. Mach. Learn. Res. vol. 15, no. 1, pp. 1929-1958, 2014. 
[31] P. Viola, M. Jones, "Robust Real Time Object Detection," in SECOND INTERNATIONAL WORKSHOP ON STATISTICAL AND COMPUTATIONAL THEORIES OF VISION – MODELING, LEARNING, COMPUTING, AND SAMPLING, VANCOUVER, CANADA, , JULY 13, 2001.. 
[32] F. Crow, “Summed-area tables for texture mapping,” In Proceedings of SIGGRAPH, volume 18(3), , pp. 207–212,, 1984. 
[33] Y. Freund, R. E. Schapire, “A Decision-Theoretic Generalization of On-Line Learningand an Application to Boosting,” Journal of Computer and System Sciences, pp. 119-139, 1997. 
[34] J. Quinlan, “Induction of decision trees,” Machine Learning 1, p. 81–106, 1986. 
[35] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in Proc. Int. Conf. Learn., 2015. 
[36] Amarjot Singh, Devendra Patil,G Meghana Reddy, SN Omkar, "Disguised Face Identification (DFI) with Facial KeyPoints using," IEEE International Conference on Computer Vision Workshops, pp. 1648-1655, 2017.