本文提出一套針對偽裝正面人臉的辨識預處理系統，旨在進行更進階偽裝辨識或人工辨識前，提供最有可能的數個辨識對象，以提升人力與電腦的效率。本文於訓練階段先使用Adaptive Boosting（Adaboost）演算法先取出全圖中的人臉，並切割為上、中、下三個圖形區塊，再使用方向梯度直方圖演算法（HOG）個別擷取三個圖形特徵，再排列成特徵矩陣與形成標籤矩陣。而後將特徵矩陣與標籤矩陣作為輸入送入標籤一致性的稀疏編碼表示法（LC-KSVD）訓練或將特徵矩陣直接送入K-SVD訓練，得出三個具鑑別力的字典以用做分類。於測試階段特徵擷取與訓練相同，而測試階段還需要個別經由HSV膚色還原分析來判定是否須用於分類器。最後由分類器綜合測試影像特徵矩陣、字典以及HSV膚色還原分析結果來計算，得到辨識結果。實驗結果顯示針對偽裝正面人臉，本系統有較好之篩選效果。

In recent years, face recognition research has become more sophisticated. Identification of the current frontal face, the recognition rate is very high. But for the researches of disguised frontal face recognition system, the number of researches and recognition rates are very low. In this paper, we use the Adaptive Boosting (Adaboost) algorithm to retrieve the human face in full picture, and then use the histogram of oriented gradients algorithm (HOG) to retrieve the front face feature. Before using the label consistency sparse coding representation (LC-KSVD), using HOG features extracted in accordance with the position of the picture into three categories first, then rearranged corresponding label matrixes. The label matrixes will train with features matrixes, getting three sparse dictionaries to be used as classification. In testing step, feature extraction are same to training step. But the testing step need to use HSV to determine whether the feature matrixes and the dictionary are necessary for the classifier. Finally, classifier calculates feature matrixes, dictionaries and HSV, obtained identification results. The results show that for disguised frontal face, based on LC-KSVD classifier of disguised frontal face preliminary screening system has a high success rate of preliminary screening.

目錄
致謝	I
中文摘要	III
英文摘要	IV
目錄	V
圖目錄	VII
第一章 緒論	1
1.1 研究背景	1
1.2 研究動機與目的	1
1.3 論文架構	2
第二章 文獻探討	3
2.1 Adaboost演算法	3
2.2 Histogram of Oriented Gradient	9
2.2.1 梯度計算	10
2.2.2 直方圖統計之方向劃分	10
2.2.3 Block選擇與Block正規化	13
2.3 K-SVD	14
2.4 Label Consistent K-SVD	17
2.4.1 傳統用於分類之字典訓練	17
2.4.2 LC-KSVD訓練	19
第三章 系統架構	20
3.1 影像前處理	23
3.2 HOG特徵擷取與PCA降維	28
3.3 LC-KSVD 字典訓練	31
3.4 測試階段	32
第四章 實驗結果	35
4.1 實驗環境	35
4.2 資料庫使用	35
4.3 實驗結果	36
4.4 不同特徵與分類器之綜合比較	42
第五章 結論與未來展望	45
參考文獻	46


 
圖目錄
圖 2. 1 范聖恩 [6]之Adaboost訓練演算法流程圖解	6
圖 2. 2 四種矩形特徵示意圖	8
圖 2. 3 Paul Viola與Michael Jones [7]提出之層疊分類原理	8
圖 2. 4 Matlab的HOG角度示意圖	11
圖 2. 5 HOG單一Cell統計示意圖	12
圖 3. 1 訓練流程圖	21
圖 3. 2 測試流程圖	22
圖 3. 3 Adaboost演算法選出之候選方框	24
圖 3. 4 Adaboost無法成功框選人臉範圍示意圖	25
圖 3. 5 經過正規化為768×768 pixels後的人臉圖像	26
圖 3. 6 切割圖形為三區塊示意圖	27
圖 3. 7 Cell 與 Block 示意圖	29
圖 3. 8 HOG特徵於切割影像計算示意圖	29
圖 3. 9 特徵矩陣排列示意圖	30
圖 3. 10 特徵矩陣與標籤矩陣排列示意圖	31
圖 3. 11 HSV膚色還原效果示意圖	33
圖 4. 1 資料庫提供之切割影像	36
圖 4. 2 第一類受測資料範例	37
圖 4. 3 第二類型受測資料範例圖	38
圖 4. 4 第三類型受測資料範例圖	39
圖 4. 5 第一類型受測資料之CMC曲線	41
圖 4. 6 第二類型受測資料之CMC曲線	41
圖 4. 7 第三類型受測資料之CMC曲線	41
圖 4. 8 HOG特徵與LBP特徵之CMC曲線比較	42
圖 4. 9 K-SVD與LC-KSVD之綜合比較	43

參考文獻
[1] 	Xiaoqing Zhang, Dong Ren, “A Survey of Approaches to Feature Extraction in Face Recognition,” 於 International Conference on Computer Science and Network Technology, 2012. 
[2] 	Billy Y.L. Li, Ajmal S. Mian, “Using Kinect for Face Recognition Under Varying Poses, Expressions, Illumination and Disguise,” 於 Applications of Computer Vision (WACV), 2013. 
[3] 	T. I. Dhamecha, R. Singh, M. Vatsa, and A. Kumar, “Recognizing Disguised Faces: Human and Machine Evaluation,” 於 PLoS ONE, 2014. 
[4] 	T. I. Dhamecha, A. Nigam, R. Singh, and M. Vatsa, “Disguise Detection and Face Recognition in Visible and Thermal Spectrums,” 於 In proceedings of International Conference on Biometrics, 2013. 
[5] 	Y. Freund, R. E. Schapire, “A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting,” Journal of Computer and System Sciences, pp. 119-139, 1997. 
[6] 	范聖恩, “以外形特徵為基礎之影像語言分類器-應用於破碎中文字合併,” 於 中央大學資訊工程研究所碩士論文, 2009. 
[7] 	P. Viola, M. Jones, “Robust Real Time Object Detection,” 於 Second International Workshop on Statistical and Computational Theories of Vision Vancouver, Canada, 2001. 
[8] 	N. Dalal, B. Triggs, “Histograms of Oriented Gradients for Human Detection,” 於 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) , San Diego, CA, USA, 2005. 
[9] 	M. Aharon, M. Elad and A. Bruckstein, “K-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation,” IEEE Transactions on Signal Processing, pp. 4311-4322, 2006. 
[10] 	Zhuolin Jiang, Zhe Lin, Larry S. Davis, “Label Consistent K-SVD: Learning a Discriminative Dictionary for Recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 2651 - 2664, 2013. 
[11] 	K. Pearson, “On Lines and Planes of Closest Fit to Systems of Points in Space.,” Philosophical Magazine, p. 559–572, 1901. 
[12] 	H. F. Hashem, “Adaptive Technique for Human Face Detection Using HSV Color Space and Neural Networks,” 於 26th NATIONAL RADIO SCIENCE CONFERENCE (NRSC2009), 2009.