指紋已經廣泛的應用在身份辨識上。然而，從犯罪現場所取得指紋資料有時遇到兩枚以上的重疊指紋，但是重疊指紋分離成兩枚清晰的單枚指紋是目前研究的課題。在本文中，提出了一種新的演算法，將指紋先分類為單枚與兩枚重疊指紋，再分離為單枚指紋。首先，利用HOG進行特徵萃取，將特徵排列成矩陣後，進行PSD運算訓練模板，將指紋分類。再利用橢圓遮罩計算單枚指紋的覆蓋面積，定義單枚指紋適當的長短軸的初始遮罩。接著，使用初始遮罩掃描重疊指紋計算覆蓋面積，計算所有遮罩之覆蓋面積。再接著，同時取兩個單枚橢圓遮罩與原圖計算上述遮罩方向與位置覆蓋面積，找出重疊面積最大者，即可從兩枚重疊指紋分離成兩枚部分清晰與部分不清晰的指紋。

Fingerprint has been used as a tool in human identification widely. But, fingerprint latent obtained occasionally from the crime scene is overlapping by two or more fingerprints. Therefore, how to separate overlapped latent into clear single fingerprint is a hot research issue. In this paper, we propose a new algorithm to separate overlapping fingerprints into single fingerprints. First use HOG to extract the feature and arrange in matrix. Perform a PSD training train template to classify the fingerprints. Next, to find the size of initial mask for fitted single fingerprint, we use elliptical mask to define the size of semi-major and semi-minor axes via statistical analysis The coverage area is calculated by using two single oval masks and the original figure. Finally, find the largest overlapping area. We can split two overlapping fingerprints into two parts clear and partially unclear fingerprints.

目錄
中文摘要 I
英文摘要 II
目錄 III
圖目錄 VI
表目錄 IX
第一章 緒論 1
1.1. 研究背景 1
1.2. 研究動機 4
1.3. 論文架構 4
第二章 指紋背景與相關研究 5
2.1. 指紋概論 5
2.2. 指紋影像前處理相關技術 12
2.2.1. 大津(Otsu)最佳二值化閥值演算法 [9] 12
2.2.2. 影像形態學(Morphology) 13
2.2.3. 方向梯度直方圖(Histogram of oriented gradient，HOG) [6] 14
2.2.4. 共變異數矩陣 [10] 19
2.2.5. 功率譜密度(Power Spectral Density, PSD) [11] 19
2.2.6. 賈柏濾波器(Gabor filter) [9] 20
2.2.7. 橢圓 20
第三章 最佳化自動重疊指紋分離 21
3.1. 前言 21
3.2. 系統流程 21
3.2.1. 單枚或兩枚重疊指紋分辨系統 23
3.2.2. 橢圓自動重疊指紋標記分離系統 27
第四章 實驗結果 36
4.1. 實驗環境 36
4.2. 指紋資料庫 36
4.3. 實驗結果 36
4.3.1. 單枚指紋與兩枚重疊指紋分辨結果 40
4.3.2. 重疊指紋辨識率 42
4.3.3. 各角度重疊指紋辨識率 43
第五章 結論與未來展望 46
5.1. 結論 46
5.2. 未來展望 46
參考文獻 47

圖目錄
圖 1.1. 淺指紋示意圖 2
圖 1.2. 被物件遮蔽的指紋圖像示意圖 3
圖 2.1. 指紋山脊和山谷示意圖 5
圖 2.2. 核心點示意圖 6
圖 2.3. 三角點示意圖 6
圖 2.4. 帳型紋示意圖 7
圖 2.5. 平弧紋示意圖 7
圖 2.6. 右箕紋示意圖 8
圖 2.7. 左箕紋示意圖 8
圖 2.8. 平螺紋示意圖 9
圖 2.9. 囊型紋(central packet whorl)示意圖 9
圖 2.11. 雙箕螺紋(double loop or twin-loop whorl)示意圖 10
圖 2.12. 雜型類(accidental)示意圖 10
圖 2.13. HOG流程圖 16
圖 2.14. HOG的cell與block示意圖 18
圖 3.1. 模板訓練流程圖 23
圖 3.2. 分辨指紋為單枚或兩枚重疊指紋流程圖 24
圖 3.3. HOG特徵矩陣排列示意圖 25
圖 3.4. 橢圓自動重疊指紋標記分離系統流程圖 27
圖 3.5. 原始指紋影像 28
圖 3.6. 運算後增加對比度的影像 28
圖 3.7. 經Otsu選定最佳化二值化閥值處理的二值化影像 29
圖 3.8. 前處理完成後的指紋區域示意圖 30
圖 3.9. 橢圓遮罩組示意圖 30
圖 3.10. 兩枚橢圓所形成遮罩示意圖 32
圖 3.11. 重疊的橢圓 33
圖 3.12. 左單枚橢圓 33
圖 3.13. 右單枚橢圓 34
圖 3.14. 由橢圓標記出的重疊區域 34
圖 3.15. 由左橢圓分離出的單枚指紋+部分重疊區域 35
圖 3.16. 由右橢圓分離出的單枚指紋+部分重疊區域 35
圖 4.1. 自然重疊指紋示意圖 38
圖 4.2. 合成重疊指紋示意圖 39

表目錄
表 2.1. 指紋細節點類型 11
表 4.1. 單枚與雙枚指紋之分辨正確率 41
表 4.2. 統整兩次測試之辨識率 41
表 4.3. 自動分離指紋之辨識率 42
表 4.4. 人工合成指紋辨識率 42
表 4.5. 共1600枚指紋辨識率 43
表 4.6文獻 [18]與本篇論文的比較 44

參考文獻
[1] M. Alkhathami, "Fingerprint Authentication Enhancement Using Combination of DWT and DCT," in IEEE 10th Conference on Industrial Electronics and Applications (ICIEA), 2015.
[2] A. Vashistha and A. M Joshii, "Fingerprint based Biometric Watermarking Architecture using Integer DCT," in TENCON 2016 - 2016 IEEE Region 10 Conference, 2016.
[3] A. Alsubari and R. J. Ramteke, "Extraction of Palmprint Texture Features using Combined DWT-DCT and Local Binary Pattern," in International Conference on Next Generation Computing Technologies (NGCT-2016), 2016.
[4] K. Tewari and R. L. Kalakoti, "Fingerprint Recognition and Feature Extraction Using Transform Domain Techniques," in International Conference on Advances in Communication and Computing Technologies, 2014.
[5] M. P. Dale and M. A. Joshi, "Fingerprint Matching Using Transform Features," in TENCON 2008 - 2008 IEEE Region 10 Conference, 2008.
[6] A. Mori and H. Takano, "Rotation Immunity Characteristics of Non-contact Fingerprint Recognition Using Local Luminance Variation," in SICE Annual Conference 2012, 2012.
[7] I. Bhardwaj and N. D. Londhe, "Feature Selection for Novel Fingerprint Dynamics Biometric Technique based on PCA," in International Conference on Advances in Computing, Communications and Informatics (ICACCI), 2016.
[8] Y. C. Wang, Biometrics-A Study of Speaker and Fingerprint Identification, Department of Electrical Engineering, Tamkang University, PhD. thesis., 2002.
[9] M. Sezgin and B. Sankur, "Survey over image thresholding techniques and quantitative performance evaluation," Journal of Electronic Imaging, vol. 13, no. 1, p. 146–165, 2004.
[10] I. Jolliffe, Principal Component Analysis, Second Edition, 2002.
[11] Yili Li and K. M. Wong, "Riemannian Distances for Signal Classification by Power Spectral Density," IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, vol. 4, Aug 2013.
[12] C.F. Chien and T.T. Lin, "An Image Processing Method To Measure Overlapped Leaf Area Using Elliptical Hough Transform," Computer And Computing Technologies In Agriculture, Volume II, pp. 1427-1433, 2007.
[13] Z. Y. X. ,. X. a. T. J. Z Ning, "Adaptive Orientation Model Fitting for Latent Overlapped Fingerprints Separation," IEEE Trans. Inf. Forensics Security, vol. 9, no. 10, Oct 2014.
[14] F. Chen, J. Feng and J. Zhou, "On Separating Overlapped Fingerprints," in IEEE, 2010.
[15] J.Moraleda, "Large scalability in document image matching using text retrieval," Pattern Recognition Letters, vol. 33, no. 7, pp. 863-871, 1 May 2012.
[16] J. Feng, Y. Shi, and J. Zhou, "Robust and efficient algorithms forseparating latent overlapped fingerprints," IEEE Trans. Inf. ForensicsSecurity, vol. 7, no. 5, pp. 1498-1510, Oct. 2012.
[17] "FVC2002: The Second International Fingerprint Verification Competition," [Online]. Available: http://bias.csr.unibo.it/fvc2002/.
[18] J.Feng, "Robust and Efficient Algorithms for Separating Latent Overlapped Fingerprints," IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, vol. 7, no. 5, 2012.