鈔券或郵票等安全文件第一階的防偽功能是讓視覺、觸覺感官可清楚辨別防偽因子。第二階防偽是採用簡易放大鏡或手機等設備觀察線條、網點及微小字，或以紫外線光源觀察螢光墨之反應。第三階防偽是由專責單位採用特殊儀器進行分析。因此會置入不同功能的防偽訊息於安全文件內，最常見的防偽因子有光變油墨(Optically Variable Ink, OVI)、光影變化箔膜(Optically Variable Device, OVD)、隱性螢光墨及網點結構，其中光變油墨及箔膜的色彩影像隨光源與視角而產生變化。本研究將採用曲面形軌跡的移動光源及非接觸式量測獲得動態影像及色彩資訊，試圖以移動光源及視覺檢測於一次拍攝後，能否實現第一、第二及第三階防偽鑑識可行性之研究。
本研究架構主要針對鈔券及彩色郵票上的重要防偽因子逐一探討，分為六大部份，第一部份為光變油墨(OVI)色彩變化及介電質反光現象的探討。第二部份以CMF(Color, Material and Finish)特性及採用卷積神經網路判斷不同廠牌的光變油墨。第三部份為曲面形軌跡之移動光源對OVD影像和色彩進行量測及辨識。第四部份為預測隱性螢光墨色彩。第五部份為二種不同網點結構組合對色彩平衡之探討。第六部份為以傅立葉轉換及深度學習法鑑識網點的特性。
上述結果的摘要為：第一部份光變油墨色彩變化及介電質反光現象的探討，可於CIELAB色空間繪製出油墨色彩軌跡圖，並觀察得知採用商業網版製程經紫外光固化後產生介電質反光現象。第二部份為三種不同距離觀察條件下取像及人工智慧判斷二種廠牌的光變油墨，以平台掃描器的近接觀察可達到99.96%的辨識率，證明在光源穩定情況下採用正常距離非接觸的觀察方式辨識真偽是可行的。第三部份為移動光源對光影變化箔膜影像及色彩進行量測及辨識，採用卷積神經網路辨識16種箔膜得到高達100%之辨識率。第四部份為預測隱性螢光墨色彩，採用了6種不同的比較方法，包含了傳統迴歸及人工智慧，其中將RGB訊號以影像方式進行卷積神經網路可得到平均色差值接近1△E*ab的精確度。第五部份為網點結構組合對色彩平衡之探討，共提出影像分區配置不同網點及二階段過網等二種創新的網點組合方法，可達到防偽功能及適合量產作業流程。第六部份為採用傅立葉轉換及深度學習方法辨識網點的特性，傅立葉轉換可快速的辨識調幅或調頻網點，而深度學習除了可辨識調幅和調頻網點外，也可直接由訓練模型預測網點大小、網點形狀、網屏線數及網屏角度。
非接觸式防偽特徵量測可使用在生產線上，確保安全文件的防偽品質，也可使用在自動化的批量檢測儀器上，提高偽製鈔券檢測的效率。未來希望整合本研究的多種非接觸量測技術，設計自動化檢測儀，取代現行多種儀器並用的耗時檢測方式，並達到一次取像可進行多階的辨識能力。

The first-level anti-counterfeiting function of security documents such as banknotes or stamps allows the visual and tactile senses to clearly identify the anti-counterfeiting factors. The second-level of anti-counterfeiting is to observe invisible dots, lines, patterns, or micro-text printed with special methods or fluorescent inks through simple devices such as magnifying glasses, mobile phones or ultraviolet lighting. The third-level anti-counterfeiting is analyzed by a special unit using special equipments. In order to perform the three-level security check, various functions of anti-counterfeiting are placed in the security documents. The most commonly used anti-counterfeiting factors are optically variable ink (OVI), optically variable device (OVD), covert fluorescent ink and the structure of halftone dots. The OVI and OVD change color and image according to the light source and the viewing geometry. In this study, a moving light source and non-contact imaging devices are used to detect the change of color and image information in order to evaluate the feasibility of the first, second and third-level anti-counterfeiting forensic identification with machine visual inspection.
This research focuses on the important anti-counterfeiting factors of banknotes and color stamps. It is divided into six parts. The first part is a study of OVI color changes based on dielectric reflection phenomena. The second part is to judge the authenticity of OVI ink based on the characteristics of CMF (Color, Material and Finish) using a convolutional neural network (CNN). The third part is to measure and identify the colors and images of OVD samples by moving the light source. The fourth part is a study of color prediction of covert fluorescent ink. The fifth part is a study on color balance of images with multiple dot-structures. The sixth part is to use Fourier spectrum and deep learning methods to identify the characteristics of printing dots.
The results are summarized as follows: In the first part, the ink color trajectory can be analyzed in CIELAB color space, and the dielectric reflections formed by the commercial screen printing after UV curing can be observed. The second part is to acquire images through different viewing distances and use CNNs to judge the authenticity of OVI inks. Among three different viewing distances, close-up viewing using a flatbed scanner achieved the highest recognition rate (99.96%). The third part is to classify OVD by a moving light source. The results show that using a CNN model to identify 16 kinds of OVD can achieve a 100% recognition rate. The fourth part is the color prediction of covert fluorescent ink. It used 6 different comparison methods, including traditional regression and modern deep learning models. The CNN can get the accuracy of the average color difference close to 1△E*ab. The fifth part is about the color balance of hybrid halftoning. Two innovative methods of dot combination are proposed, each of which has a different application field. The sixth part is the use of Fourier spectrum and deep learning methods to identify the characteristics of the dots. Fourier spectrum can quickly identify AM and FM dots. In addition to identifying the screen type, deep learning can also distinguish the dot size, dot shape, screen density and screen angle. The non-contact anti-counterfeiting feature measurement can be used on the production line to ensure the anti-counterfeiting quality of the security documents, and it can also be used on the automated batch detection instrument to improve the efficiency of the detection of counterfeit banknotes. In the future, we hope to combine the various non-contact measurement technologies proposed in this study to design an automated detection instrument to replace the current time-consuming detection method of multiple instruments.

[1] Anton Bleikolm, Olivier Rozumek, Edgar Muller, Optically Variable Pigments Providing a Color Shift between Two Distinct Colors, Coating Composition Comprising the Same, Method for Producing the Same and Substrate Coated with the Coating Composition, US Patent, #US 6521036 B1, 2003.
[2] Brian Holmes, Holographic Security Device having Diffractive Image Generating Structures, US Patent, #US 8,625,181 B2, 2014.
[3] 羅梅君，數位色彩管理科學-色彩度量學，藍海文化, 2011.
[4] CIE015:2018, Colorimetry, 4th Ed., Technical Report, Vienna, Austria. CIE, 2018
[5] Radiantzemax,https://www.radiantzemax.com/measurement-systems/imaging-sphere/, 2014.
[6] Stephen R. Marschner, Stephen H. Westin, Eric P. F. Lafortune, and Kenneth E. Torrance, Image-Based Bidirectional Reflectance Distribution Function Measurement, Applied Optics, 39(16): 2592-2600, 2000.
[7] Jeffrey Warda, The AIC Guide to Digital Photography and Conservation Documentation, 2nd edition, American Institute for Conservation of Historic and Artistic Works, American Institute for Conservation(AIC), 125. 2011.
[8] Frank P. Nanna, John Jereb, Gloss Measurement System, US Patent, #US 5,552,890.1995.
[9] Richard Szeliski, Computer Vision – Algorithms and Applications, Springer, 516-517, 2011.
[10] 湯一平，基於單目多視角機器視覺的珍珠顏色光澤度線上自動分級裝置，CN102967586A, 2013.
[11] Wei Ji, Georgios Koutsidis, Ronnier Luo, John Hutchings, Mahmood Akhtar, Francisco Megias, Mick Butterworth, A Digital Imaging Method for Measuring Banana Ripeness, Color Research and Application, 38(5): 364-374. 2013.
[12] Guowei Hong, M. Ronnier Luo, Peter A. Rhodes, A Study of Digital Camera Colorimetric Characterization based on Polynomial Modeling, Color Research and Application, 26(1):76-84, 2001.
[13] Shoji Tominaga, Daisuke Nishioka, Takahiko Horiuchi, An Integrated Spectral Imaging System for Producing Color Images of Static and Moving Objects, Color Research and Application, 40(4), 2015.
[14] R. M. Haralick, K. Shanmugam and I. Dinsten, Texture Feature for Image Classification, IEEE Transactions on System, Man and Cybernetics, 3(6):610-621, 1973.
[15] M. M. Galloway, Texture Classification using Gray Level Length, Computer Graphics and Image Processing, 4:172-179,1975.
[16] Jan Morovic, Julian Shaw, Pei-Li Sun, A Fast, Non-Iterative and Exact Histogram Matching Algorithm, Pattern Recognition Letters,23(1):127-135, 2002.
[17] T. Randen, J. H. Husoy, Filtering for Texture Classification: A Comparative Study, IEEE Transactions on Pattern Analysis and Machine Intelligence, 21(4): 291-310, 1999.
[18] 孫沛立、廖子騰、楊智、闕家彬、劉美廷，智慧圖紋生成技術之探討，印刷科技，第三十六卷第二期，Jun/2020.
[19] R. W. Conners, C. A. Harlow, Toward a Structural Textureal Analyzer based on Statistical Methods, Computer Graphics and Image Processing, 12:224-256, 1980.
[20] G. R. Cross, A. K. Jain, Markov Random Field Texture Model, IEEE Transactions on Pattern Analysis and Machine Intelligence, 5(1): 25-39,1983.
[21] A P. Pentland, Fractal Based Description of Natural Scenes, IEEE Transactions on Pattern Analysis and Machine Intelligence,6(6): 661-6741,1984.
[22] NARLabs 國家實驗研究院儀器科技研究中心：光學薄膜之原理與應用http://www.itrc.narl.org.tw/Publication/Newsletter/no67/p14.php
[23] Shoji Tominaga, Surface Identification using the Dichromatic Reflection Model. IEEE Transactions on Pattern Analysis and Machine Intelligence. 13(7):658-670, 1991.
[24] Shoji Tominaga, Norihiro Tanaka, Estimating Reflection Parameters from a Single Color Image. IEEE Computer Graphics and Applications, 20(5): 58-66, 2000.
[25] Y. T. Tsai, H. C. Wang, and Y. L. Chiu, Value-Added Applications for the Integration of Dot-Matrix Hologram’s Iridescent Effects and Movable Type Printing Technology, Proceedings of the International Conference on Applied System Innovation, ICASI 2015.
[26] Leo Kenen, Robert Jones, and Robert Durst, Optically Variable Devices with Encrypted Embedded Data for Authentication of Identification Documents. US Patent, #US 0010776A1, 2005.
[27] Dennis Gabor, Microscopy by Recorded Wavefronts. Proceedings of the Royal Society (London). 197 (1051): 454–487. doi:10.1098/rspa.1949.0075, 1949.
[28] 黃士剛，安全印刷產業之創新突破科技，印刷科技，第27卷第3期，2011.
[29] Jan Morovic, Pei-Li Sun, Predicting Image Differences in Color Reproduction from their Colorimetric Correlates, Journal of Imaging Science and Technology, 47(6): 509-516, 2003.
[30] Mahnaz Mohammadi, Developing an Imaging Bi-Spectrometer for Fluorescent Materials, Ph.D. thesis of the Chester F. Carlson Center for Imaging Science Rochester Institute of Technology, 2009.
[31] Shoji Tominaga, Keita Hirai, and Takahiko Horiuchi, Estimation of Bispectral Donaldson Matrices of Fluorescent Objects by using Two Illuminant Projections, Journal of the Optical Society of America A,32(6)1068-1078, 2015.
[32] G. Cybenko, Approximation by Superpositions of a Sigmoidal Function Mathematics of Control, Signals, and Systems, 2(4):303–314, 1989.
[33] Convolutional Neural Networks (LeNet) – Deep Learning 0.1 Documentation. Deep Learning 0.1. LISA Lab. 31 August 2013.
[34] Adhy Rizaldy and Heru Agus Santoso, Performance Improvement of Support Vector Machine (SVM) with Information Gain on Categorization of Indonesian News Documents. In Application for Technology of Information and Communication (iSemantic), 227~232, 2017.
[35] Alex Krizhevsky, Ilya Sutskever, Geoffrey E.Hinton, ImageNet Classification with Deep Convolution Neural Networks, University of Toronto. 2012.
[36] K. He, X. Zhang, S. Ren, and J. Sun, Deep Residual Learning for Image Recognition. arXiv preprint arXiv:1512.03385,2015.
[37] Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, and Alex Alemi, Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning. arXiv:1602.07261v2,2016.
[38] M. D. Fairchild, Color Appearance Models, Addison Wesley, Inc., 1998.
[39] Eric R. Fossum, Digital Camera System on a Chip, IEEE, 8-15, 1998.
[40] G.Wyszecki and W. S. Stiles, Color Science: Concepts and Methods, Quantitative Data and Formulae. Wiley, 1982.
[41] Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton, Imagenet Classification with Deep Convolutional Neural Networks. Advances in Neural Information Processing Systems. 1: 1097–1105, 2012.
[42] Benjamin Graham, Fractional Max-Pooling. 12-18, 2014.
[43] Christopher Bishop, Pattern Recognition and Machine Learning. Berlin: Springer. ISBN 0-387-31073-8. 2006.
[44] Po-Tong Wang, Jui-Jen Chou, Chiu-Wang Tseng, Colorimetric Characterization of Color Image Sensors based on Convolutional Neural Network Modeling, Sensors and Materials. 31(5):1513–1522, 2019.
[45] M. R. Samworth, Digital Halftoning Combining Multiple Screens within a Single Image, US Patent, #US 6,118,935,2000.
[46] Sasan Gooran, Hybrid Halftoning - A Useful Method for Flexography, IS&T, 49(1): 85-95, 2005.
[47] Sasan Gooran, A Novel Hybrid Amplitude Modulated/Frequency Modulated Halftoning based on Multilevel Halftoning, IS&T, 50(2):157-167, 2006.
[48] Eugenio Carvajal, A Comparison Study of Concentric Screening versus Conventional AM Screening and FM Screening in Offset Printing, Thesis for the degree of Masters Science in the School of Printing Management and Science of the Rochester Institute of Technology, 2006.
[49] Michael Kriss, Handbook of Digital Imaging, Vol.2: 930~933, 2015.
[50] Zhen He and Charles A. Bouman, AM/FM Halftoning: Digital Halftoning through Simultaneous Modulation of Dot Size and Dot Density, Journal of Electronic Imaging, 13(2):286–302, 2004.
[51] Mohammad Mehdi Faghih, Mohsen Ebrahimi Moghaddam, Neural Gray: A Color Constancy Technique using Neural Network, Color Research and, Application, 39(6):571-581, 2014.
[52] Ginter Bestmann, Altenholz (DE), Method for Gray Balance Correction of a Printing Process, US Patent, #US 8,537,420, B2.
[53] Kuang-Hua Sun. A Study of Mechanical Dot Gain for Different Dot Shapes Based on the Border Zone Theory, Thesis for the degree of Masters Science in the School of Printing Management and Science of the Rochester Institute of Technology, 1991.
[54] Michael Kriss, Handbook of Digital Imaging, 2: 927-928.2015.
[55] Daniel L. Lau, Gonzalo R. Arce, Modern Digital Halftoning, Marcel Dekker Inc., New York.2001.
[56] Imatest SQF (Subjective Quality Factor) and Acutance, http://www.i matest.com/docs/sqf.html (accessed 31 January 2014).
[57] J. A. C. Yule, Principle of Color Reproduction, John Wiley & Sons, Inc., New York, 1967.
[58] Bence Adam, Method and Apparatus for Generating and Authenticating Security Documents, US Patent, #US 9,652,814, B2.
[59] R. N. Bracewell, Fourier Spectrum and its Applications, McGraw-Hill Education, 1980.
[60] Michael R. Nofi, The Color Determination of Optically Variable Flake Pigments, Forensic Analysis on the Cutting Edge: New Methods for Trace Evidence Analysis, John Wiley & Sons, Inc., 375-397, 2007.
[61] Aline R. Novais Rodrigues, Fábio Luiz Melquiades, Carlos Roberto Appoloni, Eduardo Neris Marques, Characterization of Brazilian Banknotes using Portable X-Ray Fluorescence and Raman Spectroscopy, Forensic Science International 302, 2019.
[62] Andrew A. Green, Mark Berman, Paul Switzer, and Maurice D. Craig, A Transformation for Ordering Multispectral Data in terms of Image Quality with Implications for Noise Removal, IEEE Transactions on Geoscience and Remote Sensing, 26(1):65-74,1988.
[63] Yushi Chen, Zhouhan Lin, Xing Zhao, Gang Wang and Yanfeng Gu, Deep Learning-Based Classification of Hyperspectral Data, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,7(6):2094-2107,2014.