筆跡辨識是生物特徵辨識技術中的一種，在鑑識科學方面經常使用這項技術。但與生理特徵相比，行為特徵容易受到內在或外在因素所影響，若未能在自然情況下取得書寫者的筆跡，則鑑識的可信度必定會有所降低，並且隨著時代的改變及科技的進步，現代人使用傳統紙筆的機會大幅減少，也提升了筆跡的取得難度，這種情況下則必須思考如何以現有資訊達到應有的效果。
在本篇論文中，我們提出了一種基於交點特徵的文本無關筆跡偽造偵測技術。在輸入蒐集到的筆跡之後，我們經由二質化及型態學處理等方式進行前處理以獲得單純的筆跡影像，接著使用細線化的技術將各筆劃的骨架提取出來，並藉由骨架的影像找出具有代表性的點進行特徵擷取。之後將各影像的特徵資料以窮舉法的方式進行排列組合並以各組合的特徵建立高斯混合模型，再對組合以外的成員進行與該組合的相似度評估。並建立一個表決系統對相似度較高的組合成員進行次數統計以找出在輸入資料中相似度高的族群。接著以此族群建立一個新的高斯混合模型並對所有輸入筆跡進行相似度評估。最後計算出所有相似度資料的標準差，所有相似度落於指定閥值以下的筆跡將會被判斷為偽造。在原作者筆跡佔相對多數的前提之下，此非監督式學習方法可以用來偵測文件中的筆跡偽造。
實驗的部分我們採用IAM Handwriting Database，該資料集蒐集了657位書寫者的筆跡，圖片的解析度固定為300dpi，並且儲存為256灰階的PNG影像。在特定比例的情況下本方法最高可達到95%的準確度，若以偽造比例平均分布於20%到60%的方式進行評估，本方法亦可達到80.52%的準確度。

Handwriting recognition is one of the biometrics method, this technique is often used in forensic science. But compared with the physical characteristics, behavioral characteristics easily affected by internal or external factors. If it is unable to get writer's handwriting in natural circumstances. the forensic reliability will be reduced. Along with the change in times and technological flourishing, people have less chance using pens and pencils. That also enhance the difficulty to obtain their handwriting. In this case, you must consider how to achieve the result we need by existing information.
In this thesis, we proposed a text independent handwriting forgery detection techniques that is based on branchlet feature. The input collected handwriting image will be preprocessed that includes binarization and morphologic to get the clear handwriting image. After that we shall use thinning algorithm to get the skeleton of every stroke, and find the essential key point by the skeleton image for feature extraction. Then permutations and combinations these feature data as a group by brute force method and create their own Gaussian Mixture Model(GMM). Then estimate the similarity between this group and other input data. In the end, we will calculate the standard deviation of similarity value. All the handwriting that similarity below assigned threshold will be predicted as forgery. In the cases that promise the original author handwriting is relative majority. This unsupervised learning method is able to detect the forgery handwriting in documents.
Of the experiment we used IAM Handwriting Database, the data set collected 657 writer's handwriting, image resolution is fixed at 300dpi, and stored as 256 gray scale PNG images. Under certain proportion of situation, this method up to 95% accuracy, in terms of the way each case equally distributed from 20% to 60% forgery evaluate this method can still achieve 80.52% accuracy.

[1] S. Pal, U. Pal and M. Blumenstein, "Signature-Based Biometric Authentication," Studies in Computational Intelligence, vol.555, pp. 285-314, 2014.
[2] M. Ammar, Y. Yoshida and T. Fukumura, "Structural description and classification of signature images," Pattern Recognition, vol. 23, no. 7, pp. 697-710, 1990.
[3] R. Bajaj and S. Chaudhury, "Signature verification using multiple neural classifiers," Pattern Recognition, vol. 30, no. 1, pp. 1-7, 1997.
[4] K. Huang and H. Yan, "Off-line signature verification based on geometric feature extraction and neural network classification," Pattern Recognition, vol. 30, no. 1, pp. 9-17, 1997.
[5] 吉田 公一, 筆跡・印章鑑定ソ実務―рユ⑦Ь解説, 東京法令出版, 長野市, 日本, 2004.
[6] Z. He, X. You and Y. Tang, "Writer identification of Chinese handwriting documents using hidden Markov tree model," Pattern Recognition, vol. 41, no. 4, pp. 1295-1307, 2008.
[7] N. Akhter, "Script independent offline writer identification using handwriting style," 3rd International Conference on Imaging for Crime Detection and Prevention, London, England, pp. 1-5, 2009.
[8] S. He, M. Wiering and L. Schomaker, "Junction detection in handwritten documents and its application to writer identification," Pattern Recognition, vol. 48, no. 12, pp. 4036-4048, 2015.
 
[9] N. Otsu, "A threshold selection method from gray-level histograms," IEEE Transactions on Systems, Man, and Cybernetics, vol. 9, no. 1, pp. 62-66, 1979.
[10] T. Zhang and C. Suen, "A fast parallel algorithm for thinning digital patterns," Communications of the ACM, vol. 27, no. 3, pp. 236-239, 1984.
[11] U. Marti and H. Bunke, "The IAM-database: an English sentence database for offline handwriting recognition," International Journal on Document Analysis and Recognition, vol. 5, no. 1, pp. 39-46, 2002.
[12] D. Reynolds and R. Rose, "Robust text-independent speaker identification using Gaussian mixture speaker models," in Proceedings of the IEEE Transactions on Speech and Audio Processing, vol. 3, no. 1, pp. 72-83, 1995.
[13] F. Borba, R. Honorato and A. Juan, "Use of Raman spectroscopy and chemometrics to distinguish blue ballpoint pen inks", Forensic Science International, vol. 249, pp. 73-82, 2015.
[14] Z. Luo, F. Shafait and A. Mian, "Localized forgery detection in hyperspectral document images", 13th International Conference on Document Analysis and Recognition, Nancy, France, pp.496-500, 2015.
[15] M. Nunkoo, M. Saib-Sunassy, H. Wah and S. Laulloo, "Forensic Analysis of Black, Blue, Red, and Green Ballpoint Pen Inks", Crystallizing Ideas – The Role of Chemistry, pp. 323-339, 2016.
[16] A. Schlapbach and H. Bunke, "Off-line writer identification and verification using gaussian mixture models," Machine Learning in Document Analysis and Recognition, val. 90, pp. 409-428, 2008.
[17] A. Gordo, A. Fornés, E. Valveny and J. Lladós, "A bag of notes approach to writer identification in old handwritten musical scores," in Proceedings of the 8th IAPR International Workshop on Document Analysis Systems, Boston, MA, pp.247-257, 2010.