基於特徵點與偏旁資訊之中文字跡真偽辨識演算法

Chinese Handwriting Verification Algorithm Using Point and Side Information

王冠驊(Guan-Hua Wang)

指導教授 ： 丁建均

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902924

摘要

字筆跡鑑定的目的在於分辨是該文件是否為當事人所寫，亦或是被別人偽造。字跡辨識在法務鑑定上扮演非常重要的角色，其鑑定的內容十分的廣，例如信用卡、賬單、遺書等等。
要準確的做到筆跡鑑定是一項極需眼力跟細心程度且費時費力的工作。筆跡鑑定的關鍵在於如何從筆跡之間的差異分辨該筆跡是否為同一人所寫。然而，由於同一個人的字跡也存在個體變異，這使得字跡真偽辨識困難許多。
在這篇論文中，我們提出了一套針對中文字的字跡真偽辨識演算法。比起其他語言中的文字，中文字擁有比較複雜的結構。因此，我們所提出的演算法使用了3種不同的特徵集;全局特徵、匹配偏旁特徵以及匹配點特徵。相對於全局特徵，匹配偏旁特徵以及匹配點特徵為局部特徵。局部特徵幫助我們從較小的維度觀察中文文字。我們希望結合這3種特徵集可以更完整的描述整個文字。考量到在做真偽辨識時，不同的特徵對於不同的字有不同的重要性，我們使用k-value來篩選合適的特徵。所有篩選出來的特徵被組合成代表該字的特徵向量。支持向量機則會根據組合特徵驗證該手寫文字是否為偽造的文字。實驗結果顯示，我們的演算法在字跡真偽辨識中的準確率達到95.85％，並明顯優於其他現有的辨識方法。此外，實驗結果也顯示我們提出的演算法具有較好的可靠性。

關鍵字

字跡 ； 辨識 ； 中文字 ； 點匹配 ； 偏旁匹配 ； 偏旁 ； 偽造

並列摘要

The purpose of handwriting verification is to identify whether the handwritten words were written by a person itself or is forged by others. Handwriting verification plays a very important role in forensics. The content is very extensive, such as credit cards, bills, testaments. Accurate handwriting verification is a work that requires a lot of eyesight and care; overall it is time-consuming and laborious. The key to handwriting verification is how to identify the authenticity of the handwriting from the difference between. However, since there is also an individual variation in the handwriting of the same person, it makes handwriting verification a challenging topic.
In this thesis, we proposed a verification algorithm for the Chinese word. Comparing with words in other languages, Chinese words often have a more complex structure. Therefore, our algorithm adopted three different sets of features, the global features, the matched side feature, the matched point feature. With respect to global features, matched side feature and matched point feature are local features, which help the algorithm to observe the Chinese word in smaller magnitude levels. We hope the combination of the three feature sets is able to completely describe the entire word. Considering that different features may have different importance for the different word, we use k-value to select suitable features for each word. All the selected features are combined and the support vector machine with a linear kernel is used as a classifier to verify whether the handwritten word was forged according to the combined feature. Experimental results show that the proposed algorithm reaches 95.85% accuracy in handwriting verification and outperforms several other methods. Moreover, our proposed algorithm shows robustness while dealing with different words.

並列關鍵字

handwriting ； verification ； Chinese word ； point matching ； word side ； side matching ； forge

參考文獻

[1] N. A. Murshed, F. Bortolozzi, and R. Sabourin, "Binary image compression using identity mapping backpropagation neural network", Electronic Imaging '97. International Society for Optics and Photonics, 1997, pp. 29-35.

[2] N. A. Murshed, R. Sabourin, and F. Bortolozzi, "A cognitive approach to off-line signature verification", International Journal of Pattern Recognition and Artificial Intelligence, vol. 11, no. 05, 1997, pp. 801-825.

[3] H. Khalajzadeh, M. Mansouri, and M. Teshnehlab, "Persian Signature Verification using Convolutional Neural Networks", International Journal of Engineering Research and Technology, vol. 1, 2012.

[4] L. Hafemann, R. Sabourin, and L. Oliveira, "Writer-independent feature learning for offline signature verification using convolutional neural networks", Neural Networks The 2016 International Joint Conference on, 2016.

[5] D. Field, "Relations between the statistics of natural images and the response properties of cortical cells," Journal of the Optical Society of America A, 1987, pp. 2379-2394