商標仿冒、產品盜版等問題，對於公司來說，一直都是令人頭痛的問題，過去都曾有許多人絞盡腦汁，想辦法不讓自家的商品被別人仿冒，就只為了讓自己研發的心血不被別人整碗端去。其中一個商品防偽的方式，就是利用特殊印刷的商標貼紙來讓別人難以仿冒，但這種方式通常需要用到特殊的印刷機器以避免被別人輕易仿製，而且需要特別的掃描器材，才可以偵測出是否為正版產品。但是這就使防偽的成本變得相當高昂，若只使用一般印刷機影印商標和用一般的相機掃描，很難達到前述的特殊商標所擁有的防偽效果，除此之外，這兩種商標都有著很突兀的外觀，若他人找出商標中的規則，便可輕易的大量製造仿冒的商標。所以本篇論文將會提出一種，特別設計的防偽二維條碼，能以一般印表機列印且用一般相機照相即可解碼訊息，並與一般的二維條碼不同，外觀上並沒有明顯的定位點，且能有一定程度的背景紋理干擾與條碼遮蔽。我們會透過深度學習的方式，以特別訓練的模型來定位條碼的位置，再以特別的二值化演算法來避免拍照時的光照環境影響，加上條碼歪斜變形的修正來提升解碼的成功率。我們還會提出一種訓練資料產生的流程，能夠用來產生特殊圖形碼的訓練資料。最後，我們會以多個資料集來測試我們方法的成功率及解碼速度，來證明我們的演算法是具有實用價值的。

Trademark counterfeiting, product piracy, etc. problem has always been a difficult problem for the company. There have been many people who have thought of many ways that don't let their products be counterfeited by others to protect their achievement. One of the ways to protect the products is to use specially printed trademark labels to make it difficult for others to copy. But this method usually requires a special printing machine to avoid being easily copied by others. And this method is needed special scanning equipment to detect whether it is a genuine product. But this makes the cost of anti-counterfeiting quite high. It is difficult to achieve the anti-counterfeiting effect of the aforementioned special trademarks by using only the photocopying trademarks of common printing machines and scanning with common cameras. In addition, both trademarks have a very awkward appearance. If others find the rules in the trademark, they can easily manufacture counterfeit trademarks in large quantities.
So this paper will present a specially designed anti-counterfeiting 2D barcode. It can be printed on a common printer and decoded by the image captured by a common camera. And unlike the general two-dimensional bar code, there is no obvious positioning pattern in appearance. Also, can have some distraction texture and marker covering. We will use a specially trained model to locate the position of the barcode through deep learning technique. A special binary algorithm is used to avoid the influence of the lighting environment when taking barcode pictures. And Fix the barcode skew deformation to improve the accuracy of decoding. We will also propose a process for generating training data that can be used to generate training data for special design labels. Finally, we will test the accuracy and decoding speed of our method with multiple data sets to prove that our system is of practical value.

中文摘要 i
Abstract ii
目錄 iii
圖目錄 v
表目錄 vi
1緒論 1
2相關研究 3
3系統概觀 4
4防偽條碼設計 6
4.1訊息編碼 6
4.2條碼外觀 7
5模型訓練 8
5.1條碼訓練資料生成 8
5.1.1單條碼訓練資料生成 8
5.1.2多條碼訓練資料生成 9
5.2 Mask-RCNN訓練 10
5.2.1訓練資料處理 10
5.2.2訓練參數 10
6條碼解碼 11
6.1條碼候選區域預測 11
6.2圖片二值化 11
6.3歪斜修正 12
6.3.1區域正規化 12
6.3.2條碼邊線搜尋 12
6.3.3區域歪斜修正 14
6.4資料解碼 15
7實驗與測試 16
7.1測試環境 16
7.2測試資料集 17
7.3 Mask-RCNN效能測試 19
7.4解碼系統測試 20
8結論 22
Bibliography 23

[1]Steffen Wachenfeld, Sebastian Terlunen, and Xiaoyi Jiang.Robust 1-D Barcode Recogni-tion on Camera Phones and Mobile Product Information Display, pages 53–69. SpringerBerlin Heidelberg, Berlin, Heidelberg, 2010. ISBN 978-3-642-12349-8. doi: 10.1007/978-3-642-12349-8_4. URLhttps://doi.org/10.1007/978-3-642-12349-8_4.
[2]Zhi Liu, Lingna He, and Ronghua Liang. Research on segmentation algorithm of 2 d colorbarcode based on mobile phone. 2012.
[3]E. Ohbuchi, H. Hanaizumi, and L. A. Hock. Barcode readers using the camera devicein mobile phones. In2004 International Conference on Cyberworlds, pages 260–265, Nov2004. doi: 10.1109/CW.2004.23.
[4]C. Creusot and A. Munawar. Real-time barcode detection in the wild. In2015 IEEEWinter Conference on Applications of Computer Vision, pages 239–245, Jan 2015. doi:10.1109/WACV.2015.39.
[5]J. Li, W. Wang, T. Rao, W. Zhu, and C. Liu. Morphological segmentation of 2-d barcodegray scale image. In2016 International Conference on Information System and ArtificialIntelligence (ISAI), pages 62–68, June 2016. doi: 10.1109/ISAI.2016.0022.
[6]Andrey Zharkov and Ivan Zagaynov. Universal barcode detector via semantic segmentation.ArXiv, abs/1906.06281, 2019.
[7]Jianjun Li, Qiang Zhao, Xu Tan, Zhenxing Luo, and Zhuo Tang. Using deep convnet forrobust 1d barcode detection. In Fatos Xhafa, Srikanta Patnaik, and Albert Y. Zomaya,editors,Advances in Intelligent Systems and Interactive Applications, pages 261–267, Cham,2018. Springer International Publishing. ISBN 978-3-319-69096-4.
[8]Qijie Zhao, Feng Ni, Yongduan Song, Yongtao Wang, and Zhi Tang. Deep dual pyramidnetwork for barcode segmentation using barcode-30k database.ArXiv, abs/1807.11886,2018.
[9]Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross B. Girshick. Mask r-cnn.2017 IEEEInternational Conference on Computer Vision (ICCV), pages 2980–2988, 2017.
[10]Alexis Hocquenghem. Codes correcteurs d＇erreurs.Chiffres, 2(2):147–56, 1959.
[11]R.C. Bose and D.K. Ray-Chaudhuri. On a class of error correcting binary group codes.Information and Control, 3(1):68 – 79, 1960. ISSN 0019-9958. doi: https://doi.org/10.1016/S0019-9958(60)90287-4. URLhttp://www.sciencedirect.com/science/article/pii/S0019995860902874.
[12]Francisco Massa and Ross Girshick. maskrcnn-benchmark: Fast, modular reference im-plementation of Instance Segmentation and Object Detection algorithms in PyTorch.https://github.com/facebookresearch/maskrcnn-benchmark, 2018. Accessed: [Insertdate here].
[13]ZXing. ZXing Multi-format 1D/2D barcode image processing library, 2012. URLhttp://code.google.com/p/zxing/.