由於資訊科技的快速發展，藉由數位典藏的方式進行歷史文件手稿或書籍資料的保存逐受重視，但是一般印刷或是手稿用紙因所選擇之材質與厚度不同也可能會造成背面的文字或圖片翻印至正面。市面上有許多掃描設備，雖然使用方便及設計客製化，但是價格多為高昂且在輸出影像上可調性也較簡單。有鑑於此，本研究提出一個以低成本設備所設計之文件影像系統，本研究使用樹梅派3B+(Raspberry 3B+)作為硬體，搭配樹莓派相機(PI NOIR CAMERA V2)，能在控制端輸入指令後自動拍照並執行演算法，以移除原稿上的雜訊及汙染並且修正背面資訊翻印至正面的問題。在演算法的部分，本研究使用自適應方向性提升式離散小波轉換(Adaptive Directional Lifting-based Discrete Wavelet Transform, ADLDWT)將影像資訊轉換成頻率域分析，分別對高頻與低頻做處理。在低頻的部分，對低低頻帶(LL subband)做區域二值化處理將文字與背景分割，其中使用積分影像(Integral Image)的概念加快運算速度。在高頻的部分，對低高頻帶(LH subband)、高低頻帶(HL subband)和高高頻帶(HH subband)採用改良式最小均方演算法(Least Mean Square, LMS)訓練流程，對三個子頻帶資訊各別進行獨特的權重遮罩訓練，且以各別子頻帶影像細部所產生不同變異數值(Variance)進行分類，再利用訓練出的權重遮罩對各別的頻帶進行卷積(Convolution)運算，讓高頻影像能保留邊緣資訊並將背景細部雜訊濾除。最後，對四個子頻帶做小波反轉換，將影像重建成原尺寸。
    實驗結果使用Document Image Binarization Competition (DIBCO)的公用樣本做測試，也使用此比賽所提供的數據測量工具與其他的方法做比較。除了做數據比較，本研究也計算出各種演算法在電腦以及樹梅派上的運算時間，結果顯示本研究所提出的演算法雖然在數據上並非全部都是最佳，但是在樹莓派上執行演算法與電腦上執行的時間差不會過大，影像結果也沒有失真。由實驗得知本研究所提出的低成本文件影像系統之樹梅派嵌入式平台具有即時的特性並且可以得到和電腦上一樣的結果。

Due to the rapid development of information technology, the preservation of hand-written historical documents and scripts archived by digitized images has been gradu-ally emphasized in recent years. However, the selection of different thickness of paper for printing or handwriting probably causes the information of back page seeping into front page. Digital image of a page obtained by sampling could be interfered by image information on the reverse side. There are many scanners on the market. Although the designs are easy to use and customized, but the price is too high and the scalability of output results is monotonous. In view of these problems, this work proposes a low cost document image system, and takes Raspberry Pi 3B+ with PI NOIR CAMERA V2 as hardware for the system. We can input commands on a PC to remote control the Rasp-berry Pi to take a picture and execute the algorithm. In the algorithm for this system, we use Adaptive Directional Lifting-based Discrete Wavelet Transform(ADLDWT) to transform image data from spatial domain to frequency domain, and perform on high frequency subbands and low frequency subbands respectively. For low frequency band, we perform local threshold to segment text from background, and use the conception of Integral Image to boost computing speed. For high frequency band, we use modified Least Mean Square (LMS) training algorithm to perform a unique weighted mask train-ing on three subbands respectively, and the individual subband image is classified into different category according to its variance. After training, we use the trained weighted masks to perform the convolution with each original subband image to preserve edge information and filter out detailed background noise. Finally, inverse discrete wavelet transform is performed to reconstruct   the four subband images to a result image with original size.
In this work, we take the open data from Document Image Binarization Competition (DIBCO) to test results, and also compare evaluation measurements with other methods which submitted to the competition. In addition to comparing evaluation measure-ments, we also calculate the operation time of many methods on PC and Raspberry Pi platform. The experimental results show that the method we propose in this work is not completely the best in all measurements though, but the difference of operation time between PC and Raspberry Pi is little, and the result images are not distorted. Accord-ing to experimental results, the proposed low cost document image system in this work which performed on Raspberry Pi embedded platform has real time property and ob-tains the same results as those performed on PC.

目錄
中文摘要............................................................................................................Ⅰ
英文摘要.........................................................................................................Ⅲ
目錄..................................................................................................................Ⅴ
圖目錄............................................................................................................Ⅶ
表目錄..........................................................................................................XIII
第一章	緒論.....................................................................................................1
1.1 研究動機…...………..........................................................................1
1.2 論文架構…...………..........................................................................6
第二章	文獻回顧…...………...............................................................................7
第三章	系統介紹.........................................................................9
第四章	影像前處理…...……….........................................................................18
4.1 設定感興趣區域 (Region of Interest, ROI)...........................................19
4.2 離散小波轉換回顧.........................................................................21
4.2.1 二維離散小波轉換.................................................................22
4.2.2 提升式離散小波轉換.............................................................23
4.2.3 方向性提升式離散小波轉換.................................................27
4.2.4 方向性提升式離散小波重建.................................................31
第五章	影像分割...........................................................................................33
5.1文字與背景分割….............................................................................33
5.1.1 積分影像….............................................................................35
5.1.2 閥值選取….............................................................................36
第六章	最小均方演算法 (Least Mean Square, LMS)........................................38
	6.1 自適應濾波器...................................................................................38
	6.2 原始最小均方演算法.......................................................................39
6.3 改良式最小均方演算法...................................................................41
第七章	實驗結果與比較…...........................................................................45
	7.1 公用樣本實驗結果….......................................................................46
	7.2 公用樣本品質數據比較...................................................................97
7.3 公用樣本硬體執行時間比較.........................................................104
7.4 樹莓派實際操作結果.....................................................................108
第八章	結論.................................................................................................116
參考文獻.......................................................................................................118

圖目錄
圖1.1 (a)DIBCO 2009 P04樣本[1] (b)DIBCO 2011 PR1樣本[3]........................2
圖1.2 (a)H-DIBCO 2012 H05樣本[4] (b)DIBCO 2011 PR7樣本[3]....................3
圖1.3 (a)DIBCO 2011 PR3樣本[3] (b)DIBCO 2013 PR2樣本[5]........................4
圖3.1 樹莓派3B+(Raspberry 3B+)[9]................................................................9
圖3.2 派相機(PI NOIR CAMERA V2)[10].........................................................10
圖3.3 樹莓派LEGO外盒…….......................................................................10
圖3.4 系統固定架..........................................................................................11
圖3.5 完整系統..............................................................................................12
圖3.6 DocsExpress DS1400AF A3文件拍攝機[32]...........................................14
圖3.7 Shoot Holic 攝手KC5M01拍攝式文件掃描器[33]...............................15
圖3.8 EPSON Perfection V39 輕薄照片/書本掃描器[34]................................15
圖3.9 Handy scan手持掃描筆[35]....................................................................16
圖4.1 系統流程圖..........................................................................................19
圖4.2 輸入影像..............................................................................................20
圖4.3 感興趣區域影像..................................................................................21
圖4.4 二維離散小波分解示意圖..................................................................22
圖4.5 Block diagram of the LDWT[37] ................................................................23
圖4.6 9/7提升式離散小波架構.....................................................................24
圖4.7 分離模組階段......................................................................................25
圖4.8 預測模組階段運算..............................................................................26
圖4.9 更新模組階段運算..............................................................................27
圖4.10 DLDWT預測模組階段選擇垂直角度θv之方向[38].........................28
圖4.11 DLDWT更新模組階段選擇垂直角度θv之方向[38]..........................29
圖4.12 方向性提升式架構空間域中九個方向............................................29
圖4.13 原影像................................................................................................30
圖4.14 9/7方向性提升式離散小波轉換結果圖...........................................31
圖4.15 二維離散小波重建示意圖................................................................32
圖5.1 DIBCO 2011 HW4 樣本[3] ...................................................................34
圖5.2 DIBCO 2011 HW4 樣本的Ground Truth[3] ............................................34
圖5.3 積分影像示意圖[40]...........................................................................35
圖6.1自適應濾波器架構圖...........................................................................39
圖6.2 原始最小均方演算法流程圖..............................................................40
圖6.3 變異數值分類流程圖..........................................................................42
圖6.4 改良式最小均方演算法多重權重遮罩運算流程圖………..………44
圖7.1 DIBCO 2009 H01 (a)原始影像 (b)實驗結果........................................46
圖7.2 DIBCO 2009 H03 (a)原始影像 (b)實驗結果........................................47
圖7.3 DIBCO 2009 H05 (a)原始影像 (b)實驗結果........................................48
圖7.4 DIBCO 2009 P01 (a)原始影像 (b)實驗結果.........................................49
圖7.5 DIBCO 2009 P04 (a)原始影像 (b)實驗結果.........................................49
圖7.6 H-DIBCO 2010 H01 (a)原始影像 (b)實驗結果....................................50
圖7.7 H-DIBCO 2010 H04 (a)原始影像 (b)實驗結果....................................51
圖7.8 H-DIBCO 2010 H05 (a)原始影像 (b)實驗結果....................................52
圖7.9 H-DIBCO 2010 H07 (a)原始影像 (b)實驗結果....................................52
圖7.10 H-DIBCO 2010 H08 (a)原始影像 (b)實驗結果..................................53
圖7.11 DIBCO 2011 HW2 (a)原始影像 (b)實驗結果.....................................54
圖7.12 DIBCO 2011 HW4 (a)原始影像 (b)實驗結果.....................................55
圖7.13 DIBCO 2011 HW7 (a)原始影像 (b)實驗結果.....................................56
圖7.14 DIBCO 2011 HW8 (a)原始影像 (b)實驗結果.....................................57
圖7.15 DIBCO 2011 PR1 (a)原始影像 (b)實驗結果.......................................58
圖7.16 DIBCO 2011 PR4 (a)原始影像 (b)實驗結果.......................................59
圖7.17 DIBCO 2011 PR6 (a)原始影像 (b)實驗結果.......................................60
圖7.18 DIBCO 2011 PR7 (a)原始影像 (b)實驗結果.......................................61
圖7.19 H-DIBCO 2012 H01 (a)原始影像 (b)實驗結果...................................62
圖7.20 H-DIBCO 2012 H03 (a)原始影像 (b)實驗結果...................................63
圖7.21 H-DIBCO 2012 H04 (a)原始影像 (b)實驗結果...................................64
圖7.22 H-DIBCO 2012 H05 (a)原始影像 (b)實驗結果...................................65
圖7.23 H-DIBCO 2012 H07 (a)原始影像 (b)實驗結果...................................66
圖7.24 H-DIBCO 2012 H11 (a)原始影像 (b)實驗結果...................................67
圖7.25 H-DIBCO 2012 H12 (a)原始影像 (b)實驗結果...................................68
圖7.26 DIBCO 2013 Hw2 (a)原始影像 (b)實驗結果......................................69
圖7.27 DIBCO 2013 Hw4 (a)原始影像 (b)實驗結果......................................70
圖7.28 DIBCO 2013 Hw5 (a)原始影像 (b)實驗結果......................................71
圖7.29 DIBCO 2013 Hw8 (a)原始影像 (b)實驗結果......................................72
圖7.30 DIBCO 2013 PR1 (a)原始影像 (b)實驗結果.......................................73
圖7.31 DIBCO 2013 PR3 (a)原始影像 (b)實驗結果.......................................74
圖7.32 DIBCO 2013 PR4 (a)原始影像 (b)實驗結果.......................................75
圖7.33 DIBCO 2013 PR5 (a)原始影像 (b)實驗結果.......................................76
圖7.34 H-DIBCO 2014 H01 (a)原始影像 (b)實驗結果..................................77
圖7.35 H-DIBCO 2014 H02 (a)原始影像 (b)實驗結果..................................78
圖7.36 H-DIBCO 2014 H08 (a)原始影像 (b)實驗結果..................................79
圖7.37 H-DIBCO 2014 H09 (a)原始影像 (b)實驗結果..................................80
圖7.38 H-DIBCO 2014 H10 (a)原始影像 (b)實驗結果..................................81
圖7.39 H-DIBCO 2016 H01 (a)原始影像 (b)實驗結果..................................82
圖7.40 H-DIBCO 2016 H02 (a)原始影像 (b)實驗結果..................................83
圖7.41 H-DIBCO 2016 H04 (a)原始影像 (b)實驗結果..................................84
圖7.42 H-DIBCO 2016 H05 (a)原始影像 (b)實驗結果..................................85
圖7.43 H-DIBCO 2016 H07 (a)原始影像 (b)實驗結果..................................86
圖7.44 DIBCO 2017 H02 (a)原始影像 (b)實驗結果......................................87
圖7.45 DIBCO 2017 H06 (a)原始影像 (b)實驗結果......................................88
圖7.46 DIBCO 2017 H07 (a)原始影像 (b)實驗結果......................................89
圖7.47 DIBCO 2017 H08 (a)原始影像 (b)實驗結果......................................90
圖7.48 DIBCO 2017 H11 (a)原始影像 (b)實驗結果......................................91
圖7.49 DIBCO 2017 H15 (a)原始影像 (b)實驗結果......................................92
圖7.50 DIBCO 2017 H16 (a)原始影像 (b)實驗結果......................................93
圖7.51 DIBCO 2017 H17 (a)原始影像 (b)實驗結果......................................94
圖7.52 DIBCO 2017 H18 (a)原始影像 (b)實驗結果......................................95
圖7.53 DIBCO 2017 H20 (a)原始影像 (b)實驗結果......................................96
圖7.54 原始拍攝影像..................................................................................109
圖7.55 旋轉後影像......................................................................................109
圖7.56 選取ROI影像..................................................................................110
圖7.57 字型Lucida Calligraphy文件影像 (a)原灰階影像 (b)實驗結果.....110
圖7.58 字型Informal Roman文件影像 (a)原灰階影像 (b)實驗結果.........111
圖7.59 字型Freestyle Script文件影像 (a)原灰階影像 (b)實驗結果..........111
圖7.60 字型Palace Script MT文件影像 (a)原灰階影像 (b)實驗結果........112
圖7.61 字型Viner Hand ITC文件影像 (a)原灰階影像 (b)實驗結果..........112
圖7.62 字型Vladimir Script文件影像 (a)原灰階影像 (b)實驗結果..........113
圖7.63 字型Blackadder ITC文件影像 (a)原灰階影像 (b)實驗結果..........113
圖7.64 字型Edwardian Script ITC文件影像 (a)原灰階影像 (b)實驗結果..114
圖7.65 字型Arial Unicode MS文件影像 (a)原灰階影像 (b)實驗結果…...114
圖7.66 字型Times New Roman文件影像 (a)原灰階影像 (b)實驗結果.....115

表目錄
表3.1 各種掃描器價格比較..........................................................................14
表3.2 不同類型掃描器的優缺點比較..........................................................17
表7.1 F-Measure平均比較表.........................................................................100
表7.2 F-Measure平均比較圖表.....................................................................100
表7.3 Pseudo-FMeasure平均比較表...............................................................101
表7.4 Pseudo-FMeasure平均比較圖表...........................................................101
表7.5 PSNR平均比較表................................................................................102
表7.6 PSNR平均比較圖表............................................................................102
表7.7 DRD平均比較表.................................................................................103
表7.8 DRD平均比較圖表.............................................................................103
表7.9 Intel i7-4790執行時間平均比較表.......................................................105
表7.10 Intel i7-4790執行時間平均比較圖表.................................................105
表7.11 Raspberry Pi 3B+執行時間平均比較表..............................................106
表7.12 Raspberry Pi 3B+執行時間平均比較圖表..........................................106
表7.13兩種硬體執行時間差平均比較表...................................................107
表7.14兩種硬體執行時間差平均比較圖表...............................................107

參考文獻
[1]	http://users.iit.demokritos.gr/~bgat/DIBCO2009/benchmark/
[2]	http://users.iit.demokritos.gr/~bgat/H-DIBCO2010/benchmark/
[3]	http://utopia.duth.gr/~ipratika/DIBCO2011/benchmark/
[4]	http://utopia.duth.gr/~ipratika/HDIBCO2012/benchmark/
[5]	http://utopia.duth.gr/~ipratika/DIBCO2013/benchmark/
[6]	http://users.iit.demokritos.gr/~bgat/HDIBCO2014/benchmark/
[7]	https://vc.ee.duth.gr/h-dibco2016/benchmark/
[8]	https://vc.ee.duth.gr/dibco2017/benchmark/
[9]	https://www.raspberrypi.org/products/raspberry-pi-3-model-b-plus/
[10]	https://www.raspberrypi.org/products/pi-noir-camera-v2/
[11]	N. Otsu, “A Threshold Selection Method from Gray-Level Histograms,” in IEEE Transactions on Systems, Man, and Cybernetics, vol. 9, no. 1, pp. 62-66, Jan. 1979.
[12]	J. Bernsen, “Dynamic Thresholding of Gray Level Image,” ICPR`86: Proceedings of International Conference on Pattern Recognition, Berlin, 1986, pp. 1251-1255.
[13]	W. Niblack, An Indroduction to Digital Image Processing, Prentice-Hall, pp. 115-116, 1986.
[14]	J. Sauvola, and M. Pietikainen, “Adaptive document image binarization,” Pattern Recognition, vol. 33, no. 2, pp. 225-236, 2000.
[15]	C. Wolf, and J-M. Jolion, “Extraction and Recognition of Artificial Text in Multime-dia Documents,” Pattern Analysis and Applications, vol. 6, no. 4, pp. 309-326, Feb. 2004.
[16]	M.-L. Feng and Y.-P. Tan, “Contrast adaptive binarization of low quality document images,” IEICE Electron. Express, vol. 1, no. 16, pp. 501-506, 2004.
[17]	D. Bradley, and G. Roth, "Adaptive Thresholding using the Integral Image,” Journal of Graphics Tools. vol. 12, no. 2. pp. 13-21, 2007.
[18]	K. Khurshid, I. Siddiqi, C. Faure, and N. Vincent, “Comparison of Niblack inspired Binarization Methods for Ancient Documents,” in Proc. Document Recognition and Retrieval XVI, DRR 2009, 16th Document Recognition and Retrieval Conference, part of the IS&T-SPIE Electronic Imaging Symposium, San Jose, CA, USA, Jan. 2009, pp. 1-10.
[19]	N. Phansalkar, S. More, A. Sabale and M. Joshi, “Adaptive local thresholding for de-tection of nuclei in diversity stained cytology images,” 2011 International Conference on Communications and Signal Processing, Calicut, 2011, pp. 218-220.
[20]	X. Chen, L. Lin and Y. Gao, “Parallel nonparametric binarization for degraded docu-ment images,” Neurocomputing, vol. 189, 2016, pp. 43-52.
[21] C. Tensmeyer and T. Martinez, “Document Image Binarization with Fully Convolu-tional Neural Networks,” 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR), Kyoto, 2017, pp. 99-104.
[22] B. Gatos, K. Ntirogiannis and I. Pratikakis, “ICDAR 2009 Document Image Binarization Contest (DIBCO 2009),” 2009 10th International Conference on Document Analysis and Recognition, Barcelona, 2009, pp. 1375-1382.
[23]	I. Pratikakis, B. Gatos and K. Ntirogiannis, “H-DIBCO 2010 - Handwritten Document Image Binarization Competition,” 2010 12th International Conference on Frontiers in Handwriting Recognition, Kolkata, 2010, pp. 727-732.
[24]	I. Pratikakis, B. Gatos and K. Ntirogiannis, “ICDAR 2011 Document Image Binarization Contest (DIBCO 2011),” 2011 International Conference on Document Analysis and Recognition, Beijing, 2011, pp. 1506-1510.
[25]	I. Pratikakis, B. Gatos and K. Ntirogiannis, “ICFHR 2012 Competition on Handwritten Document Image Binarization (H-DIBCO 2012),” 2012 International Conference on Frontiers in Handwriting Recognition, Bari, 2012, pp. 817-822.
[26]	I. Pratikakis, B. Gatos and K. Ntirogiannis, “ICDAR 2013 Document Image Binari-zation Contest (DIBCO 2013),” 2013 12th International Conference on Document Analysis and Recognition, Washington, DC, 2013, pp. 1471-1476.
[27]	K. Ntirogiannis, B. Gatos and I. Pratikakis, “ICFHR2014 Competition on Handwrit-ten Document Image Binarization (H-DIBCO 2014),” 2014 14th International Con-ference on Frontiers in Handwriting Recognition, Heraklion, 2014, pp. 809-813.
[28]	I. Pratikakis, K. Zagoris, G. Barlas and B. Gatos, “ICFHR2016 Handwritten Docu-ment Image Binarization Contest (H-DIBCO 2016),” 2016 15th International Con-ference on Frontiers in Handwriting Recognition (ICFHR), Shenzhen, 2016, pp. 619-623.
[29]	I. Pratikakis, K. Zagoris, G. Barlas and B. Gatos, “ICDAR2017 Competition on Document Image Binarization (DIBCO 2017),” 2017 14th IAPR International Con-ference on Document Analysis and Recognition (ICDAR), Kyoto, 2017, pp. 1395-1403.
[30]	S. Haykin, Adaptive Filter Theory, 2nd edition, Pentice Hall, 1991.
[31]	O. N. Gerek and A. E. Cetin, “A 2-D orientation-adaptive prediction filter in lifting structures for image coding,” in IEEE Transactions on Image Processing, vol. 15, no. 1, pp. 106-111, Jan. 2006.
[32]	http://www.newimage.com.tw/Tw/ProductList.asp?SortID=8
[33]	https://www.etmall.com.tw/Shoot-Holic-%E6%94%9D%E6%89%8BKC5M01%E6%8B%8D%E6%94%9D%E5%BC%8F%E6%96%87%E4%BB%B6%E6%8E%83%E6%8F%8F%E5%99%A8-%E6%90%AD%E8%B4%88OCR%E6%96%87%E5%AD%97%E6%A2%9D%E7%A2%BC%E8%BE%A8%E8%AD%98%E5%8A%9F%E8%83%BD/i/1726960
[34]	https://www.epson.com.tw/
[35]https://tw.mall.yahoo.com/item/%E2%98%86%E9%99%90%E6%99%82%E2%86%99%E4%B8%8B%E6%AE%BA%E2%98%86handy-scan-%E6%89%8B%E6%8C%81%E6%8E%83%E6%8F%8F%E6%A9%9F-%E5%84%80-%E6%8E%83%E6%8F%8F-%E7%AD%86-Co-p050457500454
[36]	I. Daubechies and W. Sweldens, “Factoring wavelet transforms into lifting scheme,” The Journal of Fourier Analysis and Applications, vol. 4, pp. 247-269, 1998.
[37]	C.-H. Hsia and J.-M. Guo, “Efficient modified directional lifting-based discrete wavelet transform for moving object detection,” Signal Processing, vol. 96, part B, pp. 138-152, Mar. 2014.
[38]	M. N. Do and M. Vetterli, “The contourlet transform: an efficient directional multi-resolution image representation,” in IEEE Transactions on Image Processing, vol. 14, no. 12, pp. 2091-2106, Dec. 2005.
[39]	https://zh.wikipedia.org/wiki/%E7%A7%AF%E5%88%86%E5%9B%BE
[40]https://cg2010studio.com/2012/04/24/%E7%A9%8D%E5%88%86%E5%BD%B1%E5%83%8F-integral-image/
[41]	K. Ntirogiannis, B. Gatos and I. Pratikakis, “Performance Evaluation Methodology for Historical Document Image Binarization,” in IEEE Transactions on Image Pro-cessing, vol. 22, no. 2, pp. 595-609, Feb. 2013.
[42] H. Lu, A. C. Kot and Y. Q. Shi, “Distance-reciprocal distortion measure for binary document images,” in IEEE Signal Processing Letters, vol. 11, no. 2, pp. 228-231, Feb. 2004.