臺灣博碩士論文加值系統

光學文字辨識(OCR)已經是個發展多年的技術，但是現今卻還沒有一個辨識率百分之百的工具。這個問題是因為辨識影像的來源有很多種不同的狀況，例如文字影像的品質、各式各樣的文章排版、不同的語言、千變萬化的字體、大大小小的文字。只要有其中一項變動，就會對辨識率有極大的影響。
本研究是要應用在本實驗室之 Korat 自動化回歸測試系統上，辨識影像的來源是由 Korat 擷取待測系統的螢幕。因為辨識的影像來源是從螢幕截圖，所以影像會是端正、乾淨且無雜訊。基於這個特性，本研究不需要面對一般 OCR 工具會遇到的情況，如影像歪斜和雜訊干擾的問題。但是會面臨到辨視率需要100%的需求，所以即使在市面上有諸多的 OCR 工具的辨識率都有 88%-95%左右，但仍然無法符合實際應用在 Korat 的系統上的要求。
本研究使用樣板比對的方法結合動態規劃的演算法，在所有可能的辨識組合中，比較剩餘的像素總和，以剩餘最小的組合視為最佳解，藉此辨識率 100%。

Optical Character Recognition has been studied for many years. However, no OCR tools can claim 100% recognition rate because of the variation in image quality, documentation layout, character fonts and sizes. When there are changes in one of them, recognition rate is often greatly impacted.
Korat is an image-based test regression tool developed in our lab. Korat captures the screen image from a system under test. Therefore, the image is clean, no noises, and no rotation. Based on this condition, we do not deal with situations like image noises, which makes OCR a difficult problem in this thesis.
In Korat's practical applications, nearly 100% recognition rate is often required. So even there are many existing OCR tools with 88-95% recognition rate, they do not meet the requirements of Korat's practical applications.
In this research, we combine the template matching and dynamic programming algorithm to find the optimal solution with the smallest sum of remaining pixels in all possible combinations of recognition so that the recognition rate could be achieved 100% at nearly.

摘要 i
Abstract ii
Contents iii
List of Figures v
Chapter 1 Introduction 1
1.1 Automated Regression Testing 1
1.2 Practical OCR Precision Requirement of Korat 2
1.3 Problems in Segmentation 3
1.4 OCR by Template Matching and Dynamic Programming 5
Chapter 2 Background and Related Works 6
2.1 Korat [3] 6
2.2 True Type Font 7
2.3 Optical Character Recognition Technology [4][5] 7
2.3.1 Pre-preprocessing [6] 8
2.3.2 Classification 10
2.3.3 Post-processing [13] 12
2.4 Existing OCR Tools 12
2.4.1 Tesseract-OCR[8][9] 12
2.4.2 ABBYY FineReader[10] 12
2.4.3 Adobe Acrobat Pro DC[11] 14
2.5 Template Matching 15
2.6 Dynamic Programming Algorithm 16
Chapter 3 Combine Tm and DP for OCR 19
3.1 Problems Analysis 19
3.2 Image Preprocessing 20
3.3 Classification by Template Matching 20
3.4 The research problems 21
3.5 Solution to deal with influence of boundary pixels 23
3.6 Solution to deal with overlapped problem 23
3.7 Problems in TMM and minimized template 24
3.8 Combination by TMM and DP 25
3.8.1 The optimization problem comes to mind 25
3.8.2 DP to solve the optimization problem 25
Chapter 4 Preparation of Template Data 28
4.1 Preparation 28
4.2 Image Processing in Template Data 28
4.3 Character Display Tool. 29
Chapter 5 Experiment 32
5.1 Description of Test Data 32
5.2 Comparison of Recognition Rate 32
5.3 Special cases 35
5.4 Speed of Recognition 36
5.5 Font Extension 37
5.6 Test Result Description 37
Chapter 6 Conclusion and Future Works 38
Chapter 7 Reference 40
 

[1] Leung, H. K. and L. White (1989). Insights into regression testing [software testing]. Software Maintenance, 1989., Proceedings., Conference on, IEEE.
[2] Wiki. Computer font. Available from: https://en.wikipedia.org/wiki/Computer_font#Outline_fonts.
[3] Chen, X.-C. Korat: An O.S.-independent Capture/Replay Test Automation System. Institute of Computer Science & Information Engineering, National Central University, National Central University. Master of Science.
[4] Zhu, W., Y. Liu, and L. Hao. A Novel OCR Approach Based on Document Layout Analysis and Text Block Classification. in 2016 12th International Conference on Computational Intelligence and Security (CIS). 2016.
[5] Govindan, V.K. and A.P. Shivaprasad, Character recognition — A review. Pattern Recognition, 1990. 23(7): p. 671-683.
[6] Wiki. "Optical character recognition." from https://en.wikipedia.org/wiki/Optical_character_recognition..
[7] Trier, Ø. D., et al. (1996). "Feature extraction methods for character recognition-a survey." Pattern recognition 29(4): 641-662.
[8] Smith, R., An Overview of the Tesseract OCR Engine, in Proceedings of the Ninth International Conference on Document Analysis and Recognition. 2007, IEEE Computer Society. p. 629-633.
[9] Smith, R.W. History of the Tesseract OCR engine: what worked and what didn't. in IS&T/SPIE Electronic Imaging. 2013. International Society for Optics and Photonics.
[10] ABBYY. Feature of ABBYY FineReader 14 Professional. Available from: https://www.abbyy.com/en-us/.
[11] Adobe. Feature of Adobe Acrobat DC. Available from: https://acrobat.adobe.com/us/en/acrobat/how-to/ocr-software-convert-pdf-to-text.html.
[12] Wiki. Template Matching. Available from: https://en.wikipedia.org/wiki/Template_matching
[13] Y. Bassil, a.M.A., OCR post-processing error correction algorithm using Google’s Online spelling suggestion. Journal of Emerging Trends in Computing and Information Sciences, 2012.
[14] Gupta, G., et al. Document Layout Analysis and Classification and Its Application in OCR. in 2006 10th IEEE International Enterprise Distributed Object Computing Conference Workshops (EDOCW'06). 2006.
[15] Wiki. Dynamic programming. Available from: https://en.wikipedia.org/wiki/Dynamic_programming.
[16] Vamvakas, G., et al. A Complete Optical Character Recognition Methodology for Historical Documents. in 2008 The Eighth IAPR International Workshop on Document Analysis Systems. 2008.
[17] OpenCV. OpenCV: Template Matching. Available from: http://docs.opencv.org/trunk/d4/dc6/tutorial_py_template_matching.html.