車牌辨識系統至今已發展多年，大多數的車牌辨識系統皆設置於固定地點，舉凡高速公路、停車場等地方，少有研究探討以動態攝影方式之車牌辨識系統，台灣汽機車密度高居世界排名，特別是機車密度為全世界第一，警方目前於查緝贓車時多為使用手動輸入車牌號碼來查緝可疑車輛是否為贓車，然而若在執勤巡邏時，將行車記錄器安裝於安全帽上或是汽車內，在行動的過程中依然可藉由車牌辨識系統做動態贓車搜尋。

此系統主要由車牌定位、車牌處理、字元辨識等模組組成，車牌定位模組主要利用邊緣偵測中Sobel的水平微分遮罩運算子來取得影像中較密集的垂直紋理，並透過車牌特徵進一步的找出車牌在影像中的位子。車牌處理模組主要針對定位後的牌照做二值化的處理並取得更精確的牌照位子，在此模組中對於車牌字色為白色的車牌會做反向的動作。字元切割模組主要將二值化後的車牌依序取出車牌字元。最後的字元辨識模組則利用開放原始碼Tesseract-OCR，利用自己所訓練出來的字庫，來做為車牌的字元辨識。

此系統在汽機車車牌定位成功率為98.61%，辨識率為90.52%。

License plate recognition systems have been developed for many years. Most of them are installed with video cameras at fixed locations such as highway toll stations or parking lots. However, very few studies have focused on license plate recognition system with moving camera. In Taiwan, the density of vehicles is ranked top in the world. Taiwan is also known as the highest motorcycle-density country in the world. Presently the police investigating stolen vehicles use system by manually entering license plate numbers to check if they are stolen vehicles or not. In this thesis, we develop a plate license automatic recognition system with moving cameras. This allows polices to record the car or motor-cycle plate numbers while patrolling in their cars.

This system comprises the following modules: image pre-processing, license plate localization, plates processing, and optical character recognition. First, pre-processing module is used to converts color input video into 8-bits grayscale images. Next, license plate localization module uses Sobel edge detection operator to find strong vertical texture and locates the license plate area in images. Then, the plate-processing module uses adaptive threshold processing to get more accurate area of license plate. In this module, if the license plate characters are white, they will be transformed into black characters. Finally, optical character recognition module is used to recognize the character. We have adapted the Tesseract-OCR engine to do the task. We use captured plate charter templates to train its characters database and get good recognition result..

Experimental results show that our system has 98.61% successful license plate areas localization rate. It achieves 90.52% successful license plate recognition.

參考文獻
[1] 趙怡欽，「零碳排放的夢想-淨煤(氣化和富氧)技術，綠引擎與碳捕捉封存」成功大學，台南（2010）。
[2] 陳建宏，「以車牌辨識技術檢驗機車定檢狀況之研究」，碩士論文，大同大學，台北（2008）。
[3] 何坤達，「嵌入式車牌辨識系統」，碩士論文，國立台灣科技大學，台北（2006）。
[4] Veengle,MD93.[Online].Available: http://www.veengle.com/s/MD93.html
[5] 陳翔傑，「自動化車牌辨識系統設計」碩士論文，國立中央大學，桃園(2005)。
[6] 莊家龍，「車輛偵測與車牌辨識系統」，碩士論文，國立中正大學，嘉義（2005）。
[7] 鐘國亮，影像處理與電腦視覺，第三版，東華書局，台北，（2006）。
[8] N.Otsu, “A threshold selection method from gray level histogram,”IEEE Trans.
On Systems,Man,and Cybernetics,pp.62-66（1979）
[9] M.Sonka,V.Hilavac,R.Boyle,ImageProcessing,Analysis,and Machine Vision,
PWS Publishing,（1998）
[10] J.P.Parker,Algorithms for Image Processing and Computer Vision,JohnWisey
&Sons,Inc.,New York,NY,(1996)
[11] E.R.Davies,Machine Vision.Theroy,Algorithm,Practicalities,3rded.,Morgan
Kaufmann,San Francisco,(2005)
[12] A.Rosenfeld,”Connectivity in Digital Pictures,”Journal of the ACM(JACM),
vol.17 no.1,pp.146-160,(1970)
[13] F.Chang,C-J.Chen,and C-J.Lu,”A Linear-Time Component-Labeling Algorithm
Using Contour Tracing Technique,”Computer Vision and Image Understanding,,vol.93,no.2,pp.206-220,(2004)
[14] 呂彥州，「不需字元切割的車牌辨識法」，碩士論文，靜宜大學，台中（2003）。
[15] R. Brunelli, Template Matching Techniques in Computer Vision: Theory and
Practice, Wiley, (2009 )
[16] 徐志融，「基於手持式行動裝置實現車牌資訊管理與辨識系統」，碩士論文，國立台灣科技大學，台北(2010)。
[17] M.J.Burton and N.G.Bourbakis,”A comparison of OCRnext term algorithm,”Proc. 9thInt.Conf. on Simulation,vol.19,(1998)
[18] Gonzalez,Woods,Digital Image Processing,3rdEdition,Prentice Hall,(2008)
[19] jTessBoxEditor.Available: http://sourceforge.net/projects/vietocr/files/jTessBoxEditor/
[20] 李志文，「嵌入式即時多標的汽機車牌照辨識系統」，碩士論文，國立台灣科技大學，台北(2008)。
[21] H.J.Lee,S.Y.Chen and S.Z.Wang,”Extraction and recognition of license plates of motorcycles and vehicles on highways,”IEEE Proceedings of the 17th International Conference on Pattern Recognition,vol 4,pp.356-359,(2004)
[22] B.Chen,W.Cao and H. Zhang, “An efficient algorithm on vehicle license plate location,”IEEETrans.on Image Processing ,pp.1386-1389,(2008)