本方法提出一自動擷取車牌位置與車號字元的系統，所針對的問題為複雜背景與靜態的大影像。本文採用統計方式找出可能字元邊界，並利用這些可能邊界搜尋邊界密集出現區域，再利用這些區域找尋近似灰階區塊（本論文針對白底黑字車牌），當邊界與色彩皆接近車牌時，再利用字元與樣板灰階的相關性進行大量影像統計，配合影像處理方法對車牌擷取與字元辨識上提出演算法進行自動偵測。
本文針對戶外車輛做為拍攝目標，戶外於影像上有相當多的不確定因素，如陰影、背景等，而針對不同的影像統計加上影像處理方法找出較適合演算法，並利用統計學方式對字體辨識，對於較相近字體取出字體與樣板相關係數及慣性矩並配合類神經網路做進一步分類。

In this thesis, we develop a method to automatically detect and recognize the vehicle license plates. By using a large number of images to study statistically several important features of the license plates, this work has developed several methods to systematically detect and recognize the license plates. In particular, these methods are used to detect edges, to locate regions with densely distributed edges, to detect region with grey color and to identify character shapes.
This work restricts its study on outdoor environment. Many environmental uncertainties such as lighting and background complexity should also be considered. By taking these factors into consideration, our algorithm can first detect the license plates. Next, our system uses a two-stage approach to recognize the characters on the plates. Most of the characters can be correctly recognized in the first stages by using conventional template-based method. However, a moment-feature based method is applied to two pairs of characters which can not be accurately classified by the template-based method.
Experimental results are given to demonstrate the effectiveness of the proposed approach. In order to improve the proposed approach in the future, this work also studies a relatively small portion of plates that can not be perfectly handled by the proposed approach.

目錄 I
表目錄 III
圖目錄 IV
第一章 緒論 1
1.1. 前言 1
1.2. 研究動機 1
1.3. 相關研究 2
1.3.1. 車牌定位 3
1.3.2. 字元辨識 4
1.4. 論文架構 5
第二章 車牌辨識系統 6
2.1. 系統概要 6
2.2. 系統流程與簡介 7
2.3. 影像資料庫建立 14
第三章 車牌定位方法 15
3.1. 邊界偵測 15
3.1.1. 邊界統計方法 15
3.1.2. 邊界限制條件 18
3.1.3. 邊界實驗結果 19
3.2. 區塊密集度偵測 22
3.3. 車牌候選區域框選 23
3.3.1. 車牌候選區域形態偵測 23
3.3.2. 色彩偏差判定 25
3.3.2.1.色彩特徵選擇 25
3.3.2.2.色彩特徵統計方式 26
3.3.2.3.色彩門檻使用與過濾結果 28
第四章 車牌字元切割與辨識 30
4.1. 字元切割 30
4.1.1. 二值化方式 30
4.1.2. 連結破碎字元 33
4.1.3. 區塊化排列切割 35
4.2. 字元樣板比對字元區域 35
4.2.1. 字元比對方式 35
4.2.2. 樣板比對統計方式 36
4.2.3. 字元比對過濾候選區域 37
4.3. 字元辨識 40
4.3.1. 樣板比對辨識車牌字元 40
4.3.2. 相似字元再次比對 42
4.3.2.1.OFMMs特徵變數 42
4.3.2.2.特徵變數選取 43
4.3.2.3.MLP網路架構 44
4.3.2.4.實驗結果 46
第五章 實驗方法與結果 48
5.1. 車牌定位結果 49
5.2. 字元切割辨識結果 50
第六章 總結與未來展望 56
附表一：Canon S110詳細規格 57
參考文獻 58

[1] Intelligent Transportation Society of Taiwan, http://www.its-taiwan.org.tw/
[2] 陳麗奾, “在未設限環境下車牌的定位與辨識”, 國立台灣師範大學資訊教育研究所, 碩士論文, 2000
[3] 李宗憲, “以VQ編碼方法為基礎的物件偵測方法”, 國立中山大學機械與機電工程研究所, 碩士論文, 2002
[4] Cheng, Y., Lu, J., Yahagi, T., “Car License Plate Recognition based on the Principal Components Analysis and Radial Basis Function Networks”, Proceedings of the IEEE International Conference on Signal Processing(ICSP `04), Vol. 2, pp. 1455-1458, 2004
[5] 王中山, “使用小波轉換於車牌偵測”,　國立中山大學機械與機電工程研究所, 碩士論文, 2004
[6]　Parisi R., Di Claudio E.D., Lucarelli G., Orlandi G.,　“Car plate recognition by neural networks and image processing”,　Proceedings of the 1998 IEEE international Symposium on Circuits and Systems, Vol.3, pp.195-198, 1998
[7]　Torres-Mendez, L.A., Ruiz-Suarez, J.C., Sucar, L.E., Gomez, G.,　“Translation, rotation, and scale-invariant object recognition”,　IEEE Transactions on Systems,Man,and Cybernetics, Vol.30, Issue:1, pp.125-130, 2000
[8]　Mi-Ae Ko, Young-Mo Kim, “A simple OCR method from strong perspective view”, Applied Imagery Pattern Recognition Workshop, 2004. Proceedings. 33rd. pp.235-240
[9]　鍾國亮,　“影像處理與電腦視覺”,　東華書局,　第二版
[10]　Chao Kan, Srinath, Mandyam D.,　“Invariant character recognition with Zernike and orthogonal Fourier-Mellin moments”,　Pattern Recognition,　Vol. 35, Issue:1, pp.143-154, 2002
[11]　P. Pudil, J. Novovicova and J. Kittler.,　“Floating search method in feature selection”,　Pattern Recognition Letter, Vol. 15, Issue:11, pp.1119-1125, 1994
[12]　陳朕寬,　“彩色車牌的偵測方法”,　國立中山大學機械與機電工程研究所, 碩士論文, 2005