應用圖像配準技術之歪斜車牌辨識系統研究

The Study of Skewed Vehicle Plate Recognition System Applying Image Registration Techniques

高世錦(Shih-Chin Kao)

指導教授 ： 王順源 ； 曾傳蘆

國立臺北科技大學/機電學院/電機工程系所/碩士(2011年)

摘要

本研究將基於特徵點之圖像配準(image registration)技術應用於車牌影像之歪斜校正 (skewness correction)，以利樣板比對法(template matching)之辨識。本文提出以最小外接矩形(minimum enclosing rectangle, MER)演算法計算由歪斜車牌字元影像之四邊形圍線端點座標，作為近似車牌四個角落之特徵點，並自動偵測出歪斜影像形變之特徵點，作為圖像配準之最佳控制點。此方法解決了多數歪斜車牌在影像辨識上僅處理影像旋轉之剛性變換(rigid transformation)問題，而忽略了拍攝車牌所形成之前後傾斜問題。
本研究將提出之車牌定位、歪斜車牌圖像配準及字元辨識方法，應用於路邊車輛車牌辨識。經實驗結果證明，車牌定位以菱形結構元素(diamond-shaped structuring element)作為群組化定位方式辨識率達98.5%，在車牌字元完全取出條件下，歪斜車牌轉正成功率趨近100 %。字元辨識方面成功率達95.86%。

關鍵字

圖像配準 ； 最小外接矩形 ； 歪斜校正 ； 菱形結構元素

並列摘要

The purpose of this study was to apply feature-pointed image registration on skewness correction of vehicle plate and improve identification of template matching. Minimum enclosing area rectangle algorithm suggests a similarity measure for registering two images by comparing vertex of quadrangle of skewed vehicle plate’s character image and feature points of vehicle plate. The feature points of skewed image are automatically detected to be the optimized control points. This technique solves the problem that the traditional skewed plate’s image identification only processes rigid transformation but neglects panned plate skewness.
Finally, street vehicle plate identification was performed to examine the proposed method. Results show that overall successful recognition rate by using diamond-shaped structuring element was 98.5%; under the condition of taking out characters completely, skewed plate has approximately 100% successfully recognized rate; and 95.86% success recognition rate of character identification is achieved.

並列關鍵字

Image Registration ； Minimum Enclosing Rectangle ； Skewness Correction ； Diamond-shaped Structuring Element

參考文獻

[2] 王振興，多標的汽機車車牌辨識系統之研究，碩士論文，元智大學資訊管理學系，桃園，2003。

[3] 呂炎州，不需字元切割的車牌辨識法，碩士論文，靜宜大學資訊管理系，臺中，2004。

[4] 陳志成，自動贓車偵測系統，碩士論文，臺灣師範大學，資訊工程研究所，臺北，2009。

[5] M. Sarfraz, S. A. Mahmoud and Z. Rasheed, “On skew estimation and correction of text,” Proc. of IEEE on Computer Graphics, Imaging and Visualisation, Aug. 2007, pp. 308-313.

[6] Y. Cheng, J. Lu and T. Yahagi, “Car license plate recognition based on the combination of principal components analysis and radial basis function networks,” Proc. ISCP Int. Conf on Signal Processing, vol. 2, 2004, pp. 1455-1458.