【作者】 杨家辉；

【导师】 王建英；

【作者基本信息】 西南交通大学 ， 计算机应用技术， 2004， 硕士

【摘要】 车辆牌照的自动识别是数字图像处理技术与模式识别技术在交通信息系统中的重要应用。随着交通管理信息化的快速发展，车辆牌照识别技术已经成为当前研究的一项重要课题。由于现实环境中各种不可预知的因素，例如光线的变化、牌照的倾斜、牌照在图像中大小及位置的不确定性、复杂的干扰背景等，都给牌照的识别增加了很大的难度。从目前的研究成果来看，车辆牌照识别系统还远未达到完善的程度，利用各种图像处理的新理论和新方法仍是当前研究的热点之一。 在车辆牌照识别系统中，牌照的定位是最为关键的环节。为了寻找一种适应性优异的牌照定位方法，论文对数学形态学闭开运算和滤波运算、图像边缘检测算法以及基于HSV(Hue、Saturation、Value)颜色模型的色彩分析等图像处理技术做了仔细的研究与实现。对各种形态运算模板尺度的动态选取、各种边缘梯度算子的效果检测以及不同光照条件下基于区间阀值的色彩的判别等关键环节作了细致分析，并通过大量图片对分析结果进行了反复实验验证。在此基础上，提出了一种利用牌照的纹理、颜色及体态等综合特征进行牌照定位的新方法。实验结果表明，该方法在各种光照和复杂背景条件下均有很好的适应性，对牌照定位有很高的准确性。 牌照字符切分效果的好坏对后续字符识别所采用的算法与识别准确度有着重要的影响。为了减少字符识别的复杂程度，论文对字符的精确切分也提出了一系列有效的预处理方法，包括基于直线拟合与角度探测法的牌照水平及垂直倾斜校正、基于彩色投影法的牌照背景纹理过滤、基于二值投影阀值处理的非字符区域的剔除等。最终分割出的牌照字符有较好的效果，为模板匹配等简单快速的字符识别方法提供了良好的前提。 论文提出的牌照定位方法对牌照的纹理、体态及彩色特征作了充分考虑并加以综合利用，避免了以往定位方法对复杂环境适应性差的弊端。提出的多重字符分割预处理方法能够对各种不良状况的牌照加以改善，使分割出的字符效果具有最小的几何形变，减少了以往方法由于粗略分割所造成的后续字符识别率的下降。更多还原

【Abstract】 Vehicle License Plate Recognition (VLPR) is an important application of digital image processing and pattern recognition in the transportation information system. With the fast development of transportation information system, VLPR has become an important research subject nowadays. In real conditions, a variety of causation would made License Plate Recognition more difficult, such as variety in illumination conditions, sloping license plate, the uncertainty of size and position of license plate in the captured image, the complex of image background, and etc. Current technique of License Plate Recognition is far away from perfect. Using new theory and new technology to reform license plate recognition technique is one of the research keys.License plate’s location is the key of VLPR system. To find out a highly adaptive plate’s location method, mathematical morphology close and open operation, image edge detect and color analysis based on HSV (Hue, Saturation, Value) model are researched and realized. And a series of key steps of plate’s location are carefully analyzed and experimented with a mount of plate images too, including calculating dynamic size of the morphology template, validating effects of a variety of image edge detect operators, color distinguish method based on range threshold. Based on these work, a new plate location method using plate’s texture, shape, color characteristics was proposed. The experimental results show that the method has strong adaptability to unconstrained illumination conditions and irregular background, and is efficient in locating license plate.The result of characters segmentation has remarkable effect on the choosing of the character recognition algorithm and the accuracy of character recognition. To reduced to the complexity of chars segmentation, paper brought forward a series of pretreatment methods to the retrieved plate, including using line imitate method and angle detect method to correct the plate’s slope angle, using color projection method to eliminate conglutinative texture edge, and using bi-value image project method to eliminate plate’s rivets and frame. The multiple pretreatments result in the favorable outcome, and provided the better preconditions for simple charactersrecognition method that based on template match. The VLPR location method proposed in this paper fully considers and utilizes the texture, shape and color features of VLPR in VLPR location, and has better adaptability than the method used before. The proposed multiple pretreatments method can effectively mend unfavorable VLPR, based on the mended VLPR, the final segmented characters has less distortion than coarse segmentation method, and can increase accuracy of characters recognition.更多还原