车牌识别技术研究与实现

【作者】 陶军；

【导师】 杨余旺；

【作者基本信息】 南京理工大学 ， 模式识别与智能系统， 2004， 硕士

【摘要】 车牌识别技术综合了计算机视觉技术和模式识别技术，在桥梁路口自动收费、停车场无人管理、违章车辆自动记录等领域有着广泛的应用。 本文对车牌识别系统的几个主要步骤：图像预处理、车牌定位、字符分割和字符识别，分别进行了已有方案的比较和算法的设计。 图像预处理：采用灰度变换、同态滤波和平滑等方法对图像进行处理，提高了图像的质量。 车牌定位：这是车牌识别系统中至关重要的一步。先对现有的各种定位算法进行了比较和分析，并总结了这些算法的共同点。在此基础上，提出了自己的定位方法，该方法包括粗定位和精定位，并且采用了灰度特征法和颜色特征法相结合的新方法。 字符分割：首先去除车牌的上下边框，然后利用连通域法分割得到单独的字符，并对粘连的字符块做进一步的处理。 字符识别：利用Foley-Sammon最佳鉴别变换进行字符识别，取得了较高的识别率和可靠性。 本文算法对牌照在图像中的位置没有限制，对牌照的倾斜、变形、字符的污染、模糊有较强的抗干扰能力，对于外界光线强度和图像对比度的变化有较强的适应能力。系统测试结果，牌照定位的准确率达96％以上，字符识别率达90％以上。更多还原

【Abstract】 The License Plate Identification System mainly depends on the computer vision technology and the pattern recognition technology, and it plays an important role in intelligent traffic control system, paking lot monitor system and automatic charging system.The algorithms of all steps for implementing a complete system include image preprocessing, license plate location, character segmentation and character recognition, and these are studied further in this thesis.Image preprocessing: The image is processed by several algorithms including gray level stretch, homomorphic processing and image smoothing, and then the quality of image gets better.License plate location: This aspect is an extremely important step in this system and is also emphased in this paper. Several algorithms proposed traditionally are compared with that in this thesis. Several common points are summarized, and a new location algorithm is then put forward. The location process consists of two stages: coarse location followed by exact location, and we take advantages of both gray character and color character.Character segmentation: First, the top and bottom frame of the license plate are cut off. Secondly, the corresponding connection region is used to divide up the isolate character. At the same time, some efficient methods are provided for conglutination treatment.Character recognition: Foley-Sammon Discriminant Vector is employed in this paper in order that a satisfactory recognition effect is ensured.The algorithm we adopted is not restricted by the location of the plate in the image, and it has strong anti-jamming ability to lean, distortion, pollution and blur. The algorithm is also adaptable to the varieties of circumstance light and image contrast. System test result: the veracity of plate location reaches above 96%, and character identification rate above 90%.更多还原