【作者】 郭娟；

【导师】 杨为民；

【作者基本信息】 东北大学 ， 检测技术与自动化装置， 2005， 硕士

【摘要】 车辆牌照自动识别是计算机视觉与模式识别技术在智能交通领域的重要应用,是实现交通管理的重要环节之一。它在高速公路、城市道路和停车场等项目管理中占有不可取代的重要地位,它的广泛应用必将有助于我国交通管理自动化的进程。本文的处理手段是对输入的汽车图像进行预处理,运用车牌定位、字符分割、字符识别的方法提取车牌上的字符串,并以文本的形式输出。其目的是识别出车牌上的字符,用于车辆资料的管理及其安全保障。 车牌定位是车辆牌照自动识别中最关键的一步。本文通过对200多幅汽车图片进行观察,得出将图像进行灰度化、仅提取图像的纵向边缘可突出车牌部分,同时消除外界环境的干扰,最后得到精确的车牌图像。字符分割的难点在于噪声、字符粘连和断裂对分割的影响,本文充分利用车牌的先验知识和分布特征,采用对比度拉伸和改进的Bernsen算法相结合对车牌图像进行二值化,运用Hough算法进行倾斜校正,再对字符进行切分,最后用气泡法对单个字符除噪。字符识别是整个车牌识别过程中最重要的一环,关键在于特征与分类器的选择和匹配。本文提出一种对字符进行归一化,将主成分分析法(PCA)和最近邻法相结合的算法对字符进行识别,由于该算法实现了特征降维,且是一种特殊的非线性分类方法,所以识别效果非常理想。 由实验所得的结果,本文算法能较准确定位、分割车牌并进行识别,系统的性能良好。因此,多种预处理与识别技术的有机结合能提高系统识别的能力,此方法对车辆牌照自动识别在实际中的应用有一定的参考和借鉴价值。更多还原

【Abstract】 Vehicle license plate automatic recognition is the important application of computer vision and pattern recognition in the intelligent transportation domain, and is one of the important parts in transportation management. It plays important roles in the project management of expressway, urban road and parking area etc. Its wide application will contribute to the process of the traffic mechanography of our country.Its measure is to deal with the inputted images, exploit license plate location character segmentation character recognition to extract the character strings, and express these strings in text form. Moreover the aim is to recognize the character for information management and safe guarantee.The key to license plate recognition is the license plate location. Observing 200 vehicle pictures, we can get the accurate license plate area via reducing the noise by graying only extracting the vertical edge. The difficulties in the character segmentation are the noise character conglutination and rupture etc. However, the paper makes full use of the prior knowledge of the license plate, combining the contrast stretching with the improved Bernsen algorithm for the binary, utilizing the Hough algorithm to correct the slant plate, and then segmenting character, finally employing bubble-method for single character to reduce noise. What’s more, the method of character recognition takes an important part in the whole system, and the key technique lies in the matching of feature and classifier. Therefore, the paper puts forward an algorithm that makes the character in the same size, and it links the PCA and nearest neighbour algorithm to recognize the character. However it has perfect effect because it is a special nonlinear method as well as descends dimension.As a result, it proves that this recognition system can be relatively accurate to locate license plate and recognize characters, and the performance of the system is good. It shows that combined pretreatments and recognition techniques can improve the ability of recognition, so the algorithm mentioned in the article has a certain reference value for the actual application of vehicle license plate system.更多还原