【作者】 张玮；

【导师】 王平；

【作者基本信息】 南昌大学 ， 通信与信息系统， 2008， 硕士

【摘要】 随着交通需求量的不断提高,基于图像处理的车牌识别技术成为解决交通违规、监控、收费等领域问题的重要手段、途径。本文主要针对该技术的图像采集、预处理、字符分割和识别几个方面,展开了详细的理论介绍、传统算法的优劣分析,并结合传统算法的优点通过改进得到本文的算法,给出实验结果。本文针对传统Berson法、Ostu法容易受到噪声干扰等缺点,改进Ostu法,得到基于Ostu的动态阈值法。它采用了分区域动态阈值更新策略,能够在有效地抑制复杂光照环境干扰的同时大大减少噪声的参与,从而为后续分割工作提供了一个较好的车牌图像环境。在字符分割模块中,改进传统的投影法,并提出计算每个字符的起始坐标的思想,得到坐标标记法。该方法有效克服噪声、字符粘连、断裂的影响,实现字符的完整正确分割。最后,在字符识别模块中,改进传统的模板匹配法,提出多方面提取字符的结构和统计特征的思想,得到基于结构特征和统计特征的模板匹配法。其主要思想是,先提取数字和字母的结构特征,进行粗分类,再提取汉字和剩下字符的统计特征:行扫和列扫的特征值、行和列投影值、水平和垂直宽度,结合二值化字符的多方面特征形成各自的模板,而且加入、提取了一部分扭曲、含许多噪声的特殊字符的特征,以扩大字符识别范围,然后,输入待识别字符进行模板匹配,得出正确的识别结果。为了验证文中算法的正确性和有效性,在车牌图像二值化、字符分割、识别模块中作了大量的编程实验工作,算法经VC++实现,实验表明,本文的改进算法效果较好,对车牌识别速度、准确率的提高作了有益的尝试,为提高车辆车牌识别技术的性能作了自己的努力。更多还原

【Abstract】 With the rising demand for transport, license plate recognition technology based on image processing is an important means of solving the problems of traffic violations, monitoring, fees, and so on. For image acquisition, preprocessing, character segmentation and character recognition, the article describes detailed theories and analyzes the merits and shortcomings of the traditional methods. At last, the author puts forward the improved algorithms combined the merits of the traditional algorithms, and experimental results are showed.Because the traditional Berson, Ostu ways are easy to be disturbed by the noise. The author improves the Ostu way and presents the way of dynamic threshold value based on Ostu. It overcomes the shortcomings of traditional binaryzation ways, and combines the merits of them. The new algorithm has used the different threshold values in different regions, so it can effectively suppress the disturbance of the complex illumination environment and greatly reduce noise participation, then it has provided a good car license image environment for the following segmentation work. In the charaters division module, the author improves the traditional projection way and proposes the thought of calculated the characters outset coordinate, then proposes the coordinate marking algorithm. This algorithm can effectively overcome the influence of noises, the character adhesion and break, and characters can be correctly divided. At last, in characters recognition module, the author improves the way of traditional template matched and presents the thought of extracted the characters structural and statistical natures, proposes the template matched way based on the structural features and the statistical natures. Firstly, the algorithm extracts the structural features of the numbers and the letters to carry on the thick classification. Secondly, it extracts the statistical natures of Chinese character and the rest characters, such as: the characteristic values of scanned in line and row, and line and row projection values, and the horizontal and vertical widths. The new way combines various characteristics of binaryzation character to compose the respective templates, moreover, it extracts some characteristics of special characters to expand character recognition scope. Finally, license plate characters are matched with templates to get correct recognition result.In order to confirm the accuracy and effectivenesss of the above algorithms, massive programming experimental work have been done in several parts of the car license image binaryzation, character segmentation and recognition. All of the algorithms are implemented in VC++, and experiments prove that the improved algorithms have a better effect. The author has made the beneficial attempt to improve character recognition speed and accuracy, and has made own effort to enhance the performance of the vehicles car license recognition technology.更多还原