【作者】 刘大宇；

【导师】 龙建忠；

【作者基本信息】 四川大学 ， 模式识别与智能系统， 2003， 硕士

【摘要】 汉字识别是模式识别中一个重要的研究方向，它在办公自动化、高速信息处理、机器视觉等多方面具有重要的理论意义和实用价值。在智能交通系统中，应用汉字识别技术，对汽车牌照进行自动识别，可以实现对车辆的监控和智能管理，具有广阔的市场前景。 本文首先介绍了针对车牌字符图像的预处理方法，包括规格化和二值化算法，根据实验效果对各种方法进行了比较，选取了适用性较强的双线性插值法和最大方差比分割法，应用到汉字识别的预处理过程中。 作者对车牌汉字识别的特征提取方法进行了研究，首先比较了多种基于统计模式识别的特征提取匹配算法，包括外围面积特征，网格特征和用于区分相似汉字的微结构投影特征。为了实现对模糊汉字灰度图的识别，提出了基于二值图形变动分析的模糊模板匹配算法及其改进方案。经初步测试，该方法取得了良好的识别效果。本章还对贝叶斯分类器，子空间模式识别和人工神经网络在字符识别中的应用进行了总结，可作为进一步研究的基础。 在车牌字符识别中引入了误识模型和多分类器集成技术。在实验分析的基础上，确定了影响系统识别性能的两个主要参数：首选距离及其与二选的距离差，并将其作为多分类器集成的判决依据。在实际系统中，我们使用了基于串行方式的两级识别器融合，车牌识别正确率达到了用户的要求，系统运行状况良好。 文章最后介绍了汉字识别的最新进展，并对其研究方向做了展望。更多还原

【Abstract】 Chinese character recognition is an important research direction in the field of pattern recognition. It is of great value both in theory and practice and has applied in many circumstances, including official automation, high speed information processing, machine vision, etc. In the Intelligent Transportation System (ITS), automatic recognition of license plate by using the technique of character recognition can give facilities for vehicles’ surveillance and management. So there’s much prospect in the application.In chapter 2, some methods of pretreatment are introduced in detail, including algorithms of normalization and binarization for the image of license plate character. By comparing these methods, we select the bilinear interpolation and thresholding segmentation of maximum variance that have good performance, and all of these have applied in the course of preprocessing of Chinese character images.A survey of character recognition methods is presented in chapter 3. Comparison of some extracted feature matching algorithms based on statistical pattern recognition is conducted. These features are profile, mesh and projection of micro structure for distinguishing similar characters. For the sake of identifying some blurred images, a fuzzy template matching algorithm based on analysis of binary images’ variability and its improved scheme is presented. Theoretical analyses and experimental results demonstrate that this method is very effective. Also, Bayesian classifier, subspace method and ANNare summarized in this chapter. They can be used for the next research.The techniques of mis-recognition model and multiple classifier combination are proposed and used in the system. On the basis of experimental analysis, two main parameters that have great effects on the recognition correctness of the system are ascertained: the first candidate’s distance and the distance difference between the first and the second candidate. By using these parameters, multiple classifier combination can be realized and represent validity. In the actual LPR system, we applied two classifier combination based on serial method and its performance has meet the needs of customs.The latest progress of Chinese character recognition and the research prospect is presented in the last of this paper.更多还原