【作者】 李国平；

【导师】 路长厚；

【作者基本信息】 山东大学 ， 机械制造及其自动化， 2007， 博士

【摘要】 标牌是生产厂家出售产品时的基本标志，承载着可以追溯生产来源的重要信息。这些信息的自动识别对产品的生产、运输、销售全过程信息化管理具有重要意义。标牌的内容可分为固定不变内容和动态数据两部分。前者一般事先印制在标牌上，而后者则是利用模具压制而成，表现为凸出或凹下标牌表面的凹凸字符，以具备永久性和防伪功能。针对标牌压印结束之后的合格检测以及生产厂家对产品信息的录入都靠人工检查和识别来完成的这一弊端，提出实现标牌凹凸字符自动识别的研究课题。本文重点从提高获取原始图像的质量着手，提出了基于莫尔技术的标牌凹凸字符图像获取与识别策略。针对标牌凹凸字符图像的特点，结合光学字符识别技术，研究了标牌凹凸字符图像的获取技术、字符图像的预处理方法、字符的特征提取和字符的分类识别方法。对不同情况下形成的莫尔条纹具有不同特征的复杂情况进行了研究，分析了常用的正弦光栅和矩形光栅所形成的莫尔条纹及其影响因素。这对设计基于莫尔技术的光学系统具有重要的理论价值和指导作用。在研究了莫尔条纹形成机理的基础上，提出利用阴影莫尔法获取字符图像的新方法。给出基于该方法的字符图像获取的理论分析；设计出光栅传感器及其实验装置；分析了实验结果。和传统方法获取的图像对比，证明可以明显改善获取的字符图像质量。在图像处理中，阈值和图像边缘都是图像二值化的重要信息。提出将最佳全局阈值二值化与局部阈值二值化在边缘提取的基础上融合起来对图像进行二值化处理的方法。给出了基于图像边缘提取的自适应的二值化方法基本原理及算法的实现。圆周投影变换是一种具有尺度、旋转、平移不变性的特征提取方法。它以图像质心为极点，将笛卡尔坐标变换为极坐标，然后计算同一半径圆周上的像素值之和。把二维字符图像，变换为一维圆周投影曲线。变换时利用伪数字圆查表法快速地计算出字符图像的圆周投影。对获得的圆周投影曲线，利用小波的多分辨率分析特性进行快速小波分解，分别得到在尺度空间内的圆周投影的大致概貌信号和在小波空间内的细节部分信号。以大致概貌信号作为字符识别特征，再利用离散余弦变换进一步对小波分解后的大致概貌信号进行特征降维，以实现快速的字符识别。在圆周投影变换特征提取的基础上，将最小二乘支持向量机引入到小字符集压印凹凸字符识别中。针对标牌上的压印凹凸字符的数字集进行了识别实验，同时与模板匹配、神经网络等其它分类方法进行了比较。实验结果表明此方法的识别率较高，在小字符集识别中具有较强的实用性。本课题为教育部博士点基金资助项目课题。更多还原

【Abstract】 The label is the basic signs that carry amounts of important information of product. The recognition of this information is of great significance for information management during the production, transportation and sales.The contents on the label are classified as the fixed and dynamic data. The former are usually printed on the label surface in advance, and the latter which are protuberant characters on the label are pressed by a protuberant mould, so that the protuberant characters are permanent and have a function of anti-counterfeiting.As to the abuse of checking and recognition completed by man after the character pressed and as to the information recorded of product by producer, subject of automatic recognition protuberant character was proposed. Stressing to improve original image quality, we presented a way of image acquisition based on Moire technology. Combining the characteristics of character image with the optical character recognition, we studied image acquisition, pre-processing of character image, feature extraction and classification recognition methods of character.We researched the features of Moire fringes formed in different cases and the influencing factors of Moire fringes formed by sine gratings and rectangle gratings. The research is of great theoretical and practical value to design of an optical system based on Moire technology.After analyzing Moire fringes, we gave a theoretical analysis of image acquisition based on shadow Moire, designed grating sensors and experimental devices, and analyzed the experimental results. Contrasted with the conventional acquired images, the images obtained by shadow Moire have higher quality.As the threshold and the image edge are the key information during the image processing, we proposed an adaptive binarized method through combining the optimal global threshold with the local threshold based on edge detection. The basic principle and the algorithm of the adaptive method are given.The circular projection transformation is a method of feature extraction that has scale, rotation, and translation invariance. It takes centroid as apex transform Descartes coordinates into the polar coordinate, and calculate the sum of pixel value on the same radius. It changes two-dimensional character image into one-dimensional circular projection curve. A quick calculation of circular projection curve is put forward through indexing the table of pseudo digit circle.The curve of circular projection can be decomposed by use of the multi-resolution analysis of wavelet. The approximate signals as the feature of character recognition are extracted in the scale space, and the detailed signals are obtained in the wavelet space. Finally, the dimension of approximate signal is further reduced through the discrete cosine transformation to realize quick character recognition.Being based on the feature extraction of the circular projection, we simulated the number set of the pressed characters on the label and compared with other classifiers, such as template matching, neural networks. The experimental results show that this method has a high recognition rate and practicability for the recognition of small character sets.This research work was supported by a grant of Programs Foundation of Ministry of Education of China.更多还原