基于机器视觉的压印字符识别系统研究

发布时间：2018-05-18 22:15

本文选题：机器视觉 + 压印字符　；参考：《广东工业大学》2017年硕士论文

【摘要】：轮胎压印字符是在工业中广泛被应用的一类标识字符,它是利用模具使物体表面产生凹凸形变压印而成。压印字符与其背景区域颜色相同,且为立体字符。随着现代制造业科技水平及生产效率要求的提高,对凹凸压印字符的自动化识别已成为工业产品生产管理智能化的一个必然要求。由于无色差这个特点使得工业相机采集的压印字符图像不易辨认,所以直接借用已有的光学字符识别研究成果较为困难。本文正是针对凹凸压印字符难以稳定识别这一问题,开展了以下的相关研究工作。设计了针对轮胎压印字符的图像采集硬件系统。本文首先简述了机器视觉系统的组成,分析了相关硬件的性能参数和选型方法,并根据本课题视觉系统的具体要求选取了合适硬件。分析了光照对压印字符图像的影响,并根据轮胎压印字符特点选择高角度环形LED光源前置照明的照明方案作为识别系统的最佳照明方案,为后续的算法处理的奠定了良好的基础。设计了针对轮胎字符信息环形排列结构的模板匹配定位方法及边缘检测分割方法。本文首先介绍了常用的字符定位算法,然后利用压印字符的特点和已知字符信息研究了基于归一化积相关(NCC)的压印字符定位问题,进而提出基于已知信息粗略定位,扇形扫描定位轮胎字符区域的方法,定位效果比较理想,能够准确地确定轮胎字符所在区域的边界。通过多种分割方法的实验效果对比,采用边缘检测对已定位并转换为水平排列的字符图像进行单个字符分割。采用了基于最小二乘支持向量机的字符识别方法,该方法很好地解决了压印字符在识别过程中出现的训练时间长,识别正确率不稳定的问题。从最终的软件实现效果来看,本文设计的压印字符识别系统的识别正确率及稳定性符合系统的设计要求,达到了很好的识别效果。
[Abstract]:Tyre embossing character is a kind of mark character widely used in industry. The embossing character is the same color as its background area and is a stereoscopic character. With the improvement of technology level and production efficiency of modern manufacturing industry, automatic recognition of embossed characters has become an inevitable requirement for intelligent production management of industrial products. Because of the characteristic of no color difference, it is difficult to identify the imprint character image collected by industrial camera, so it is difficult to directly borrow the existing research results of optical character recognition. Aiming at the problem that the embossing characters are difficult to recognize stably, the following research work is carried out in this paper. A hardware system of image acquisition for tire embossing characters is designed. In this paper, the composition of the machine vision system is briefly introduced, and the performance parameters and selection methods of the related hardware are analyzed, and the appropriate hardware is selected according to the specific requirements of the vision system in this paper. The influence of illumination on imprint character image is analyzed, and the lighting scheme of high angle ring LED light source preillumination is selected as the best lighting scheme of recognition system according to the character characteristics of tire embossing character. It lays a good foundation for the subsequent algorithm processing. A template matching location method and edge detection segmentation method are designed for the ring arrangement of tire character information. This paper first introduces common character localization algorithms, then studies the imprint character location problem based on normalized product correlation (NCC) using the characteristics of imprint characters and known character information, and then proposes rough location based on known information. The method of sector scanning and locating tire character region is ideal, and it can accurately determine the boundary of tire character region. By comparing the experimental results of many segmentation methods, the single character segmentation of the character image which has been located and converted to horizontal arrangement is carried out by edge detection. The method of character recognition based on least squares support vector machine (LS-SVM) is used to solve the problems of long training time and unstable recognition accuracy of imprint characters. From the final effect of software implementation, the recognition accuracy and stability of the imprint character recognition system designed in this paper accord with the design requirements of the system, and achieve a very good recognition effect.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TQ330.493

【参考文献】