【作者】 徐铭杰；

【导师】 姚明海；

【作者基本信息】 浙江工业大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 图像处理技术的发展自上世纪50年代,在20世纪60年代迅速发展并成为了一门新学科。到今天,数字图像处理在科学研究、工农业生产、军事技术、卫生医疗、政府部门等许多领域中,发挥着越来越重要的作用。传统的图像识别在PC机上实现,识别算法多基于神经网络和模板匹配等。此类方法需要在样本数目足够多的前提下才能获得较好结果。系统在便携移动性方面具有很大的局限性。随着时代的发展,人们对于系统的实时处理要求越来越高,字符识别系统在保证系统运算速度、识别精度高的前提下,逐渐向集成化、便携性方向发展。本文基于这样的背景,提出了在嵌入式平台上通过支持向量机算法实现字符图像识别。(1)图像预处理。对字符定位、倾斜校正、字符分割、字符归一化等方面进行了研究,比较了各算法的优缺点,对于关键的字符定位和字符分割两方面提出了自己解决方法。并且具有一定的通用性。(2)字符识别。深入学习支持向量机算法,在提取字符图像的轮廓线数特征的基础上,利用傅立叶描绘子提取字符外轮廓线特征,将两者相结合作为SVM的向量进行训练、识别,最终实现字符识别。(3)系统总体方案的提出。设计了字符识别系统的软件和硬件的总体框架。对嵌入式Linux系统开发进行了深入的研究和探讨。(4)在嵌入式Linux平台上实现了以下功能:图像采集模块的开发过程中,对Linux下的USB摄像头驱动和V4L标准进行了深入的研究,并在此基础上实现了图像采集存储。移植并实现了PPP协议,使得终端能够无线上网。编写了socket服务器端和客户端程序,使得两者之间能进行数据交换,传输识别的字符。在此功能基础上,结合本文实现的字符识别功能,构成了一个完整的字符识别系统。更多还原

【Abstract】 Image processing technology began in the 1950s, it became a new subject in the 1960s quickly. With the development of the society, the image processing technology has played a more and more important role in our life, such as in science research, military technology, medicial and health service, industrial production and government department. However, traditional statistic mode for recognition realized on the base of PC, such as template matching and neural network. For this way, it could get good effectiveness only under the condition of adequate samples. Its system has great limitations on the portable and moving hands. With the development of the society, people have the increasing need of real time data processing. On based of this present situation, this paper put forward a way to realize character image recognition using support vector machine algorithm on the platform of embedded Linux operation system.(1) For the pretreatment of the image, on the hands of image capture, character location, character segmentation, author compares the good points and bad points of all kinds of algorithm. Meanwhile, for the important two hands (character location, character segmentation), author also puts forward his new thoughts and solvements.(2) Character recognition. After deeply learning the Support Vector Machine algorithm, on the base of Fourier descriptors and support vector machine is presented. Character features are extracted from the outer contour by Fourier descriptors, which constructs the input vector of the multi-classes support vector machine. The test results show the effectiveness of the algorithm even with limited samples.(3) The overall design of the system. This paper firstly summarizes the development course and current situation of the research for character recognition system. Then it designs the framework of the software and hardware of the character recognition system. It also has a serious study and discussion for the development of embedded Linux system.(4) On the platform of Linux operation system, it will realize the following functions. Firstly, it realizes the video-capturing module based on the research of the V4L standard and the USB camera driver under Linux. Second, it realizes PPP protocol on the embedded Linux system, which endows the intelligent terminal with the functions of wireless internet connection and transfer a large amount of data.更多还原