【作者】 宋晓东；

【导师】 阙大顺；

【作者基本信息】 武汉理工大学 ， 信号与信息处理， 2003， 硕士

【摘要】 数字图像处理是对在空间上离散、幅度上量化分层的图像进行某些特定数理模式加工处理的过程。而图像分割是把图像分成各具特性的区域并提出感兴趣目标的技术和过程，图像分割在图像工程中占有重要的位置。 本文以湖北省科技厅重点科技发展计划项目资助课题——“智能运输系统的视频信号采集及识别算法研究”为背景，以课题中数字图像处理部分主要环节之一——车牌图像的分割问题为主要研究目标，在参阅大量文献资料的基础上，对数字图像分割方法进行了学习和研究，并结合特定的数学理论，如数学形态学、小波分析等，对车牌图像分割方法进行了着重地探讨与实现。图像处理过程在去噪的同时，尽量保留边缘特征是图像分割首要顾及的问题。本文采用小波多分辨分析和阈值去噪的处理方法，取得了较好的效果。为了将车牌区域从复杂背景中分割出来，需要在一定程度模仿人类视觉系统在分割与识别图像时具有的对多种信息综合运用的能力。小波多分辨率分析方法的主要思路是在不同的分辨率层次上对图像进行分割，粗分割结果对精分割具有一定的指导作用。数学形态学是一种非线性滤波方法，其特点是能将复杂的形状进行分解，并将有意义的形状分量从无用的信息中提取出来。利用数学形态学这一特性来对小波分解出来的图像的细节分量进行处理，在计算开销和定位精度这一对矛盾上可达到较好的平衡。针对智能运输系统视频信号采集识别中车牌图像的分割问题，本文提出了结合运用小波多分辨分析和数学形态学来处理图像分割问题的实验方案。实验结果证明该方案是可行和有效的，具有一定的实用价值。 本文主要分六部分。第一部分介绍图像处理的分类、特点和相关技术，图像分割在图像工程中的位置；第二部分以图像变换的有关数学理论为基础分析了数字图像处理的特点与优势；第三部分论述了典型的数字图像预处理的方法；第四部分阐述了各种图像分割的主要算法原理，并对它们进行了比较、归纳、分类，着重研究了小波分析和数学形态学等在图像分割中的应用；第五部分简要介绍了数字图像分割的程序设计常规工具，提出了综合运用多种图像信息的图像分割实验方案；最后在第六部分对所做的主要工作进行了总结与展望。更多还原

【Abstract】 Digital image processing is a mathematical procedure which calculated by computer. Image segmentation is a technology that divide image into characteristic areas and put forward the interested target. Image segmentation is a key step from digital image processing and analysis, so it has an important position in the technology of the image processing.The background of this paper is "the research on the recognition algorithm of video signal sampling in ITS", which is a scientific & technical development plan of HuBei province. Image segmentation mainly has following several treatments. Firstly, proceeding from the whole graph. Judge each pixel’s classification based on collected groups of attribute spaces, finally get the areas. Secondly, setting out from pixels. Form an area by collecting near pixels with unanimous attribute. Thirdly, division-collection method. Divide the image into several regular blocks, don’t separate the blocks without consistent attribute and unite the blocks with the same attribute, until forming a district map. The ability that man can distinguish different goals from the complicated scenery quickly, at least partly benefit from many kinds of information in the image, such as the grey level, border, texture, etc. It illumine people to create methods on how to use mis information to segmentation. Edge information is the most important high frequency information of an image. Therefore we should try to maintain more edge information in the process of denoising. We present a new image denoising and segmentation method based on wavelet analysis and mathematical morphology. In some extent, It simulate the ability of human vision system by this means. The experiment results show that the segmentation method is effective, it has some practicality.This thesis can divide into six parts. The first part is to introduce the classification, characteristic, and relevant technology of image processing, and segmentation’s position in the image engineering. The second part will analyse the characteristic and advantage of digital image transformation, basing on the relevant mathematics theories. The third part is to introduce several typical digital pretreatment of the image. The fourth part will put an emphasis on various kinds of main algorithm of the image segmentation, followed by the comparison, the conclusion, and classification. The fifth part will introduce some general tools of program designs for image processing, and put forward a segmentation experiment method on vehicle license plates. The last part is a summary and prospect.更多还原