【作者】 田建国；

【导师】 龚劬；

【作者基本信息】 重庆大学 ， 计算数学， 2008， 硕士

【摘要】 图像分割是计算机图像识别与理解中一个十分活跃的研究领域,是计算机图像理解方法实现的基础。主动轮廓模型是图像分割的一种重要的方法,能够准确地实现图像中目标对象的分割,因而成为众多研究者研究的热点问题。小波分析作为信号和图像处理的一种强有力的工具,其理论框架已基本形成。小波分析具有其他许多信号分析方法(比如Fourier变换、Gabor变换等)所不具备的一些优良特性。小波分析的多分辨特性能够提供图像在不同分辨率下不同通道上的图像特征。本文在推广梯度矢量流的基础上,提出了基于二进小波变换的改进外力场的计算方法。通过综合利用图像小波分解的高频信息,迭代计算获得其外力场,使Snake在图像的多尺度空间中搜索目标轮廓。针对Snake模型对初始轮廓的依赖性问题,给出了在一维最大熵阈值分割后的图像上获取初始轮廓点的方法,减少了人工的干预,加快了Snake的收敛速度和效果。实验结果表明该模型能有效地排除噪声的干扰,搜索凹陷轮廓,而且对脆弱轮廓有很好的逼近能力。随着信息技术和智能技术的发展,交通管理系统的信息化、智能化已成为发展的趋势。车牌识别系统(LPR)是智能交通系统(ITS)的核心组成部分,在现代交通收费管理系统中发挥着举足轻重的作用。近年来,对车牌识别系统中关键技术的研究已经成为智能交通领域的一个热点问题。本文对车牌识别系统中的车牌定位,车牌图像倾斜校正以及字符分割等几大关键部分进行了比较深入,全面的论述。车牌定位是车牌识别的基础,针对复杂背景,复杂环境车牌定位难的问题,综合利用了车牌区域水平灰度跳变特征,自适应地实现图像的二值化;利用数学形态学对图像进行一系列形态运算,消除了大量无用信息和噪声;根据区域分析提取出候选车牌区域;最后结合先验知识,将车牌准确地定位出来。然后通过Harris算法得到车牌的角点信息,并在此基础上,根据Hough变换得到车牌的倾斜角度,实现车牌的倾斜校正,接着对车牌进行去除边框的处理,最后利用投影法实现车牌字符的切分。通过对采集于各种真实环境的图像进行实验,结果表明,本文所采用的方法能达到较好的车牌定位和字符分割效果,具有一定的鲁棒性和实时性。更多还原

【Abstract】 Image segmentation is thoroughly active research domain in the field of computer image recognition and comprehension. It’s the basic of image comprehension. Activate contour model is a important method of image segmentation, and can segment the object accurately, so it has already become an important research field. Wavelet transform is a powerful tool in the signal and image processing, and its fundamental theory has been formed. Wavelet transform has many properties which other signal analysis techniques (for example, Fourier analysis, Gabor Transform, etc) do not have. Because of the multiresolution property, the wavelet transform provides image’s characters on different channel and different scale.This paper generalizes the gradient vector flow and then introduces an improved external force based on dyadic wavelet transform. The detail coefficients of image wavelet transform are used to derive the new external force, so the snake can track the object contours through the multi-scale space of an image. In additional, aimed at the snake’s dependence on initial contour points, one-dimensional maximum-entropy threshold algorithm is adopted to segment the image firstly and then obtain the initial contour points, which reduce the manual operation as well as improve the convergence of snake speed. Experimental results proved that the new model can greatly get rid of influence of noise and search concave contours. Furthermore, it is efficiently convergence to weak boundary.With the development of information technology and intelligence technology, the informatization and intelligentizing of traffic management is the trend. License plate recognition system (LPR) is the core of intelligent traffic system (ITS). It is very important in modern traffic management systems. In recent years, the study on the critical techniques for LPR has already become an important research field of scientific circles.In this paper we discuss the main parts of LPR: license plate location, slant correction and character segmentation, and make a deep research on some important and key technology. Automatic vehicle location license plate is the basic of LPR. It’s difficult for realizing this technology in complex background and complex environment. According to a great deal of horizontal texture of vehicle image, some preprocesses such as adaptive binary conversion, a series of morphological operations which can greatly decrease the noise and useless information in vehicle image; then the method of region analysis can select some candidate regions; lastly using prior knowledge locate the license plate region accurately. Next step is the characters segmentation. We can get the corners of the license image using the Harris algorithm. At the same time, the slant angle of license plate is necessary to use the method combined with Hough transform and corner information. Then we excise the border of the characters and segment the characters. Based on the experiments dealing with images taken under various real world conditions, the results prove that this proposed method can relatively locate license plate and segment characters. It shows that the performance of the system is promising, robust and timely.更多还原