【作者】 王宏；

【导师】 关宇东；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2008， 硕士

【摘要】 随着计算机技术的飞速发展数字图像处理技术也得到了快速发展,人们越来越追求更高的图像效果。然而在成像过程中受到许多不可避免的因素影响,图像出现了降质现象。图像阴影就是图像一个典型的降质现象,它的存在直接影响了图像匹配的精度、模式识别的准确度以及目标提取的自动化程度,使智能视频监控系统中后续的目标跟踪、识别等工作的出错率大大增加。因此,阴影消除对于提升智能视频监控系统的工作性能很有必要。本文针对大多数分割算法输出目标携带阴影的问题,分析了阴影形成的原理和光学属性,构造阴影形成的光照模型。从阴影相关理论出发利用数字图像处理的相关理论和信息论的有关知识,设计了三种有效的去除阴影的算法。三种算法各有其特点,各有其适合处理的图像,结合起来实现自适应去除阴影的算法。论文在绪论中阐述了课题的背景、意义、来源以及国内外现状,明确了研究的主要目的。第一章中研究了阴影的相关理论,其是阴影去除的重要依据,根据其做了如下工作。首先根据图像处理的直方图原理和聚类技术,提出了基于直方图和聚类技术去除阴影的算法,此方法实现了多个目标的阴影去除。利用直方图的相关知识能够确定目标的个数,目标的大概宽度、边界,阴影的方向,阴影的大致区域;结合灰度图像的聚类分析得到精确的阴影区域,实现了单个或者多个目标阴影的去除。其次利用阴影颜色和纹理不变的光学属性,建立了基于色度畸变和局部交叉熵去除阴影的算法。在RGB彩色空间运用颜色矢量计算阴影与对应背景颜色变化的角度,即色度畸变角,通过阈值确定阴影区域。利用信息论中的交叉熵进一步区分目标与阴影。色度畸变和交叉熵的结合,能够有效地去除分割目标的阴影。再次根据阴影区域灰度连续、平坦的特点,设计了基于多梯度分析和线扫描去除阴影的算法。梯度体现了灰度的变化,灰度变化越剧烈梯度越大,线扫描可以检测出一个像素宽度的线。梯度可以检测出灰度连续的阴影区域,线扫面可以去除连续区域面积小的非阴影区域,两者结合进而实现了去除阴影的目的。最后根据三种算法的原理和应用范围,设计了判定程序实现了图像自适应的选择去影算法,减少人为判断,提高了算法的鲁棒性。引入了一种量化评估方法,对各算法进行客观评定。更多还原

【Abstract】 With the rapid development of computer, the technology of digital image processing is developing fast. And people have a growing pursuit of higher image processing effect. However, the effect of digital images is degraded more or less by many inescapability factors during the imaging process. Image shadows is typical phenomenon of the drop of image quality. It will directly affect the precision of image matching, the accuracy of pattern recognition and the automation of objects extraction, which will cause more errors in tracking and recognition of object at post-treatment of intelligent video monitoring. Therefore, it is very important to wipe off shadows to improve work quality of intelligent video monitoring.Against the problem that many methods of image segmentation often bring shadow, this thesis analyzed the principle and optical properties of shadow whose illumination model is also made. Based on the theories of shadow, three effectively algorithms to remove shadow were designed by theories of digital image processing and information. Because the algorithms have different features and one algorithm is adapted to a kind of image, automation algorithm of removement shadows was designed by the algorithms.This thesis explains background, significance, source and present situations of shadow removing in domestic and foreign to clear the purpose of this research. And Chapter One studied the theory of shadow which is an important basis for the removal of the shadow, according to it, do the jobs as follows.Firstly, based on histogram and clustering techniques, a shadow eliminating method which depended on those techniques was presented. It can remove multi-object shadows. The quantity, probably width, borderline of objects, direction and approximately area of shadows could be confirmed by theory of histogram. And this method combined with clustering techniques of gray image, this method could obtain exact area of shadows in order to eliminate shadow of one-object or multi-objects.Secondly, by the use of the optical characteristic of shadow (such as color, veins), a shadow removing algorithm based on chrominance distortion and local Cross Entropy was set up. In the RGB space, we use color vector to calculate the angle between shadow and corresponding background. And based on the limit we can confirm the area of shadows. By the use of Kullback-Leibler divergence in information theory, we can gain further distinction between objects and shadows. The combine of chrominance distortion and Cross Entropy can eliminate shadow of segmentation object in the effect.Thirdly, the shadow removing algorithms based on multi-gradient analysis and line scan were designed according to the continuum and smooth of shadow area gray. Gradient represented the variety of gray and gradient is also much bigger at gray acute variety area. Line scan can search the lines of the image element point width. Gradient can find the continuum area of shadow and line scan can remove little area of non- area of shadow so as to realize purpose of eliminating shadow.Finally, based on the theory and application of three algorithms, the judgment process was designed and the choice of adaptive algorithms of shadows removing was also realized. It reduced human judgments and improves the universality of algorithm. A quantitative method is introduced to evaluate the algorithms on a benchmark.更多还原