【作者】 余庆军；

【导师】 谢胜利；

【作者基本信息】 华南理工大学 ， 通信与信息系统， 2004， 博士

【摘要】 数字图像处理是一门涉及光学、数学、物理学、生理学、计算机科学、通信理论、信号分析、控制理论以及系统工程等方面知识的综合性的学科。它通常根据不同的应用、不同的要求而采用不同的处理方法。其目的就是对图像信息进行加工处理,以满足人的视觉心理或实际应用的要求。自20世纪60年代诞生以来,数字图像处理技术一直受到了广泛、深入的关注,并逐渐渗透到社会的各个领域。当前,随着Internet、移动通信以及多媒体技术的迅猛发展,无线多媒体的需求日益增长,数字图像处理技术也就不可避免地成为无线多媒体技术中一个十分活跃的研究领域。本论文主要围绕模糊技术、计算机视觉以及心理物理学理论等智能技术在图像处理中的应用展开研究,同时对所提出的智能图像处理技术如何应用于无线多媒体业务进行了深入的探讨。本论文首先研究和讨论了如何提高图像质量,使模糊图像变清晰的图像预处理技术。对于现存的各向异性扩散图像平滑技术,本论文研究分析得出,由于其扩散系数或者非线性扩散方程式严重地依赖于图像梯度,导致它们通常不能很好地处理受噪声污染的图像,甚至是不稳定的。而且,所有这些算法几乎都没有考虑到人类视觉系统的影响。为了克服这些问题,本论文在人类视觉系统的基础上提出了三个改进的各向异性扩散图像平滑方法。方法一充分利用了人类视觉噪声可见度函数所提供的有关信息;方法二改进了通常的根均方对比度测量模型,并利用了它对噪声抑制的良好作用;方法三则充分利用了人类视觉的选择性。对于基于模糊逻辑的图像滤波技术,本论文在最小二乘法以及考虑相邻像素相互兼容度的基础上,提出了一些新的模糊加权平均滤波器改进了Choi等的算法。此外,本论文还在视觉心理与心理物理学的基础上,结合模糊逻辑的方法,提出了一个新的边缘敏感图像去噪算法。实验结果表明,在图像滤除各种不同类型的噪声的过程中,所有这些方法在噪声消除和边缘保留方面均能获得良好的性能。其次,本论文研究和讨论了图像质量评估方法。本论文假设图像由平滑区域、边缘区域以及纹理区域等组成,而且不同的区域应采用不同的方法进行质量评估。在此基础上,本论文给出了三个分别适用于不同区域的图像质量评估因子,并提出了一个灵活的模糊图像质量评估方法。再次,本论文还详细地研究和讨论了便于图像传输和存储的有关图像变换和颜色量化方面的技术。考虑到视觉感知与心理物理学对于图像中感兴趣区域的影响,本论文提出了一个新的模糊图像插值算法以用于图像的缩放;此外,更多还原

【Abstract】 Digital image processing is a multidisciplinary field which falldiverse aspects of optics, mathmatics, physics, physiology, computerscience, communication theory, signal analysis, control theory andsystem engineering, etc. Its aim is to satisfy the human visualperception or the requirements of practical application. Digital imageprocessing has been attracting much extensive and in-depth attention,and has been gradually applied to every field of society with varyingdegrees of success. Currently, the prompt development of Internet,wireless communication and multimedia technology brings on theincreasing demands of wireless multimedia, which causes digital imageprocessing unavoidably becomes one of the most active research field inthe wireless multimedia techniques. We mainly study how to applyintelligent techniques in this dissertation, including fuzzy logic,computer vision and psychophysics theory, to image processing, anddeeply discusses how to apply the proposed intelligent image processingtechniques to wireless multimedia service.Firstly, we study and discuss some image preprocessing techniquesin this dissertation, which are used to improve image quality throughimage filtering. By studying and analyzing the existing anisotropicdiffusion image smoothing techniques, we deduce that some of them areusually not able to process noisy images well, and even not stable.Moreover, all of them hardly consider the influence of the human visualsystem. To overcome these problems, we develop three modifiedanisotropic diffusion algorithms for image smoothing on the basis of thehuman visual system in this dissertation. The first algorithm makes useof the information offered by the visibility function in human vision;the second one improves the general root mean square contrast measuredmodel and utilizes its good suppression effect to noise;and the thirdone utilizes the selectivity of human vision. As for the fuzzy imagefiltering techniques, a fuzzy weighted average filter was developed byconsidering the degrees of compatibility between the neighboring pixelson the basis of the least square method and the algorithm proposed byChoi in this dissertation. Futhermore, we develop a new edge-sensitivityimage denoise algorithm through integrating visual perception andpsychophysics with fuzzy logic in this dissertation. The experimentresults show that all these methods are able to achieve a good effectin the respect of noise removal and edge preservation during the processof image preprocessing.Secondly, we study and discuss the issues of image quality assessmentin this dissertation. By assuming that an image is composed of edge,texture and smooth regions, and a different quality measure should beapplied to its corresponding region, we present three image qualityassessment factors, and develop a flexible fuzzy logic-based imagequality assessment approach in this dissertation.Thirdly, we study and discuss the techniques detailedly that are fitfor image transmission and storage in this dissertation, such as imagetransform and color quantization techniques. By considering theinfluence of visual perception and psychophysics to the interestingregions in the images, we develop a new fuzzy image interpolationalgorithm to resize the image in this dissertation. In addition, wepresent a fast fuzzy cluster-based color reduction algorithm in thisdissertation to compress image by considering the visual redundanc andthe good characteristics of data link.Lastly, we discuss how to apply the above-mentioned intelligentimage processing techniques to the wireless multimedia service in thisdissertation. To adapt to the development of multimedia messagingservice, this dissertation develops a image warping-based morphanimation generating mechanism, which can not only generate novel andhistrionic GIF animation in real time, but also make it easier thattransmits rich and colorful image files in the wireless network byrealizing adaptive image processing according to the differentcharacteristics of different mobile terminals.更多还原