【作者】 王舟；

【导师】 余英林； 张大鹏；

【作者基本信息】 华南理工大学 ， 通信与电子系统， 1998， 博士

【摘要】 远程相关性是自然界图像中普遍存在的一种特殊的信息冗余形式,人眼和人脑所组成的智能视觉系统能够利用远程相关信息实现图像表示、图像恢复与图像增强等功能。本论文旨在把远程相关性应用到解决实际图像处理问题上,研究了基于远程相关性的基本方法和具体实现算法。本论文的工作总结如下: Ⅰ.分形图像编码方面的分析和发展 1) 对分形图像编码技术提出了一些不同的新看法。我们认为,分形图像编码之所以取得成功,最重要的原因在于利用了自然界图像中的远程相关冗余性。 2) 提出余域中的分形方块编码方法。该方法既可提高编码效率,又可减少解码迭代次数。优化方法和快速解码结构的提出,使解码计算时间进一步减少。 3) 提出并讨论了部分分形映射的概念和性质,给出一类混合图像编码系统的通用理论和算法框架。这一框架可以派生出许多具体的混合图像编解码方案,我们还给出一个具体实例,显示了混合编码系统的特点和优越性。 Ⅱ.基于远程相关性的图像信息恢复 1) 首次提出了利用远程相关性进行图像信息恢复的思路和基本算法,并将其归纳为五个基本步骤,即提取、搜索、匹配、竞争和恢复。 2) 提出利用远程相关技术对图像通信中方块丢失进行差错补偿的方法。针对连续图像块丢失的的情况,给出了逐级迭代差错补偿算法,使在图像块丢失比例非常高的情况下,仍然能得到合理的恢复结果。 3) 针对图像采集与通信过程中经常遇到的脉冲噪声问题,提出基于远程相关性的脉冲噪声消除方法。对不同噪声形式和噪声比例下的测试结果显示,该方法在近年来许多其他学者提出的各种算法的基础上又有了一个明显的提高。 4) 提出了将模糊技术与远程相关技术相结合的脉冲噪声消除算法,使得对图像中不能准确估计是否是噪声的图像点也能够给出理想的处理效果。 5) 提出用基于远程相关性的方法来消除图像中的“方块效应”。对IJG的JPEG版本6.0解码图像的后处理实验显示,图像的主客观质量都有明显提高。更多还原

【Abstract】 Long range correlation is a special kind of information redundancy existing in natural images. The intelligent vision system composed of human eyes and brains can make use of it to implement the functions of image representation, image restoration and image enhancement. This dissertation aims at using it to solve some real problems in image processing. A general method of image processing using long-range correlation is proposed and some implementation algorithms are provided. The work of this dissertation can be summarized as follows:I. Analysis and Progress on Fractal Image Coding1) The fractal image coding technique is discussed from a new point of view. We think its success mainly relies in the use of long range correlation in natural images.2) The Fractal Image Coding in Residue Domain (FBCRD) method is presented. It can improve the coding efficiency while reduce the decoding iteration steps. An optimization method and a fast decoding structure are also introduced, that can further reduce the decoding time.3) The concept and the properties of partial fractal mapping are presented and discussed, which leads to a general theory and framework of a class of hybrid image coding systems. Many concrete image coding/decoding systems can be derived from such a framework. Our example shows the advantageous of the hybrid image coding method.II. Image Information Restoration Based on Long Range Correlation1) For the first time, we propose the idea and the basic algorithm for image restoration using long range correlation, which can be summarized into five basic steps: Extracting, Searching. Matching, Competing and Recovering.2) An error concealment method for information loss in block-based image coding systems is presented. For the cases of contiguous block losses, we provide a progressive error concealment method, so that even when the images are very highly corrupted, we still can get reasonable restoration results.3) Impulse noises are often encountered in image acquisition and communication. We present an impulse noise removal method using long range correlation. The tests on different types of noise and different noise ratios show that our method outperforms many recent published approaches provided by other researchers.4) An impulse noise removal algorithm combing fuzzy technique with long range correlation method is introduced. By using fuzzy rules, some ambiguous pixels in the image can also be processed very well.5) The long range correlation method is used to reduce "Blocking Effect" in images. Experiments show that our method achieves enhanced quality for the IJG version 6.0 of JPEG decoded images both subjectively and objectively.更多还原