1、 图像稀疏分解论文：数字图像差分进化稀疏分解及压缩【中文摘要】随着社会信息化程度的不断提高,图像信息随处可见,而图像压缩作为图像处理中的一个重要环节,也被广泛应用于现代科学技术的多个领域。目前,人们已经提出了多种图像压缩方法,并形成了一系列图像压缩标准。这些压缩标准大都是基于图像正交变换的,在高比特率压缩时可以取得良好的效果,但在低比特率条件下的解码图像质量却不尽如人意,满足不了人们的图像压缩需求。比如：JPEG 压缩标准一般在 0.2bpp 以下便会出现方块效应,而JPEG2000 压缩标准则在低比特率条件下会出现蚊式噪声,因此需要发展一种在低比特率情况下有效的图像压缩方法。近年来兴起的稀疏
2、分解是一种非正交的分解,将图像分解在过完备原子库上从而得到图像的稀疏表示,该表示结果非常简洁,也符合人眼的视觉特性。由于图像稀疏分解的优良特性,使其成为解决低比特率下图像压缩问题的新途径。本文则主要围绕利用稀疏分解实现图像压缩这一问题展开研究,首先针对图像稀疏分解中计算量大的问题引入差分进化算法,并对这种算法进行深入研究,从而得到图像的稀疏表示,在此基础上研究高效的编码方法。本论文的主要工作及研究成果如下：(1)介绍了图像稀疏分解的核心思想以及图像的稀疏表示,并就图像稀疏分解过程中计算量大的问题引入差分进化算法,给出基于差分进化算法的图像稀疏分解流程和实现过程。(2)为了在图像稀疏分解之后得到
3、高质量的原子和投影分量信息,进一步研究差分进化算法,给出一种基于种群多样性的差分进化算法,并将其应用到图像稀疏分解。首先分析了现有的五种不同差分策略的性能特点,通过稀疏分解重建图像质量的对比选取最佳策略；其次充分考虑了寻优过程中随着迭代次数的增加,种群的多样性不断下降的问题,以种群多样性的度量值代替固定的迭代次数作为终止条件进行寻优。实验结果表明,与文献中基于粒子群算法的图像稀疏分解方法及传统差分进化算法相比,同等条件下该算法可以有效的提高重建图像的峰值信噪比,能更准确、有效的得到图像的稀疏表示。(3)根据图像稀疏表示的六个分量的分布规律,分析了传统的排序差分编码的不足,给出一种变码长编码的压
4、缩算法。该算法首先对分解结果数据中的投影分量进行排序差分处理,接着再对信息量比较大的差分分量、两个平移分量和一个旋转分量按照原子个数均各自分配一个码长,而对于分布范围比较小的两个尺度分量采用定长编码。实验结果表明与文献中排序差分压缩算法比较,同等条件下该算法在压缩重建图像峰值信噪比相同的情况下可以获得更高的压缩比,有效的减少了编码冗余,提高了压缩效率。(4)为了进一步提高图像的压缩比,根据图像稀疏表示之后的参数分量的特点,将行程编码的思想应用到基于稀疏分解的图像压缩,形成一种采用行程编码的图像压缩算法。实验结果表明,与变码长算法及文献中排序差分压缩算法相比,该算法同条件下可获得更高的压缩比,能
5、有效地提高图像的压缩效率。【英文摘要】With the improvement of information society, we can see the image information everywhere. The image compression, as an important part of image processing, is also widely used in modern science and technology fields. At present, people have made a variety of image compression metho
6、ds, which formed a series of image compression standard. These standard are mostly based on image orthogonal transformation, which can achieve good results in the high bit rate compression, but the decoding image is not satisfactory under the conditions of low bit rate, and it can not meet peoples d
7、emand for image compression. For example:for the JPEG compression standard, the following box will appear under the 0.2-bit, while for the JPEG2000 compression standard, mosquito-like noise will appear in conditions of low bit rate. Therefore it is necessary to develop an effective image compression
8、 method of low bit rate.In the recent years, sparse decomposition becomes very popular in the study of image processing. It is one kind of non-orthogonal decomposition, which decompose the image on the over-complete dictionary so as to get the image of the sparse representation. The decomposition re
9、sult is very simple and consistent with human visual characteristics. It has become a new way of solving the image compression in low bit rate because that it can transform an image into a spare formation,.This paper mainly focuses on image compression based on the sparse decomposition. First to the
10、 large computation of image sparse decomposition, differential evolution algorithm is used; secondly this article has in-depth research to get the image of the sparse representation. On the basis of this, this paper researches some efficient coding methods. The main work and research results are as
11、follows:1. The principle of image sparse decomposition and image sparse representation are introduced. Fast algorithm based on differential evolution algorithm is used for the large computation issue of image sparse decomposition, and the decomposition process and implementation process is given.2.
12、In order to get the high-quality image sparse decomposition atoms and projection components, this paper has the further study for differential evolution, and presents the differential evolution algorithm based on the population diversity, which is used in image sparse decomposition. Firstly, the alg
13、orithm analyzes five different existing differential strategy and select the best one by comparing the reconstructed image quality of the sparse decomposition; Secondly considering of the issue that the diversity of population is down with the increase in the number of iterations in the optimization
14、 process, and measuring population diversity instead of a fixed number of iterations to find optimal conditions for the termination. The experimental results show that comparing with the particle swarm optimization algorithm of sparse decomposition and traditional differential evolution algorithm, t
15、his algorithm can effectively improve the peak signal noise ratio of reconstructed image under the same conditions, which can get the image representation more efficiently, quickly and accurately.3. According to the distribution of the six components of the image Sparse representation, a variable co
16、de length encoding compression algorithm is given after analysis of the traditional sort of lack of differential encoding. Firstly, the projection component of decomposition data is processed by the sort of projection differential, and then assigning a number of yards long by the component of atoms
17、for each the differential component, two translational and one rotational component which have the larger amount of information, finally using fixed-length encoding for the two small-scale distribution components. The experimental results show that in the same conditions the algorithm can get a high
18、er compression ratio compared with the algorithm of the literature under the same peak signal noise ratio of the compression reconstructed image, which can effectively reduce the coding redundancy and improve compression efficiency.4. To further improve the image compression ratio, the compression a
19、lgorithm with Run-Length Encoding is given according to the parameters components characteristics of the image sparse representation, which use the idea of run-length encoding in the image compression based on the spare decomposition. The experimental results show that the algorithm can get a higher
20、 compression ratio at the same conditions by comparing with the varying code length algorithm and literature algorithm, which can improve the image compression efficiency.【关键词】图像稀疏分解 差分进化算法 差分策略 种群多样性 变码长编码 行程编码【英文关键词】image spare decomposition differential evolution algorithm differential strategy p
21、opulation diversity varying length encoding run-length encoding【目录】数字图像差分进化稀疏分解及压缩 摘要 6-8 Abstract 8-9 第 1 章 绪论 12-16 1.1 引言 12 1.2 论文的提出及研究意义 12 1.3 国内外现状分析 12-14 1.3.1 图像稀疏分解研究现状 12-13 1.3.2 图像压缩技术发展现状 13-14 1.4 论文的主要工作 14-15 1.5 论文的结构安排 15-16 第 2 章 数字图像稀疏分解的差分进化实现 16-23 2.1 引言 16 2.2 图像的稀疏分解 16-1
22、9 2.2.1 基本思想 16-17 2.2.2 非对称原子库 17-18 2.2.3 图像稀疏分解效果评价 18-19 2.3 图像稀疏分解实现 19-22 2.3.1 差分进化算法 19-20 2.3.2 基于差分进化算法的图像稀疏分解 20-22 2.4 小结 22-23 第 3章 基于种群多样性的差分进化自适应稀疏分解算法 23-42 3.1 引言 23 3.2 差分策略研究 23-32 3.2.1 性能对比 23-24 3.2.2 实验对比 24-32 3.3 基于种群多样性的自适应稀疏分解算法 32-41 3.3.1 种群多样性度量 32-33 3.3.2 基于种群多样性的自适应分
23、解算法 33-35 3.3.3 实验结果及分析 35-41 3.4 小结 41-42 第 4 章 变码长图像压缩算法 42-52 4.1 引言 42 4.2 图像压缩效果评价 42 4.3 变码长压缩算法 42-46 4.3.1 稀疏分解结果分布规律 43-44 4.3.2 算法描述 44-45 4.3.3 编码 45-46 4.4 实验结果及分析 46-51 4.5 小结 51-52 第 5 章 采用行程编码的图像压缩算法 52-61 5.1 引言 52 5.2 行程编码压缩算法 52-54 5.2.1 行程编码原理 52 5.2.2 算法描述 52-54 5.2.3 编码 54 5.3 实验结果及分析 54-60 5.4 小结 60-61 工作总结与展望 61-63 致谢 63-64 参考文献 64-68 攻读硕士学位期间发表的论文及科研成果 68
