矩阵低秩逼近在图像压缩中的应用-应用数学专业毕业论文.docx

摘 要 图像是人类获取信息，表达信息和传递信息的重要手段。但是图像占有相当 大的数据量，如果直接进行存储和传播，就需要太大的存储器和带宽，因此，图 像压缩非常有必要。 本文致力于研究矩阵的低秩逼近实现图像压缩,提出了一种矩阵广义低秩逼 近算法中获得 L 和 R 的非迭代算法—SVGLRAM。首先，本文回顾了矩阵低秩逼 近的一种重要算法—奇异值逼近。详细介绍了奇异值分解的原理，以及如何用奇 异值分解实现压缩图像，该算法具有最小的误差，但是具有较高的时间和空间复 杂度。其次，本文讨论另一种低秩逼近算法—广义低秩逼近。这种算法可以一次 对若干幅相似的图像实现压缩，并且具有较小的时间和空间复杂度。最后,本文提 出了获得 L 和 R 的非迭代算法。通过在不同的人脸数据库上选取不同的参数做实 验得知，这种算法能获得和 GLRAM 接近的误差和压缩比，而且具有很小的时间 和空间复杂度。 关键词：图像压缩 奇异值 低秩逼近 压缩率 Abstract Images are important media to obtain, express and transmit messages, while they o ccupy a large amount of data space. It is very necessary that we compress the images be cause a large memory and bandwidth will be needed if they are stored and transmitted st raight. This paper mainly focuses on the realization of image compression. And the main innovation of this paper is SVGLRAMa non-Iterative Algorithm of Generalized Low Rank Approximations of Matrices and it will be proved theoretically and practically as follows: Firstly, this paper reviews an important algorithm of Low Rank Approximations of Matrices, that is, singular value approximation, its development, principle, and how to use the singular value decomposition to compress image, it has its advantage of the minimum error but the disadvantage of high time and space complexity. Secondly, this paper discusses another recently developed low rank approximation iterative algorithm of generalized low rank approximation. This algorithm can achieve compression of a number of images at one time, owning low time and space complexity as well; experiments are carried out to test its correctness in this part. Finally，this paper presents SVGLRAMa creative and non iterative algorithm. By selecting different parameters from different face databases, the experimental results show that it can not only obtain the error and the compression rate close to GLRAM, but also low time and space complexity. Keywords: Image compression Singular value Low rank approximation Compression rate 目 录 第一章 绪论 1 1.1 课题的来源与发展概况 1