sparse coding中的字典学习方法研究毕业论文

 2021-04-10 10:04

摘 要

稀疏编码,是指将数据向量用基本元素的稀疏线性组合表示的理论。这一理论广泛应用于机器学习,神经网络,信号处理和统计学中。在最近的研究中,大量的多种学科性质的研究已经致力于稀疏编码及其应用中。在统计学和机器学习中,我们用稀疏原则来进行模型选择,即在众多的模型中选择最简单且最适合的一个。在信号处理领域,稀疏编码是指用一些字典元素的线性组合来表示数据,这些字典元素就是小波。在后来的发展中,一些科学领域,比如神经科学,生物信息或计算机视觉中,相对应的系数编码工具已经被广泛地应用。

当今世界是科技发展日新月异的世界。尤其计算机科学飞速发展,网络让整个世界变得更加紧密,信息交换量迅猛增长,我们迎来了大数据云计算的时代。无论在机器学习,神经网络,信号处理,图像处理还是统计学中,稀疏编码理论都有着非常重要的应用和现实意义。本文尝试从稀疏编码理论在图像处理这一冰山一角出发,阐述稀疏编码理论的重要应用。

本篇论文的目标就是介绍字典学习和稀疏编码的研究背景和意义,讨论字典学习和稀疏编码在各个领域的发展历程。接着,我们要介绍信号的稀疏表示,稀疏理论的基础还有要用到的相关数学基础知识。然后,介绍MATLAB基础语法知识和要用到的函数。并对实验机器的配置进行说明,以更好地衡量后面讨论的字典学习时间。紧接着,我们介绍预处理过程和多种字典学习方法的理论知识,介绍用到的字典学习方法的出处。在实践上我们从多个角度,多个维度考察不同的字典学习算法,并试图从中找出最优的字典学习算法。更进一步,我们会将学习得的字典应用到实践中。利用不同的字典学习和矩阵分解算法,通过字典和系数矩阵我们分别还原黑白图像和彩色图像,并进行比较和具体生活实践中的应用。

关键词:稀疏编码;图像处理;字典学习;图像还原;MATLAB

ABSTRACT

Sparse coding---that is, modelling data vectors as sparse linear combinations of basic elements---is widely used in machine learning, neuroscience, signal processing and statistics. In recent years, a large amount of multi-disciplinary research has been conducted on sparse models and their applications. In statistics and machine learning, the sparsity principle is used to perform model selection—that is, automatically selecting a simple model among a large collection of them. In signal processing, sparse coding consists of representing data with linear combinations of a few dictionary elements. Subsequently, the corresponding tools have been widely adopted by several

scientific communities such as neuroscience, bioinformatics, or computer vision.

In this era of technology, things keep developing day by day. Especially the computer science has been developing the most. The network makes the whole world more and more communicable. As the amount of information exchanging increases, the age of big data and cloud computing comes. No matter in machine learning, neuroscience, signal processing, image processing or statistics, sparse coding has always been playing key roles and has practical significance. This paper tries to state the significant applications of sparse coding theory in image processing, which is just the basic and simple part of the theory.

The goal of this paper is to introduce the background and the significance of dictionary learning and sparse coding and to talk about the developing history of dictionary learning and sparse coding in many fields. Then, we will introduce the sparse represent of data, the fundament of sparse coding and the basic knowledge of relative math used. And the basic knowledge of MATLAB and several functions will be presented. We will give the configuration of the PC we used just to measure the computing time of dictionary learning approaches. Subsequently, we are going to introduce the pre-processing and several dictionary learning methods. We will test different dictionary learning methods at multiple perspectives and try to find out the optimal one among them. Further, we will apply the dictionary we learnt to the practice. Using distinct dictionary learning methods and matrix factorization algorithms, We restore the grey and colorful images by dictionaries and coefficients ,compare them and talk about the application of them in practice.

Key words: sparse coding; dictionary learning; image processing; image restoring; MATLAB

目 录

第一章 绪论 1

1.1 研究背景和意义 1

1.2 研究现状 1

1.3 本文主要研究内容及论文组织结构 3

1.3.1本文的主要研究内容 3

1.3.2 论文组织结构 4

第二章 背景知识介绍 5

2.1 数学知识 5

2.1.1 范数 5

2.1.2 罚函数 5

2.1.3 Frobenius范数 5

2.2 稀疏编码理论 6

2.2.1 小波与信号处理 6

2.2.2 字典学习概述 7

第三章 MATLAB基础知识 9

3.1 MATLAB基础知识 9

第四章 字典学习方法比较 12

4.1 实验用PC的配置 12

4.2 预处理 12

4.2.1 中心化(centering) 12

4.2.2 对比度归一化(contrast normalization) 12

4.2.3 彩色图形块的预处理 13

4.3 不同的字典学习算法介绍 13

4.4 字典学习算法比较 14

4.4.1 比较不同字典学习算法的性能(学习速度,精度) 14

4.4.2 同一算法在对字典不同限制情况下的性能(学习速度,精度) 17

4.4.3 同一算法在不同iter,lambda情况下的性能(学习速度,精度) 20

4.4.4比较mexTrainDL和mexTrainDL_Memory的性能(学习速度,精度) 23

第五章 字典学习方法的应用 27

5.1黑白图像还原(一) 27

5.1.1代码(核心代码,详细代码见附录) 27

5.1.2运行结果 28

5.1.3 结果分析 28

5.2黑白图像还原(二) 29

5.2.1代码 29

5.2.2运行结果 30

5.2.3 结果分析 30

5.3彩色图像还原 31

5.3.1代码 31

5.3.2 运行结果 32

5.3.3 结果分析 33

第六章 结束语 35

致谢 36

附录 38

4.4.1代码 38

4.4.2 代码 41

4.4.3 代码 44

4.4.4 代码 (略) 47

5.1.1 代码 47

5.2.1 代码 49

5.3.1 代码 51

您需要先支付 80元 才能查看全部内容!立即支付

课题毕业论文、开题报告、任务书、外文翻译、程序设计、图纸设计等资料可联系客服协助查找,优先添加企业微信。