没有合适的资源?快使用搜索试试~ 我知道了~
目前, 稀疏表示的应用范围基本为自然信号形成的图像、音频以及文本等, 对于非自然信号或数据的应用尚未有文献涉及。在应用方面, 可大体划分为两类: 基于重构的应用 此类应 用 有 图 像 去 噪、 压 缩 与 超 分 辨 、S A R 成像 、 缺失图像重构 以及音频修复 等。这些应用主要将目标的特征用若干参数来表示, 这些特征构成稀疏向量, 利用稀疏表示方法得到稀疏向量, 根据数学模型进行数据或图像重构。在这些应用中, 观测数据一般含有噪声。 基于分类的应用 这类应用的本质是模式识别 , 将表征对象主要的或本质的特征构造稀疏向量, 这些特征具有类间的强区分性。利用稀疏表示方法得到这些特征的值, 并根据稀疏向量与某类标准值的距离, 或稀疏向量间的距离判别完成模式识别或分类过程, 例如盲源分离、 音乐表示与分类、 人脸识别 、文本检测。
资源推荐
资源详情
资源评论
Foundations and Trends
R
in
Computer Graphics and Vision
Vol. 8, No. 2-3 (2012) 85–283
c
2014 J. Mairal, F. Bach and J. Ponce
DOI: 10.1561/0600000058
Sparse Modeling for Image and
Vision Processing
Julien Mairal
Inria
1
julien.mairal@inria.fr
Francis Bach
Inria
2
francis.bach@inria.fr
Jean Ponce
Ecole Normale Supérieure
3
jean.ponce@ens.fr
1
LEAR team, laboratoire Jean Kuntzmann, CNRS, Univ. Grenoble Alpes,
France.
2
SIERRA team, département d’informatique de l’Ecole Normale Supérieure,
ENS/CNRS/Inria UMR 8548, France.
3
WILLOW team, département d’informatique de l’Ecole Normale Supérieure,
ENS/CNRS/Inria UMR 8548, France.
Contents
1 A Short Introduction to Parsimony 2
1.1 Early concepts of parsimony in statistics . . . . . . . . . . 6
1.2 Wavelets in signal processing . . . . . . . . . . . . . . . . 8
1.3 Modern parsimony: the ℓ
1
-norm and other variants . . . . 14
1.4 Dictionary learning . . . . . . . . . . . . . . . . . . . . . . 32
1.5 Compressed sensing and sparse recovery . . . . . . . . . . 35
1.6 Theoretical results about dictionary learning . . . . . . . . 39
2 Discovering the Structure of Natural Images 44
2.1 Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . 46
2.2 Principal component analysis . . . . . . . . . . . . . . . . 52
2.3 Clustering or vector quantization . . . . . . . . . . . . . . 56
2.4 Dictionary learning . . . . . . . . . . . . . . . . . . . . . . 59
2.5 Structured dictionary learning . . . . . . . . . . . . . . . . 60
2.6 Other matrix factorization methods . . . . . . . . . . . . . 64
2.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3 Sparse Models for Image Processing 75
3.1 Image denoising . . . . . . . . . . . . . . . . . . . . . . . 76
3.2 Image inpainting . . . . . . . . . . . . . . . . . . . . . . . 82
3.3 Image demosaicking . . . . . . . . . . . . . . . . . . . . . 84
ii
iii
3.4 Image up-scaling . . . . . . . . . . . . . . . . . . . . . . . 87
3.5 Inverting nonlinear local transformations . . . . . . . . . . 92
3.6 Video processing . . . . . . . . . . . . . . . . . . . . . . . 94
3.7 Face compression . . . . . . . . . . . . . . . . . . . . . . 96
3.8 Other patch modeling approaches . . . . . . . . . . . . . 99
4 Sparse Coding for Visual Recognition 106
4.1 A coding and pooli ng approach to image modeling . . . . 107
4.2 The botany of sparse feature coding . . . . . . . . . . . . 115
4.3 Face recognition . . . . . . . . . . . . . . . . . . . . . . . 122
4.4 Patch classification and edge detection . . . . . . . . . . . 124
4.5 Connections with neural networks . . . . . . . . . . . . . . 130
4.6 Other applications . . . . . . . . . . . . . . . . . . . . . . 135
5 Optimization Algorithms 140
5.1 Sparse reconstruction with the ℓ
0
-penalty . . . . . . . . . 141
5.2 Sparse reconstruction with the ℓ
1
-norm . . . . . . . . . . . 148
5.3 Iterative reweighted-ℓ
1
methods . . . . . . . . . . . . . . . 154
5.4 Iterative reweighted-ℓ
2
methods . . . . . . . . . . . . . . . 156
5.5 Optimization for dictionary learning . . . . . . . . . . . . . 158
5.6 Other optimization techniques . . . . . . . . . . . . . . . 169
6 Conclusions 170
Acknowledgments 172
References 173
Abstract
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 rep-
resenting data with linear combinations of a few dictionary elements.
Subsequently, the corresponding tools have been widely adopted by sev-
eral scientific communities such as neuroscience, bioinformatics, or com-
puter vision. The goal of this monograph is to offer a self-contained view
of sparse modeling for visual recognition and image processing. More
specifically, we fo cus on applications where the dictionary is learned
and adapted to data, yielding a compact representation that has been
successful in various contexts.
J. Mairal, F. Bach and J. Ponce. Sparse Modeling for Image and
Vision Processing. Foundations and Trends
R
in Computer Graphics and Vision,
vol. 8, no. 2-3, pp. 85–283, 2012.
DOI: 10.1561/0600000058.
1
A Short Introduction to Parsimony
In its most general definition, the principle of sparsity, or parsimony,
consists of representing some phenomenon with as few variables as
possible. It appears to be central to many research fields and is often
considered to be inspired from an early doctrine formulated by the
philosopher and theologian William of Ockham in the 14th century,
which essentially favors simple theories over more complex ones. Of
course, the link between Ockham and the tools presented in this mono-
graph is rather thin, and more modern views seem to appear later in
the beginning of the 20th century. Discussing the scientific method,
Wrinch and Jeffreys [1921] introduce indeed a simplicity principle re-
lated to parsimony as follows:
The existence of simple laws is, then, appa rently, to be re-
garded as a quality of nature; and accordingly we may infer
that it is justifiable to prefer a simple law to a more complex
one that fits our ob serv atio ns slightly better.
Remarkably, Wrinch and Jeffreys [1921] further discuss statistical mod-
eling of physical observations and r elate the concept of “simplicity” to
the number of learning parameters; as a matter of fact, this concept is
relatively close to the contemporary view of parsimony.
2
剩余201页未读,继续阅读
资源评论
qq_24599599
- 粉丝: 1
- 资源: 19
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功