没有合适的资源?快使用搜索试试~ 我知道了~
首页基于SVD的感知非局部均值图像去噪技术
本文主要探讨了"PERCEPTUAL NON LOCAL MEAN (P-NLM) DENOISING"这一主题,它是一种基于非局部均值的图像复原技术。该研究由W. Souidene等人在L2TI实验室进行,位于法国Villetaneuse的99 avenue J.B Clément。作者们提出了一种创新的图像去噪方法,利用SVD(奇异值分解)为基础的图像质量度量作为邻域相似性的衡量标准。通过这种方法,他们计算出空间上的高斯加权核,这有助于提高去噪效果。 传统意义上的图像去噪往往依赖于局部像素间的相似性,而P-NLM方法则超越了这一局限,考虑了全局的、感知上的相似性。SVD在这里的作用在于,它能够捕捉到图像中的关键特征,从而更准确地识别出哪些区域应该被保留或降噪。这种全局的相似性衡量使得P-NLM在处理噪声和保持图像细节方面更具优势。 此外,论文还着重提到了优化计算方案的开发,通过并行架构来加速滤波过程。这不仅适用于多台机器之间的协作,也适用于单机上多个核心的高效利用,显著提高了处理速度,使得大规模图像处理成为可能。 作者们指出,尽管P-NLM去噪技术是基于一种主观感知模型,但它在客观图像相似性评估中展现出了很好的性能。这表明,尽管它的设计初衷是为了更好地符合人类视觉系统的感知,但其结果在量化评价标准下也表现出色。 关键词包括:图像去噪、非局部均值滤波、感知模型、感知滤波和客观图像相似性。P-NLM方法在图像处理领域具有很大的潜力,为提高图像质量和处理效率提供了一种新颖且有效的策略。未来的研究可能会进一步探索如何优化这种算法,以适应更多复杂的图像环境和应用场景。
资源详情
资源推荐
PERCEPTUAL NON LOCAL MEAN (P-NLM) DENOISING
W. Souidene
1
, S. Megrhi
1,2
1
L2TI, 99 avenue J.B Cl
´
ement
93430, Villetaneuse
wided.mseddi, beghdadi@univ-paris13.fr
A. Beghdadi
1
, C. Ben Amar
2
2
Ecole Nationale d
0
Ing
´
enieurs de Sfax,
D
´
epartement de G
´
enie Electrique
Route de Soukra, B.P. 1173, 3038, Sfax, Tunisie
chokri.benamar@ieee.org, sam.khanfir@gmail.com
ABSTRACT
In this paper we propose a denoising technique based on non-
local means using an image similarity measure. The idea is
to use the SVD-based image quality metric as a measure of
neighborhood similarity. This measure is then used in the
computation of the spatial Gaussian weighting kernel. We
also develop an optimization computation scheme using a par-
allel architecture in order to accelerate the filtering process
on different machines or different cores on the same machine.
The obtained results are very promising.
Keywords : image denoising, non local means filtering,
perceptual model, perceptual filtering, objective image simi-
larity.
1. INTRODUCTION
Image denoising is one of the most attractive research fields
in image processing. A number of recent studies have shown
that exploiting the spatial and photometric redundancy of the
image in the design of the filtering process yield interesting
results and overcome the classical approaches. This redun-
dancy appears not only on some textured images but also in
many natural images. The Non Local Mean (NLM) technique
introduced in [1] restores the original image by considering
non local neighborhoods of a given pixel. It has been shown
that the concept of non local neighborhoods is very relevant
for natural as well as textured images. In fact, it exploits the
redundancy and allows a better contribution of different im-
age structures to denoise similar ones. The notion of similar-
ity is very often computed as an MSE-based measure which
is not actually correlated with the human perception. In this
article, we first present an overview of the NLM as introduced
in [1] then we propose a different scheme of similarity calcu-
lation inspired by some characteristics of the on the Human
Visual System (HVS). The basic idea is to use a more consis-
tent measure of similarity between windows in order to select
the most perceptual relevant windows in the weighting pro-
cess. The filtering is then performed based on these selected
windows. However, the improvement, in terms of denoising
quality, is gained at the expense of complexity and time con-
sumption. To this end, we propose an optimized and rapid
implementation of the filtering based on a parallel architec-
ture. The paper is organized as follows. Section 2 introduces
the NLM concept and the a most important steps. The follow-
ing section presents the proposed idea. Section 4 gives some
details on the parallelization scheme implementation. The re-
sults are given in section 5. Finaly the last section is devoted
to conclusion and perspectives.
2. PROBLEM STATEMENT
The Non Local Means denoising technique has been recently
developed to overcome the limitations of classical linear fil-
tering methods [2][1] [3][4][5]. Its principle is quite simple
and it was used in the spatial domain as well as in the trans-
form domain [6]. Using a noisy version g of the original
image f, we restore each pixel using its neighbors in a pre-
specified region of the image. The definition of a neighbor is
no longer a ”spatial” definition but a definition based on sim-
ilarity between pixels. So taht, pixels i and j will be consid-
ered are neighbors if they belong to similar structures of the
image. Let I be the support of original and noisy images (we
suppose that they have the same size). Let (n
1
, n
2
) be a pixel
in noisy image g. The estimated value of the pixel (n
1
, n
2
)
in the original image is computed as a weighted average of all
the pixels in the image (or in a portion of the image):
f(n
1
, n
2
) =
X
(k
1
,k
2
)∈[1,m
f
]×[1,n
f
]
w
n
1
,n
2
(k
1
, k
2
)g(k
1
, k
2
)
(1)
where w
n
1
,n
2
(k
1
, k
2
) is a weight associated with the similar-
ity between pixels (n
1
, n
2
) and (k
1
, k
2
).
w
n
1
,n
2
(k
1
, k
2
) =
1
Z(n
1
, n
2
)
exp(−
kg
n
− g
k
k
2
2
γ
2
) (2)
where g
n
and g
k
denote the neighborhoods of pixels (n
1
, n
2
)
and (k
1
, k
2
) respectively and γ is a constant that controls the
decay of the exponential function. Z is a normalizing factor
Proceedings of the 5th International Symposium on Communications, Control and Signal Processing,
ISCCSP 2012, Rome, Italy, 2-4 May 2012
978-1-4673-0276-0/12/$31.00 ©2012 IEEE
下载后可阅读完整内容,剩余3页未读,立即下载
rainpgy
- 粉丝: 3
- 资源: 17
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
- 深入理解23种设计模式
- 制作与调试:声控开关电路详解
- 腾讯2008年软件开发笔试题解析
- WebService开发指南:从入门到精通
- 栈数据结构实现的密码设置算法
- 提升逻辑与英语能力:揭秘IBM笔试核心词汇及题型
- SOPC技术探索:理论与实践
- 计算图中节点介数中心性的函数
- 电子元器件详解:电阻、电容、电感与传感器
- MIT经典:统计自然语言处理基础
- CMD命令大全详解与实用指南
- 数据结构复习重点:逻辑结构与存储结构
- ACM算法必读书籍推荐:权威指南与实战解析
- Ubuntu命令行与终端:从Shell到rxvt-unicode
- 深入理解VC_MFC编程:窗口、类、消息处理与绘图
- AT89S52单片机实现的温湿度智能检测与控制系统
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