A non local algorithm for image denoising
2013-02-15 21:20
387 查看
[/code]
本文为Non-localprinciple应用于图像去噪中的论文“Anonlocalalgorithmforimagedenoising”的读书笔记。 Non-localprinciple近年被用于imagematting中,取得了很好的效果。 参考文献: A.Buades,B.Coll,J.M.Morel,"Anonlocalalgorithmforimagedenoising", IEEEComputerVisionandPatternRecognition2005,Vol2,pp:60-65,2005. A.Buades,B.Coll,J.M.Morel,“NonlocalImageandMovieDenoising”,IntJComputVis(2008)76:123–139. Non-localMatting,Lee,CVPR2011. KNNMatting,Chen,CVPR2012.
Anonlocalalgorithmforimagedenoising
Sason@CSDN
1.介绍 定义去噪方法Dh为一个分解: v=Dh*v+n(Dh,v), 其中v是有噪图像,h为滤波参数,常取决于噪声的标准差。理想情况下Dh*v比v更平滑,n(Dh,v)接近于白噪声。 去噪方法不应改变原始图像u。但是许多去噪方法退化或去掉了u中的细节和纹理。介绍并分析方法噪声(methodnoise)用于更好的理解去噪方法导致的结果。 2.方法噪声(methodnoise) 定义:u为图像,Dh为依赖于滤波参数h的去噪算子,定义方法噪声为图像间的差异 u−Dh*u. 对多种经典的局部平滑滤波器的方法噪声进行计算和分析:高斯滤波法(theGaussianfiltering),各向异性滤波法(theanisotropicfiltering),全变分最小化法(theTotalVariationminimization)以及邻域滤波法(theneighborhoodfiltering)。 3.NL-means算法 对于离散有噪声图像v={v(i)|i∈I},对于像素i的估计值NL[v](i),为图像中所有像素的加权平均值。其中的权值为像素i和像素j的相似度,满足条件:权值在[0,1]之间,且总和为1。 像素i和像素j的相似度定义为灰度值矢量v(Ni)和v(Nj)间的欧氏距离,Nk表示以像素k为中心的固定尺寸方形邻域。 有着与v(Ni)相似的灰度邻域的像素在加权平均时有较大的权值。 4.NL-means一致性 定理4说明NL-means算法校正有噪图像,而不是从真实图像中分离噪声; 定理5说明条件期望值是V(Ni\{i})的函数,用于最小化真实图像u的均方根误差。 5.讨论与实验 实验环节比较thelocalsmoothingfilters和NL-means算法的性能,使用3个判据:方法噪声(themethodnoise),视觉质量(thevisualqualityoftherestoredimage)以及均方根误差(themeansquareerror)。 实验中设置searchwindow为21*21像素,similaritysquareneighborhoodNi为7*7像素。若图像像素为N^2,则算法复杂度为49*441*(N^2)。Thefilteringparameterh设置为10∗σ,σ为噪声的标准差。 1.通过计算方法噪声获得差异图像,可见NL-Means方法噪声不表示任何可注意到的几何结构; 2.从视觉质量来看,NL-Means方法更好; 3.NL-Means方法求得的均方根误差约为其他方法的1/2。
欢迎来到我的CSDN博客:http://blog.csdn.net/anshan1984/
相关文章推荐
- A non-local algorithm for image denoising
- Non-local Sparse Models for Image Restoration
- denoising 之 Non local means
- 《Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising》学习笔记
- Hyperspectral image super-resolution via non-local sparse tensor factorization
- 局部特征用于图像检索 Aggregating local features for Image Retrieval
- 去雾 - Non-Local Image Dehazing
- Non-Photorealistic Rendering (Domain transform for edge-aware image and video processing)
- 论文笔记之:Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic
- A general iterated shinkage algorithm for non-convex sparse coding.
- deep learning Deep Belief networks for image denoising
- 论文笔记之:Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic
- Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image
- Passing address of non-local object to __autoreleasing parameter for write-back
- Incoherent dictionary learning for sparse representation based image denoising(一)
- Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image
- DL去燥之 DnCNNs:Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
- Discriminative Learned Dictionaries for Local Image Analysis学习笔记
- Passing address of non-local object to __autoreleasing parameter for write-back
- Local image caching solution for Android: Square Picasso vs Universal Image Loader