DCT子空间的非局部均值去噪算法_胡金蓉.pdf在整个图像块像素灰度值向量空间中，非局部均值（ｎｏｎ

文件名称: DCT子空间的非局部均值去噪算法_胡金蓉.pdf

所属分类: 机器学习

开发工具:

文件大小: 1mb

下载次数: 0

上传时间: 2019-07-21

提供者: imbe****

下载 (1mb)

不能下载？报告错误

详细说明：在整个图像块像素灰度值向量空间中，非局部均值（ｎｏｎｌｏｃａｌｍｅａｎｓ，ＮＬＭ）算法度量像素间的相似性不仅计算复杂度高，而且当噪声存在时还不能准确地计算出像素间的相似性权重值，影响了对图像冗余性质的利用，使得去噪结果图像对比度和清晰度低．针对ＮＬＭ算法的这一缺陷，利用离散余弦变换（ｄｉｓｃｒｅｔｅｃｏｓｉｎｅｔｒａｎｓｆｏｒｍ，ＤＣＴ）的低数据相关性和高能量紧致性，将ＤＣＴ与ＮＬＭ算法相结合，对图像块进行ＤＣＴ，并在ＤＣＴ低频系数子空间内度量像素间的相似性．实验结果表明，与ＮＬＭ算法相比，该方法能够在保护图像结构信息、对比度和清晰度的前提下更有效地去除噪声，峰值DCT DCT NLM NLM-DCTS DCT u M (3) p(x 2t(i) Za(i) p(r, y) d∈[1, x(2x+1)u 丌(zy+1) (2)M DCT Cos COS 7 x,y=0,1,2∴…,r-1,r L·了 ,(ck(N)一 0.1,2 cLd」(N)-cLd」(N,)‖";cLdl(N √1{r=0 7l≠0 c[d(N2)]={C(u,v)=c(l)a(v) P(N 1r=0 x(2x+1)l (2y+1) 2/rv≠0 COS DCT d=M NLM DCT NLM-DCTS 1a.1b 128×128 1c,1d 1d13% DCT NLM NLM-DCTS a 41×41;2d 25 (21,21) Lena子图 b Barbara子图 NLM- DCTS NLM 7×7 (41×41) gag 20% d=10.2b,2e 2 b NLM c重构的Lena子图 d重构的 Barbara子图 1 DCI NLM NLM 2c,2f NLM-DCTS DCT DCT NLM NLM-DCTS DCT 21994-2014ChinaAcademicJournalElectronicPublishingHousealLrightsreservedhttp://www.cnki.net 24 a原始图像 baf权值分布(NM) a的权值分布( NLM-DCTS) d噪声图像σ=25 ed的权值分布(NLM) fd的权值分布( NLM-DCTS) BF SVD, NLM-PNDLLOJ BM3D[17 PSNR NLM-DCTS NLM-DCTS PSNR (a=25)dB NLM.NLM- Ho P∈pp SVDLS. NLM-PND0 3 NLM-DCIS 31.64 Boat House. Pe eppes BE 280 Brain web MRI NLMLlJ 30.31 28.42 30.81 28.77 NLM-SVD-3I 0.93 (i)=u(i)+n(t),,v(i) NLM-PND1131.1289131.6529.33 (i) BM3D-1YI 3206 29.85 3284 30.11 PSNR NLM-DCTS PSNR RPsN-101g 255/Ems (4) (li(i)-u(i))2/ NLM 1,10 NLM NLM-DCTS 21×217×7; Zigzag NLM-PND PSNR 20% DCT NLM-SⅤD NLM-DCTS 10 PSNR BM3D NLM-DCT NLM-DCTS house 3b~3h PM BE 21994-2014ChinaAcademicJournalElectronicPublishingHousealLrightsreservedhttp://www.cnki.net . DCT 93 NLM. NLM-SVD, NLM-PND. BM3D NLM DCTS house NLM-DCTS NLM-DCTS NLM NLM-PND NLM-SVD 2 NLM 山山 a噪声图像 bPM算法去噪结果 cBF算法去噪结果 dNLM算法去噪结果 e NLM-SV算法去噪结果fNLM-PND算法去噪结果 gBM3U算法去噪结果 hNLM-DIS算法去噪结果 i噪声残差图像 jPM算法噪声图像 kBF算法噪声图像 NLM算法噪声图像 m NLM-SV算法噪声图像 n NLM-PND算法噪声图像 0BND算法噪声图像 p NM-DCTS算法噪声图像 ouse NLM-DCTS NLM PSNR NLM 5,10,15,20,25,50,75100 PSNR NLM PSNR PSNR C 0=5 0=10),NLM-DC TS Brain web MRI 21994-2014ChinaAcademicJournalElectronicPublishingHousealLrightsreservedhttp://www.cnki.net 94 24 NLM-DCTS PSNR NLM NLM NLM-DCTS 10 NLM 35 35 35 汤 15 一NLM-DcTS NLM-DCTS 15-E-NLM-DCTS - NLM NLM NLM 噪声图像 e—噪声图像 -噪声图像 5 15 75 75 25 75 a House b Lena c Peppers 35 一 NLM-DCTS 15}-一 NLM-DCTS E NLM-DCTS NLM =早NLM 一NLM 噪声图像母一噪声图像一噪声图像 e Brain We TI f Brain Web t2 号一 NLM.DCTS - NLM 噪声图像 g Brain Web PD 4 NLM-DCTS NLM PSNR NLM-DCTS NLM 5c,5f Lena pper rain n web t1 NLM- NLM-DCTS DCTS NLM-DCTS 5 n 21994-2014ChinaAcademicJournalElectronicPublishingHousealLrightsreservedhttp://www.cnki.net DCT 9 5 o DCT 50) NLM-DCTS PSNR a Lena a=25bNLM对a的去噪结果 C NLM-DCTS对 d lena a=50eNLM对d的去噪结果 f NLM-DCTS对 a的去噪结果 d的去噪结果 g Peppers a=25hNLM对g的去噪结果 I NLM-DCTS对 Peppers a=50kNLM对j的去噪结果1 NLM-DCTS对的去噪结果 j的去噪结果 m BrainWeb T=25nNLM对m的去噪结果0 NLM-DCTS对 o BrainWeb T1σ=50qMM对p的去噪结果 NLM-DCTS对 m的去噪结果的去噪结果 5 NLM-DCTS NLM-DCTS NI NLM O(|g·S|·M) 1/10, JP NLM-DCTS NLM NLM- DCTS NLM LM-DCTS MATLAB NLM House, Peppers 256×256,Lena 512×512 2 NLM-DCTS NLM S House Peppers Le Boat NLM 114 468 473 DCT NLM NLM-DCTS 36 36 21994-2014ChinaAcademicJournalElectronicPublishingHousealLrightsreservedhttp://www.cnki.net 96 24 Zigzag means algorithm for image denoising [J]. Journal of Computer DCT Science and Technology, 2008, 23(2): 270-279 [9 Orchard J, Ebrahimi M, Wong A. Efficient nonlocal-means DCL denoising using the SVD[C]//Proceedings of the 15th IEEE DCT International Conference on Image Processing. Los Alamitos IEEE Computer Society Press, 2008: 1732-1735 [10 Tasdizen T. Principal neighborhood dictionaries for nonlocal (References means image denoising [J. IEEE Transactions on Image Proccssing,2C09,18(12):2649-2600 [11 Ji Z X. Chen Q. Sun Q S, el ul. A moment-based 1 Buades A, Coll B, Morel J M. A review of image denoising nonlocal-means algorithm for image denoising [J] algorithms, with a new onc [J. Multiscale Modeling and Information Processing Letters, 2C09, 109(25/21): 12 Simulation,2005,4(2):490-530 1244 _] Tomasi C, Manduchi R. Bilateral filtering for gray and color [12] Sun Weifeng, Peng Yuhua. An improved non-lccal mears ges [C] //Proceedings of the 6th International Conference de-noising approach [J. Acta Electronica Sinica, 2010, 38 on Computer Vision. Washington D C: IEEE Computer (4):923-928(in Chinese Society press. 1998:839-846 3 Cai Chao, Ding Mingyue, Zhou Chengping, et al. Bilateral 2010,38(4):923-928) filtering in the wavelet domain [. Acta Electronica Sinica [13 Chen Q, Sun Q S, Xia D S. Homogencity similarity bascd 2004,32(1):128-131( in chinese) image denoising I. Pattern Recognition, 2010, 43(12) [JI 4089-4100 ,2004,32(1):128-131) [14J Ahmed N, Natarajan T, Rao K R. Discrete cosine transform 4 Perona P, Malik J. Scale-space and edge detection using LJ. IEEE Transactions cn Computers. 1974,C-23(1):90- anisotropic diffusion J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990,12(7): 629-639 [15 Hu Guangshu. Digital signal processing theory, algorith 5 Rudin L, Osher S, Fatemi E. Nonlinear total variation based and implementation [M. Bcijing 'Tsinghua Univcrsity noIse rene algorithms [J]. Physica D: Nonlinear Press. 2003, 363-408(in Chinese Phenomena,192,50(1/4):259258 LMI 6 Coifman RR, Donoho D L. Translation- invariant de-noising 203)3:363-40)8) [C//Proceedings cf Wavelets and Statist ics, Springer L16 Khayam S A. The discrete cosine transform dct ): theory Lecture Notes in Statistics, New York Springer, 1995, 103: and application [R]. Michigan State: Michigan State 125-150 University. Department of Electrical & Computer 7 Mahmoudi M, Sapiro g. Fast image and video denoising via eering, 2003 nonlocal means of similar neighborhoods [J_. IEEE Signal [17] Dalov K. Foi A. Ka koynik v. ei al. Image denoising by Processing Letters, 2005, 12( 12): 839-812 sparse 3D transform-domain collaborative filtering[J. IEEE 8 Liu Y L, Wang J, Chen X, et al. A robust and fast non-lccal transactions cn Image processing, 2007,16(8): 2080-2095 21994-2014ChinaAcademicJournalElectronicPublishingHousealLrightsreservedhttp://www.cnki.net

(系统自动生成,下载前可以参看下载内容)