2.765

2022影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

无监督多重非局部融合的图像去噪方法

陈叶飞 赵广社 李国齐 王鼎衡

陈叶飞, 赵广社, 李国齐, 王鼎衡. 无监督多重非局部融合的图像去噪方法. 自动化学报, 2022, 48(1): 87−102 doi: 10.16383/j.aas.c200138
引用本文: 陈叶飞, 赵广社, 李国齐, 王鼎衡. 无监督多重非局部融合的图像去噪方法. 自动化学报, 2022, 48(1): 87−102 doi: 10.16383/j.aas.c200138
Chen Ye-Fei, Zhao Guang-She, Li Guo-Qi, Wang Ding-Heng. Unsupervised Multi-non-local Fusion Image Denoising Method. Acta Automatica Sinica, 2022, 48(1): 87−102 doi: 10.16383/j.aas.c200138
Citation: Chen Ye-Fei, Zhao Guang-She, Li Guo-Qi, Wang Ding-Heng. Unsupervised Multi-non-local Fusion Image Denoising Method. Acta Automatica Sinica, 2022, 48(1): 87−102 doi: 10.16383/j.aas.c200138

无监督多重非局部融合的图像去噪方法

doi: 10.16383/j.aas.c200138
基金项目: 国家重点研发计划 (2018TFE0200200), 教育部重大科技创新研究项目, 北京智源人工智能研究院资助
详细信息
    作者简介:

    陈叶飞:西安交通大学电子与信息学部自动化科学与工程学院硕士研究生. 主要研究方向为模式识别, 图像去噪.E-mail: yefeichen@stu.xjtu.edu.cn

    赵广社:西安交通大学电子与信息学部自动化科学与工程学院教授. 主要研究方向为深度学习与图像识别, 大数据与群体智能, 多智能体系统协同导航与控制.E-mail: zhaogs@mail.xjtu.edu.cn

    李国齐:清华大学精密仪器系类脑计算研究中心副教授. 主要研究方向为信号处理, 类脑计算, 复杂网络. 本文通信作者.E-mail: liguoqi@mail.tsinghua.edu.cn

    王鼎衡:西安交通大学电子与信息学部自动化科学与工程学院博士研究生. 主要研究方向为图像处理, 模型优化.E-mail: wangdai11@stu.xjtu.edu.cn

Unsupervised Multi-non-local Fusion Image Denoising Method

Funds: Supported by National Key R & D Program of China (2018YFE0200200), Key Scientific Technological Innovation Research Project by Ministry of Education, Beijing Academy of Artificial Intelligence
More Information
    Author Bio:

    CHEN Ye-Fei Master student at the School of Automation Science and Engineering, Faculty of Electronic and Information Engineering, Xi'an Jiaotong University. His research interest covers pattern recognition, image denoising

    ZHAO Guang-She Professor at the School of Automation Science and Engineering, Faculty of Electronic and Information Engineering, Xi'an Jiaotong University. His research interest covers deep learning, image recogition, big data, swarm intelligence, multi-agent system cooperative navigation and control

    LI Guo-Qi Associate professor at the Center for Brain Inspired Computing Research, Department of Precision Instrument, Tsinghua University. His research interest covers signal processing, brain-like computing, and complex network. Corresponding author of this paper

    WANG Ding-Heng Ph. D. candidate at the School of Automation Science and Engineering, Faculty of Electronic and Information Engineering, Xi'an Jiaotong University. His research interest covers image processing and model optimization

  • 摘要: 非局部均值去噪 (Non-local means, NLM) 算法利用图像的自相似性, 取得了很好的去噪效果. 然而, NLM 算法对图像中不相似的邻域块分配了过大的权重, 此外算法的搜索窗大小和滤波参数等通常是固定的且无法根据图像内容的变化做出自适应的调整. 针对上述问题, 本文提出一种无监督多重非局部融合 (Unsupervised multi-non-local fusion, UM-NLF) 的图像去噪方法, 即变换搜索窗等组合参数得到多个去噪结果, 并利用 SURE (Stein's unbiased risk estimator) 对这些结果进行无监督的随机线性组合以获得最终结果. 首先, 为了滤除不相似或者相似度较低的邻域块, 本文引入一种基于可微分硬阈值函数的非局部均值 (Non-local means with a differential hard threshold function, NLM-DT) 算法, 并结合快速傅里叶变换 (Fast Fourier transformation, FFT), 初步提升算法的去噪效果和速度; 其次, 针对不同的组合参数, 利用快速 NLM-DT 算法串联生成多个去噪结果; 然后, 采用蒙特卡洛随机采样的思想对上述多个去噪结果进行随机的线性组合, 并利用基于 SURE 特征加权的移动平均滤波算法来抑制多个去噪结果组合引起的抖动噪声; 最后, 利用噪声图像和移动平均滤波后图像的 SURE 进行梯度的反向传递来优化随机线性组合的系数. 在公开数据集上的实验结果表明: UM-NLF 算法去噪结果的峰值信噪比 (Peak signal to noise ratio, PSNR) 超过了 NLM 及其大部分改进算法, 以及在部分图像上超过了 BM3D 算法. 同时, UM-NLF 相比于 BM3D 算法在视觉上产生更少的振铃伪影, 改善了图像的视觉质量.
    1)   收稿日期 2020-03-16 录用日期 2020-06-11  Manuscript received March 16, 2020; accepted June 11, 2020 国家重点研发计划 (2018TFE0200200), 教育部重大科技创新研究项目, 北京智源人工智能研究院资助 Supported by National Key Research and Development Program of China (2018YFE0200200), Key Scientific Technological Innovation Research Project by Ministry of Education, Beijing Academy of Artificial Intelligence 本文责任编委 黄庆明
    2)  Recommended by Associate Editor HUANG Qing-Ming  1. 西安交通大学电子与信息学部自动化科学与工程学院 西安 710049 2. 清华大学精密仪器系类脑计算研究中心 北京 100084 1. School of Automation Science and Engineering, Faculty of Electronic and Information Engineering, Xi' an Jiaotong University, Xi' an 710049 2. Center for Brain Inspired Computing Research, Department of Precision Instrument, Tsinghua University, Beijing 100084
  • 图  1  传统非局部均值算法的执行流程图

    Fig.  1  Schematic diagrams of non-local means denoising algorithm

    图  2  参数$ h $对算法去噪效果的影响

    Fig.  2  The effect of parameter $ h $ on the denoising effect of the algorithm

    图  3  UM-NLF去噪算法的整体流程图

    Fig.  3  The overall flowchart of the UM-NLF denoising algorithm

    图  4  硬阈值加入前后相似权重值的对比

    Fig.  4  Comparisons of similar weight values before and after adding a hard threshold

    图  5  可微分的硬阈值函数

    Fig.  5  Differentiable hard threshold function

    图  6  不同算法对噪声等级$ \sigma = 20 $Airplane噪声图像的去噪结果PSNR (dB)和SSIM ((a) 原始无失真图像;(b) 噪声图像(22.08 dB/0.4397); (c) NLM (28.33 dB/0.8328); (d) NLM-SAP (28.94 dB/0.8526); (e) PNLM(29.04 dB/0.8490); (f) LJS-NLM (28.37 dB/0.8410); (g) ANLM (27.86 dB/0.8489); (h) BM3D (29.55 dB/0.8755); (i) NLM-DT (28.60 dB/0.8418); (j) UM-NLF (29.83 dB/0.8760))

    Fig.  6  Denoising results PSNR (dB) and SSIM on noisy image Airplane with noise level $ \sigma = 20 $ by different methods ((a) Ground truth; (b) Noisy image (22.08 dB/0.4397); (c) NLM (28.33 dB/0.8328); (d) NLM-SAP (28.94 dB/0.8526); (e) PNLM (29.04 dB/0.8490); (f) LJS-NLM (28.37 dB/0.8410); (g) ANLM (27.86 dB/0.8489); (h) BM3D (29.55 dB/0.8755); (i) NLM-DT (28.60 dB/0.8418); (j) UM-NLF (29.83 dB/0.8760))

    图  7  不同算法对噪声等级$ \sigma = 35 $“Test016” 噪声图像的去噪结果PSNR (dB)和SSIM ((a) 原始无失真图像;(b) 噪声图像(17.25 dB/0.4201); (c) NLM (24.41 dB/0.7202); (d) NLM-SAP (24.75 dB/0.7293); (e) PNLM(24.85 dB/0.7376); (f) LJS-NLM (23.94 dB/0.7168); (g) ANLM (24.89 dB/0.7540); (h) BM3D(25.48 dB/0.7850); (i) NLM-DT (25.04 dB/0.7422); (j) UM-NLF (25.91 dB/0.7853))

    Fig.  7  Denoising results PSNR (dB) and SSIM on noisy image “Test016” with noise level $ \sigma = 35 $ by different methods ((a) Ground truth; (b) Noisy image (17.25 dB/0.4201); (c) NLM (24.41 dB/0.7202); (d) NLM-SAP (24.75 dB/0.7293); (e) PNLM (24.85 dB/0.7376); (f) LJS-NLM (23.94 dB/0.7168); (g) ANLM (24.89 dB/0.7540); (h) BM3D (25.48 dB/0.7850); (i) NLM-DT (25.04 dB/0.7422); (j) UM-NLF (25.91 dB/0.7853))

    图  8  不同算法对噪声等级$ \sigma = 50 $Baboon噪声图像的去噪结果PSNR (dB)和SSIM ((a) 原始无失真图像;(b) 噪声图像 (14.16 dB/0.2838); (c) NLM (21.40 dB/0.4674); (d) NLM-SAP (21.33 dB/0.4386); (e) PNLM(21.48 dB/0.4793); (f) LJS-NLM (21.35 dB/0.4866); (g) ANLM (21.40 dB/0.4529); (h) BM3D(22.35 dB/0.5489); (i) NLM-DT (21.62 dB/0.4840); (j) UM-NLF (22.53 dB/0.5739))

    Fig.  8  Denoising results PSNR (dB) and SSIM on noisy image Baboon with noise level $ \sigma = 50 $ by different methods ((a) Ground truth; (b) Noisy image (14.16 dB/0.2838); (c) NLM (21.40 dB/0.4674); (d) NLM-SAP (21.33 dB/0.4386); (e) PNLM (21.48 dB/0.4793); (f) LJS-NLM (21.35 dB/0.4866); (g) ANLM (21.40 dB/0.4529); (h) BM3D (22.35 dB/0.5489); (i) NLM-DT (21.62 dB/0.4840); (j) UM-NLF (22.53 dB/0.5739))

    图  9  不同算法对Baboon噪声图像去噪后并经过XDoG滤波的结果

    Fig.  9  XDoG filtered results of the denoised images of different algorithms on noisy image Baboon

    表  1  UM-NLF算法的参数选择

    Table  1  Parameter selection of UM-NLF algorithm

    图像大小参数参数值
    邻域块的直径$[5, 7, 11]$
    搜索窗的直径$[13, 21]$
    高斯核的系数$[1, 2, 4]$
    $256\times 256$滤波参数$[0.8, 1.0, 1.2, \cdots, 2.4]$
    硬阈值参数$[0, 0.02, 0.04]\times \rm exp(-\frac{1}{\sigma^2})$
    线性组合的数目40
    蒙特卡洛的次数80
    加权移动平均的数目8
    邻域块的直径$[5, 7, 11]$
    搜索窗的直径$[13, 21]$
    高斯核的系数$[1, 2, 4]$
    $512\times 512$滤波参数$[0.8, 0.95, 1.1,\cdots, 2.3]$
    硬阈值参数$[0, 0.04, 0.08]\times \rm exp(-\frac{1}{\sigma^2})$
    线性组合的数目70
    蒙特卡洛的次数150
    加权移动平均的数目8
    下载: 导出CSV

    表  2  在13张噪声水平$\sigma$分别为10、15、20、25、35和50的噪声图像上不同算法去噪结果的PSNR (dB)

    Table  2  The PSNR (dB) results of different methods on 13 gray images with noise level $\sigma$ at 10、15、20、25、35 and 50

    LevelImagesC.manHousePeppersStarfishMonarchAirplaneParrotLenaBarbaraBoatManCoupleBaboon
    NLM31.8335.3533.2332.2232.4731.0231.4234.8533.8032.7032.6632.5928.55
    NLM-SAP33.5035.4934.2032.4433.6932.8632.9235.0633.7732.9733.1833.0829.78
    PNLM33.5035.2833.6632.7433.4932.8532.8234.6333.3932.9033.1532.7829.92
    LJS-NLM33.0835.1833.2732.1032.5432.2332.5534.5433.5632.6432.8232.5630.11
    $\sigma=10$ANLM30.3334.6632.0530.6930.9929.4429.8434.3332.7331.8832.0631.8626.97
    BM3D34.1936.7134.6833.3034.1233.3333.5735.9334.9833.9233.9834.0430.58
    NLM-DT32.0635.3933.2332.0533.3131.0331.5334.9033.6332.8732.9232.7428.64
    UM-NLF34.1336.0734.7133.5434.3433.5433.6035.7034.4433.6833.9833.6430.66
    NLM30.3533.7531.3530.3730.7529.5930.0632.8931.6730.8030.7130.4926.92
    NLM-SAP31.1833.8532.2430.3731.2830.5430.6833.3231.9331.0231.0330.9427.41
    PNLM31.2533.4631.4030.4231.1330.5830.6432.6031.0630.8630.9530.4527.59
    LJS-NLM30.8633.2631.0529.8930.2929.9330.3532.4331.1930.4530.5830.1127.54
    $\sigma=15$ANLM29.3733.3430.8329.5429.7928.6528.9832.8731.3230.5630.5830.4726.18
    BM3D31.9134.9332.6931.1431.8531.0731.3734.2633.1032.1331.9232.1028.18
    NLM-DT30.5533.7931.5230.3430.6929.6730.1233.0931.8431.0831.0830.8727.18
    UM-NLF31.9234.4732.7031.3632.1231.3431.4434.0132.5731.8231.9531.6128.32
    NLM29.2432.2629.8628.6829.3528.3328.9631.4129.9129.3229.3128.7925.44
    NLM-SAP29.6932.4830.7728.7829.5828.9429.2331.9230.3829.6029.5829.3125.94
    PNLM29.7331.9529.7928.6929.5129.0429.2431.1529.2729.3829.4428.7626.00
    LJS-NLM29.3531.6929.4328.1528.7528.3728.9230.9629.3828.9129.0528.3425.82
    $\sigma=20$ANLM28.5632.4129.6528.4728.7627.8628.2331.6830.1229.4529.4129.2525.27
    BM3D30.4933.7731.2929.6730.3529.5529.9633.0531.7830.8830.5930.7626.61
    NLM-DT29.3732.5130.1028.9729.3828.6029.0131.6830.2829.7029.7429.3525.97
    UM-NLF30.4433.2931.2229.8030.6129.8230.0332.7731.1730.5130.5930.2326.80
    NLM28.2930.8928.6227.2528.1927.2728.0530.2628.4928.1728.2427.4624.26
    NLM-SAP28.6131.2029.5527.4328.2627.7528.2430.7629.0528.4728.4827.9824.74
    PNLM28.5830.6828.5227.3328.2627.8328.2130.0227.8528.2328.3027.4624.77
    LJS-NLM28.1630.3928.1126.7627.5327.1627.8429.8627.9527.7427.9127.0424.57
    $\sigma=25$ANLM27.9131.6228.6627.4927.8827.0727.5730.6929.0528.5028.4928.1624.38
    BM3D29.4532.8530.1628.5629.2528.4228.9332.0730.7129.9029.6129.7125.46
    NLM-DT28.4731.2428.9527.7228.2927.7028.1130.5428.9728.6428.6928.1024.97
    UM-NLF29.3532.2930.0428.5629.4228.6729.0031.7930.0729.5129.5629.1225.69
    NLM26.5728.6426.6625.1626.3025.4826.5828.5326.3526.4426.6725.6722.69
    NLM-SAP26.8828.8827.4725.2726.3025.7626.7028.9426.9026.7226.8926.0622.89
    PNLM26.7628.5626.5425.3226.3425.8626.6528.2825.7626.4726.6225.6623.02
    LJS-NLM26.1928.2226.1124.7525.5825.1626.2228.2025.8626.0126.3025.3022.86
    $\sigma=35$ANLM26.8230.0527.1625.7626.4325.7926.4629.1627.2426.9727.1126.3122.86
    BM3D27.9231.3628.5126.8627.5826.8327.4030.5628.9828.4328.2228.1523.82
    NLM-DT26.9829.0527.0325.6526.5426.0126.7328.7626.8526.9127.0326.1523.38
    UM-NLF27.7730.6428.2626.7027.7026.8627.5430.2228.4227.9628.0727.4524.09
    NLM24.3926.1424.4823.1724.0023.3624.8226.6124.2624.6525.0724.0621.40
    NLM-SAP24.7526.3225.1023.1424.0723.3724.9527.0724.6324.9425.3624.3921.33
    PNLM24.6725.9624.2323.1723.9923.5424.8426.3423.7124.6124.8923.9621.48
    LJS-NLM24.0625.7923.8422.8323.3123.0624.3826.3423.8324.3024.7023.7121.35
    $\sigma=50$ANLM25.4327.9525.4423.8424.8224.2425.2127.6125.1925.3625.6924.5321.40
    BM3D26.1329.6926.6825.0425.8225.1025.9029.0527.2226.7826.8126.4622.35
    NLM-DT25.0826.6224.9423.6924.4924.0225.0426.7924.6525.0525.3024.3821.81
    UM-NLF26.0628.5726.2924.8025.8524.9926.0428.5126.5826.3126.5225.7122.53
    注: 最佳的两个结果分别以粗体和斜体突出显示
    下载: 导出CSV

    表  3  不同算法在BSD68灰度数据集上的平均PSNR (dB)和SSIM

    Table  3  Average PSNR (dB) and SSIM results of different methods on BSD68 gray dataset

    $\sigma$NLMNLM-SAPPNLMLJS-NLMANLMBM3DNLM-DTUM-NLF
    1031.37/0.880932.47/0.900632.52/0.900032.30/0.898430.72/0.883433.32/0.916331.57/0.893433.35/0.9184
    1529.64/0.828130.29/0.847230.25/0.847229.95/0.844829.44/0.841231.07/0.872029.96/0.846431.16/0.8760
    2028.31/0.780328.84/0.797328.73/0.799428.38/0.796728.41/0.802929.62/0.834228.72/0.801529.70/0.8367
    2527.28/0.736827.75/0.755027.60/0.756327.22/0.753927.56/0.768528.57/0.801727.73/0.758528.64/0.8034
    3525.75/0.660226.17/0.685425.95/0.681225.58/0.681426.25/0.709927.08/0.748226.20/0.676727.12/0.7451
    5024.20/0.564324.59/0.606924.29/0.589323.99/0.594524.88/0.640325.62/0.686924.52/0.568525.63/0.6768
    下载: 导出CSV

    表  4  不同算法在大小为$256\times 256$灰度噪声图像上的去噪时间(秒)

    Table  4  Denoising time (seconds) of different algorithms on gray noisy images with a size of $ 256\times 256$

    $\sigma$NLMNLM-SAPPNLMLJS-NLMANLMBM3DNLM-DTUM-NLF
    1071.74$\pm$0.8210.89$\pm$0.217.39$\pm$0.641.41$\pm$0.016.35$\pm$0.340.67$\pm$0.0968.49$\pm$1.4525.55$\pm$0.59
    1572.85$\pm$3.4511.11$\pm$0.207.79$\pm$0.581.41$\pm$0.036.31$\pm$0.040.68$\pm$0.0960.01$\pm$1.4224.41$\pm$1.23
    2072.58$\pm$4.3411.23$\pm$0.158.09$\pm$0.461.40$\pm$0.026.25$\pm$0.080.70$\pm$0.0970.68$\pm$0.8625.27$\pm$1.56
    2572.76$\pm$1.6211.36$\pm$0.108.37$\pm$0.581.40$\pm$0.016.26$\pm$0.200.68$\pm$0.1164.71$\pm$2.4524.98$\pm$3.12
    3571.68$\pm$0.7011.43$\pm$0.128.74$\pm$0.311.39$\pm$0.036.25$\pm$0.130.67$\pm$0.1060.40$\pm$0.7824.68$\pm$2.32
    5071.48$\pm$1.0711.48$\pm$0.058.95$\pm$0.401.39$\pm$0.016.27$\pm$0.190.87$\pm$0.0562.52$\pm$1.4524.30$\pm$1.22
    下载: 导出CSV

    表  5  不同算法在大小为$512\times 512$灰度噪声图像上的去噪时间(秒)

    Table  5  Denoising time (seconds) of different algorithms on gray noisy images with a size of $512\times 512$

    $\sigma$NLMNLM-SAPPNLMLJS-NLMANLMBM3DNLM-DTUM-NLF
    10284.37$\pm$13.8266.46$\pm$0.6638.48$\pm$3.4610.37$\pm$0.4025.27$\pm$0.653.16$\pm$0.13281.63$\pm$11.13155.64$\pm$8.90
    15282.18$\pm$11.2567.77$\pm$2.2241.25$\pm$3.5510.30$\pm$0.3325.46$\pm$0.683.26$\pm$0.11281.31$\pm$10.23160.28$\pm$5.34
    20285.50$\pm$11.3468.70$\pm$1.8242.85$\pm$2.5010.15$\pm$0.4925.70$\pm$0.663.32$\pm$0.11287.31$\pm$13.33155.94$\pm$9.91
    25283.65$\pm$12.7269.53$\pm$0.6344.46$\pm$4.0110.18$\pm$0.3125.54$\pm$0.323.33$\pm$0.08275.43$\pm$14.23151.56$\pm$5.23
    35284.15$\pm$15.7070.41$\pm$0.6046.88$\pm$1.8910.21$\pm$0.2525.40$\pm$0.713.20$\pm$0.17279.51$\pm$13.32150.65$\pm$4.43
    50279.63$\pm$19.0771.46$\pm$2.1149.37$\pm$2.8310.36$\pm$0.1825.40$\pm$0.493.86$\pm$0.09289.26$\pm$14.24158.75$\pm$6.45
    下载: 导出CSV

    表  6  传统算法和深度学习算法在BSD68灰度数据集上的平均PSNR (dB)和SSIM

    Table  6  Average PSNR (dB) and SSIM results of traditional and deep learning methods on BSD68 gray dataset

    传统的无监督去噪方法无监督式神经网络去噪方法有监督式神经网络去噪方法
    $\sigma$BM3DUM-NLFNoise2VoidGCBDTNRDDnCNN-S
    1531.07/0.872031.16/0.8760−/−31.59/−31.42/0.882231.73/0.8906
    2528.57/0.801728.64/0.803427.71/−29.15/−28.92/0.814829.23/0.8278
    5025.62/0.686925.62/0.6762−/−−/−25.97/0.702126.23/0.7189
    注: “−”表示原始论文中没有给出相应的实验结果
    下载: 导出CSV
  • [1] Tian C W, Fei L K, Zheng W X, Xu Y, Zuo W M, Lin C W. Deep learning on image denoising: an overview [online], available: http://arxiv.org/abs/1912.13171, January 16, 2020
    [2] Dong W S, Wang P Y, Yin W T, Shi G M, Wu F F, Lu X T. Denoising prior driven deep neural network for image restoration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(10): 2305−2318 doi: 10.1109/TPAMI.2018.2873610
    [3] Geman S, Geman D. Stochastic relaxation, gibbs distributions, and the bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1984, 6(5-6): 721−741
    [4] Buades A, Coll B, Morel J M. A non-local algorithm for image denoising. In: Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, California, USA: IEEE, 2005.60−65
    [5] Mahmoudi M, Sapiro G. Fast image and video denoising via non-local means of similar neighborhoods. IEEE Signal Processing Letters, 2005, 12(12), 839−842 doi: 10.1109/LSP.2005.859509
    [6] Coupe P, Yger P, Prima S, Hellier P, Kervrann C, Barillot C. An optimized blockwise nonlocal means denoising filter for 3-D magnetic resonance images. IEEE Transactions on Medical Imaging, 2008, 27(4), 425−441 doi: 10.1109/TMI.2007.906087
    [7] Vignesh R, Oh B T, Kuo C C J. Fast non-local means (NLM) computation with probabilistic early termination. IEEE Signal Processing Letters, 2010, 17(3), 277−280 doi: 10.1109/LSP.2009.2038956
    [8] Tasdizen T. Principal neighborhood dictionaries for nonlocal means image denoising. IEEE Transactions on Image Processing, 2009, 18(12): 2649−2660 doi: 10.1109/TIP.2009.2028259
    [9] Chan S H, Zickler T, Lu Y M. Monte carlo non-local means: random sampling for large-scale image filtering. IEEE Transactions on Image Processing, 2014, 23(8): 3711−3725 doi: 10.1109/TIP.2014.2327813
    [10] Karam C, Hirakawa K. Monte-carlo acceleration of bilateral filter and non-local means. IEEE Transactions on Image Processing, 2018, 27(3): 1462−1474 doi: 10.1109/TIP.2017.2777182
    [11] Wang J, Guo Y W, Ying Y T, Liu Y L, Peng Q S. Fast nonlocal algorithm for image denoising. In: Proceedings of the 2006 IEEE International Conference on Image Processing. Atlanda, GA, USA: IEEE, 2006.1429−1432
    [12] Chaudhury K N, Singer A. Non-local euclidean medians. IEEE Signal Processing Letters, 2012, 19(11), 745−748 doi: 10.1109/LSP.2012.2217329
    [13] Wu Y, Tracey B, Natarajan P, Noonan J P. Probabilistic non-local means. IEEE Signal Processing Letters, 2013, 20(8), 763−766 doi: 10.1109/LSP.2013.2263135
    [14] Deledalle C A, Duval V, Salmon J. Non-local methods with shape-adaptive patches (NLM-SAP). Journal of Mathematical Imaging and Vision, 2012, 43(2): 103−120 doi: 10.1007/s10851-011-0294-y
    [15] Grewenig S, Zimmer S, Weickert J. Rotationally invariant similarity measures for nonlocal image denoising. Journal of Visual Communication and Image Representation, 2011, 22(2): 117−130 doi: 10.1016/j.jvcir.2010.11.001
    [16] Wu Y, Tracey B, Natarajan P, Noonan J P. James-stein type center pixel weights for non-local means image denoising. IEEE Signal Processing Letters, 2013, 20(4), 411−414 doi: 10.1109/LSP.2013.2247755
    [17] Nguyen M P, Chun S Y. Bounded self-weights estimation method for non-local means image denoising using minimax estimators. IEEE Transactions on Image Processing, 2017, 26(4): 1637−1649 doi: 10.1109/TIP.2017.2658941
    [18] Salmon J. On two parameters for denoising with nonlocal means. IEEE Signal Processing Letters, 2010, 17(3), 269−272 doi: 10.1109/LSP.2009.2038954
    [19] Van D V D, Kocher M. SURE-based non-local means. IEEE Signal Processing Letters, 2009, 16(11), 973−976 doi: 10.1109/LSP.2009.2027669
    [20] Stein C M. Estimation of the mean of a multivariate normal distribution. The Annals of Statistics, 1981, 9(6), 1135−1151 doi: 10.1214/aos/1176345632
    [21] Dong W S, Zhang L, Shi G M, Li X. Nonlocally centralized sparse representation for image restoration. IEEE Transactions on Image Processing, 2013, 22(4): 1620−1630 doi: 10.1109/TIP.2012.2235847
    [22] May V, Keller Y, Sharon N, Shkolnisky Y. An algorithm for improving non-local means operators via low-rank approximation. IEEE Transactions on Image Processing, 2016, 25(3): 1340−1353 doi: 10.1109/TIP.2016.2518805
    [23] Dabov K, Foi A, Katkovnik V, Egiazarian K. Image denoising by sparse 3-D transform-domain collaborative filtering. IEEE Transactions on Image Processing, 2007, 16(8): 2080−2095 doi: 10.1109/TIP.2007.901238
    [24] Li X Y, Zhou Y C, Zhang J, Wang L H. Multipatch unbiased distance non-local adaptive means with wavelet shrinkage. IEEE Transactions on Image Processing, 2020, 29: 157−169 doi: 10.1109/TIP.2019.2928644
    [25] Wan L, Zeiler M D, Zhang S, Yann L C, Fergus R. Regularization of neural networks using dropconnect. In: Proceedings of the 2013 IEEE International Conference on Machine Learning. Atlanta, GA, USA: IEEE, 2013.1058−1066
    [26] 邢笑笑, 王海龙, 李健, 张选德. 渐近非局部平均图像去噪算法. 自动化学报, 2020, 46(9): 1952−1960

    Xing Xiao-Xiao, Wang Hai-Long, Li Jian, Zhang Xuan-De. Asymptotic non-local means image denoising algorithm. Acta Automatica Sinica, 2020, 46(9): 1952−1960
    [27] Chen Y J, Pock T. Trainable nonlinear reaction diffusion: a flexible framework for fast and effective image restoration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1256−1272 doi: 10.1109/TPAMI.2016.2596743
    [28] Zhang K, Zuo W M, Chen Y J, Meng D Y, Zhang L. Beyond a gaussian denoiser: residual learning of deep CNN for image denoising. IEEE Transactions on Image Processing, 2017, 26(7): 3142−3155 doi: 10.1109/TIP.2017.2662206
    [29] Chen J W, Chen J W, Cao H Y, Yang M. Image blind denoising with generative adversarial network based noise modeling. In: Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, Utah, USA: IEEE, 2018. 3155−3164
    [30] Krull A, Buchholz T O, Jug F. Noise2Void-learning denoising from single noisy images. In: Proceedings of the 2019 IEEE Conference on Computer Vision and Pattern Recognition. Long Beach, CA, USA: IEEE, 2019. 2129−2137
    [31] Roth S, Black M J. Fields of experts. International Journal of Computer Vision, 2009, 82(2): 205−229 doi: 10.1007/s11263-008-0197-6
    [32] Wang Z, Bovik A C, Sheikh H R, Simoncelli E P. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 2004, 13(4): 600−612 doi: 10.1109/TIP.2003.819861
    [33] Winnemoller H, Kyprianidis J E, Olsen S C. XDoG: an extended difference-of-gaussians compendium including advanced image stylization. Computers & Graphics, 2012, 36(6): 740−753
  • 加载中
图(9) / 表(6)
计量
  • 文章访问数:  1551
  • HTML全文浏览量:  469
  • PDF下载量:  447
  • 被引次数: 0
出版历程
  • 收稿日期:  2020-03-16
  • 录用日期:  2020-06-11
  • 网络出版日期:  2021-12-06
  • 刊出日期:  2022-01-25

目录

    /

    返回文章
    返回