面向图像去噪的深度双层群稀疏编码网络

doi:10.3778/j.issn.1002-8331.2206-0216

摘要/Abstract

摘要： 当前基于深度学习的图像去噪方法主要是利用深度神经网络将噪声图像直接映射到干净图像，忽略了图像去噪任务知识。针对该问题，提出了一种基于双层群稀疏编码的深度图像去噪网络。为了充分利用图像中相似结构以及对相似块之间的特异性有效表示，提出了双层群稀疏编码图像去噪模型，并表示成ℓ2,1 - ℓ1范数优化问题。利用算法展开技术将所提去噪模型的优化解转化成“端到端”的深度神经网络。为了进一步提高网络训练的稳定性，所提去噪网络中引入一种改进的残差连接。在BSD68、Set12、CBSD68、Kodak24和Urban100等常用数据集上的实验结果表明，所提算法在主观视觉质量和客观评价指标上优于一些主流的去噪方法。特别地，针对噪声等级为75，所提算法在CBSD68数据集上比经典的DnCNN算法平均PSNR指标提高了1.3 dB。

关键词: 图像去噪, 群稀疏编码, 深度神经网络, 算法展开, 残差连接

Abstract: The current deep learning based image denoising methods directly map the noisy image to its clean one using by deep neural network, which ignore the domain knowledge of image denoising task. According to this problem, this paper proposes a two-layer group sparse coding based image denoising network. Firstly, to effectively exploit the similarity structure of image and represent the specificity among similarpatterns, a two-layer group sparse coding is developed for image denoising, modeled as ℓ2,1 - ℓ1-norm optimization problem. Then, the optimization solution of this denoising model is converted into an end-to-end deep neural network through algorithm unrolling. Moreover, a modified residual connection is applied to further improve the stability of network training. The experimental results on BSD68, Set12, CBSD68, Kodak24 and Urban100 datasets demonstrate that the proposed method outperforms some popular denoising methods. Specifically, the proposed method achieves 1.3 dB average PSNR gain over the classical DnCNN on CBSD68 at noise level 75.

Key words: image denoising, group sparse coding, deep neural network, algorithm unrolling, residual connection

方祯煜, 尹海涛. 面向图像去噪的深度双层群稀疏编码网络[J]. 计算机工程与应用, 2023, 59(21): 195-203.

FANG Zhenyu, YIN Haitao. Deep Two-Layer Group Sparse Coding Network for Image Denoising[J]. Computer Engineering and Applications, 2023, 59(21): 195-203.

参考文献

[1] ZHUANG L，BIOUCAS-DIAS J M.Fast hyperspectral image denoising and inpainting based on low-rank and sparse representations[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing，2018，11（3）：730-742.
[2] SHARIF S M A，NAQVI R A，BISWAS M.Learning medical image denoising with deep dynamic residual attention network[J].Mathematics，2020，8（12）：2192.
[3] 唐权华，雷金娥，周艳，等.一种时空联合的视频去噪方法[J].计算机工程与应用，2010，46（6）：163-165.
TANG Q H，LEI J E，ZHOU Y，et al.Spatial-temporal method for video denoising[J].Computer Engineering and Applications，2010，46（6）：163-165.
[4] PRONINA V，KOKKINOS F，DYLOV D V，et al.Microscopy image restoration with deep wiener-kolmogorov filters[C]//European Conference on Computer Vision，2020：185-201.
[5] GAVASKAR R G，CHAUDHURY K N.Fast adaptive bilateral filtering[J].IEEE Transactions on Image Processing，2018，28（2）：779-790.
[6] 祝严刚，张桂梅.一种改进的非局部均值图像去噪算法[J].计算机工程与应用，2017，53（18）：192-198.
ZHU Y G，ZHANG G M.Improved non-local means denoising algorithm[J].Computer Engineering and Applications，2017，53（18）：192-198.
[7] DABOV K，FOI A，KATKOVNIK V，et al.Image denoising by sparse 3-D transform-domain collaborative filtering[J].IEEE Transactions on Image Processing，2007，16（8）：2080-2095.
[8] CHAN T F，SHEN J，ZHOU H M.Total variation wavelet inpainting[J].Journal of Mathematical Imaging and Vision，2006，25（1）：107-125.
[9] ElAD M，AHARON M.Image denoising via sparse and redundant representations over learned dictionaries[J].IEEE Transactions on Image Processing，2006，15（12）：3736-3745.
[10] KE Z，HUANG W，CUI Z X，et al.Learned low-rank priors in dynamic MR imaging[J].IEEE Transactions on Medical Imaging，2021，40（12）：3698-3710.
[11] ZHANG K，ZUO W，CHEN Y，et al.Beyond a Gaussian denoiser：residual learning of deep CNN for image denoising[J].IEEE Transactions on Image Processing，2017，26（7）：3142-3155.
[12] ZHANG K，ZUO W，ZHANG L.FFDNet：toward a fast and flexible solution for CNN-based image denoising[J].IEEE Transactions on Image Processing，2018，27（9）：4608-4622.
[13] TIAN C，XU Y，LI Z，et al.Attention-guided CNN for image denoising[J].Neural Networks，2020，124：117-129.
[14] HU X，MA R，LIU Z，et al.Pseudo 3D auto-correlation network for real image denoising[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2021：16175-16184.
[15] DONG G，ZHANG Y，XIONG Z.Spatial hierarchy aware residual pyramid network for time-of-flight depth denoising[C]//European Conference on Computer Vision，2020：35-50.
[16] CHEN H，WANG Y，GUO T，et al.Pre-trained image processing transformer[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2021：12299-12310.
[17] MONGA V，LI Y，ELDAR Y C.Algorithm unrolling：interpretable，efficient deep learning for signal and image processing[J].IEEE Signal Processing Magazine，2021，38（2）：18-44.
[18] GREGOR K，LECUN Y.Learning fast approximations of sparse coding[C]//International Conference on Machine Learning，2010：399-406.
[19] ZHANG J，GHANEM B.ISTA-Net：interpretable optimization-inspired deep network for image compressive sensing[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2018：1828-1837.
[20] SIMON D，ELAD M.Rethinking the CSC model for natural images[C]//Advances in Neural Information Processing Systems，2019：2274-2284.
[21] LECOUAT B，PONCE J，MAIRAL J.Fully trainable and interpretable non-local sparse models for image restoration[C]//European Conference on Computer Vision，2020：238-254.
[22] QIAN Q，XIONG F，ZHOU J.Deep unfolded iterative shrinkage-thresholding model for hyperspectral unmixing[C]//IEEE International Geoscience and Remote Sensing Symposium，2019：2151-2154.
[23] LI Y，TOFIGHI M，GENG J，et al.Efficient and interpretable deep blind image deblurring via algorithm unrolling[J].IEEE Transactions on Computational Imaging，2020：666-681.
[24] ZHANG J，ZHAO D，GAO W.Group-based sparse representation for image restoration[J].IEEE Transactions on Image Processing，2014，23（8）：3336-3351.
[25] FAN Y，YU J，MEI Y，et al.Neural sparse representation for imagerestoration[C]//Advances in Neural Information Processing Systems，2020：15394-15404.