Research on Image Denoising Based on Improved CycleGAN

doi:10.3778/j.issn.1002-8331.2209-0050

Abstract

Abstract: Image denoising research is a very critical technology in image processing. At present, most of the common image denoising methods require paired training data sets, and the generated denoising images are mostly different from the real images in terms of color distribution, edge connection and other details. Therefore, an image denoising method based on improved cycle-consistent adversarial networks is proposed. A multi-scale discriminator is introduced and a new target loss function is proposed in the improved network. In the new target loss function, pixel loss and feature loss are introduced and the L1 norm loss in the original network is replaced with the Smooth L1 norm loss. The experimental results show that the improved network has a certain improvement in performance compared with the original network. Compared with the original network, the peak signal-to-noise ratio of the image denoised by the improved network is increased from 25.24 dB to 29.02 dB, an increase of 15.0%. The structural similarity index is increased from 0.862 to 0.956, an increase of 10.9%.

Key words: cycle-consistent adversarial networks, loss function, multi-scale discriminator, image denoising, image processing, deep learning

摘要： 图像去噪研究是图像处理中非常关键的一项技术。目前，常见的图像去噪方法大部分都需要成对的训练数据集，并且所生成的去噪图像大都会在颜色分布、边缘衔接等细节信息上和真实图像存在一定的差异，因此提出了一种基于改进的循环生成对抗网络的图像去噪方法。这种改进方法在原网络的基础上引入了多尺度判别器并提出了新的目标损失函数。其中，新的损失函数引入了像素损失和特征损失，还用Smooth L1范数损失代替了原网络中的L1范数损失。实验结果表明，提出的改进网络相较于原网络的性能有一定提升。和原网络相比，利用改进后的网络进行去噪的图片峰值信噪比从25.24?dB提高到29.02?dB，提高了15.0%；结构相似性指数从0.862提高到0.956，提高了10.9%。

关键词: 循环生成对抗网络, 损失函数, 多尺度判别器, 图像去噪, 图像处理, 深度学习

FU Jin, HUANG Shan. Research on Image Denoising Based on Improved CycleGAN[J]. Computer Engineering and Applications, 2023, 59(17): 178-186.

伏锦, 黄山. 基于改进循环生成对抗网络的图像去噪研究[J]. 计算机工程与应用, 2023, 59(17): 178-186.

References

[1] BHARATI S，KHAN T Z，PODDER P，et al.A comparative analysis of image denoising problem：noise models，denoising filters and applications[M].Berlin：Springer，2020：49-66.
[2] LUO E，CHAN S H，NGUYEN T Q.Adaptive image denoising by mixture adaptation[J].IEEE Transactions on Image Processing，2016，25（10）：4489-4503.
[3] 刘迪，贾金露，赵玉卿，等.基于深度学习的图像去噪方法研究综述[J].计算机工程与应用，2021，57（7）：1-13.
LIU D，JIA J L，ZHAO Y Q，et al.Research review of image denoising methods based on deep learning[J].Computer Engineering and Applications，2021，57（7）：1-13.
[4] KIM D W，RYUN CHUNG J，JUNG S W.GRDN：grouped residual dense network for real image denoising and gan-based real-world noise modeling[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2019.
[5] WOO S，PARK J，LEE J Y，et al.CBAM：convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision，2018：3-19.
[6] CHEN J，CHAO H.Image blind denoising with generative adversarial network based noise modeling[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2018：3155-3164.
[7] GOODFELLOW I J，POUGET-ABADIE J，MIRZA M，et al.Generative adversarial networks[C]//Advances in Neural Information Processing Systems，2014：2672-2680.
[8] LIN K，LI T H，LIU S，et al.Real photographs denoising with noise domain adaptation and attentive generative adversarial network[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2019.
[9] LI Z H，HUANG J Z，YU L F，et al.Low-dose CT image denoising using cycle-consistent adversarial networks[C]//Proceedings of the IEEE Conference on Nuclear Science Symposium and Medical Imaging，2019：1-3.
[10] ZHU J Y，PARK T，ISOLA P，et al.Unpaired image-to-image translation using cycle-consistent adversarial networks[J].arXiv：1703.10593，2017.
[11] ZHOU L，SCHAEFFERKOETTER J D，THAM I，et al.Supervised learning with CycleGAN for low-dose FDG PET image denoising[J].Medical Image Analysis，2020，65（7）：101770.
[12] HUANG Z，CHEN Z，ZHANG Q，et al.CaGAN：a cycle-consistent generative adversarial network with attention for low-dose CT imaging[J].IEEE Transactions on Computational Imaging，2020，6：1203-1218.
[13] LI Z，ZHOU S，HUANG J，et al.Investigation of low-dose CT image denoising using unpaired deep learning methods[J].IEEE Transactions on Radiation and Plasma Medical Sciences，2020，5（2）：224-234.
[14] TAN C，YANG M，YOU Z，et al.A selective kernel-based cycle-consistent generative adversarial network for unpaired low-dose CT denoising[J].Precision Clinical Medicine，2022，5（2）：12.
[15] TAIGMAN Y，POLYAK A，WOLF L.Unsupervised cross-domain image generation[J].arXiv：1611.02200，2016.
[16] PATHAK D，KPAHENBUHL P，DONAHUE J，et al.Contex encoders：feature learning by impainting[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Piscataway：IEEE，2016：2536-2541.