[1]卢娟,陈纯毅.融合递归自编解码器的蒙卡画面重构降噪方法[J].智能系统学报,2023,18(3):459-467.[doi:10.11992/tis.202112011]
LU Juan,CHEN Chunyi.A denoised method by fusing recursive auto-encoder decorder for Monte Carlo rendering image reconstruction[J].CAAI Transactions on Intelligent Systems,2023,18(3):459-467.[doi:10.11992/tis.202112011]
点击复制
LU Juan,CHEN Chunyi.A denoised method by fusing recursive auto-encoder decorder for Monte Carlo rendering image reconstruction[J].CAAI Transactions on Intelligent Systems,2023,18(3):459-467.[doi:10.11992/tis.202112011]
融合递归自编解码器的蒙卡画面重构降噪方法
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
18
期数:
2023年第3期
页码:
459-467
栏目:
学术论文—机器学习
出版日期:
2023-07-05
- Title:
- A denoised method by fusing recursive auto-encoder decorder for Monte Carlo rendering image reconstruction
- Keywords:
- recursive residual network; multi-scale convolution; generative adversarial networks; Monte Carlo rendering screen; image processing; image denoising; deep learning; auto-encoder network
- 分类号:
- TP183
- 摘要:
- 针对生成对抗模型降噪结果出现的伪影和模糊问题，提出一种基于对抗生成递归自编码器的蒙卡渲染画面降噪方法。在模型上，设计多尺度卷积编码结构，以多尺度残差自编码模型为生成器，通过组合连接实现不同层次特征提取，融合不同感受野的特征信息。以设计的递归残差网络模型为判别器，判断蒙卡渲染画面真伪，在对抗中提高网络性能。以端到端的方式将辅助信息特征、含有噪声的蒙卡画面和8192采样率下的画面输入到融合递归自编解码器中进行降噪处理。实验表明，该方法在测试场景下的平均峰值信噪比为32.44 dB，比生成对抗网络方法和残差网络方法分别提升4.80%和3.13%；平均结构相似性为0.92，比2种已有的算法分别提高2.54%和1.01%。
- Abstract:
- Aiming at the problem of artifacts and blurring in the noise reduction results of generative adversarial model, a noise reduction method for Monte Carlo rendering image based on generative adversarial recurrent auto-encoder is proposed in this paper. On the model, the multi-scale convolutional coding structure is designed, and the multi-scale residual auto-encoder model is used as the generator. The feature extraction at different levels is realized by combination connection, fusing the feature information of different receptive fields. The designed recursive residual network model is used as the discriminator to judge the authenticity of the Monte Carlo rendering picture and improve the network performance in the confrontation. In an end-to-end way, the auxiliary information features, the noise-containing Monte Carlo pictures and the pictures with the sampling rate of 8192 are input into the recursive auto-encoder decoder for noise reduction. Experiments show that the average peak signal-to-noise ratio of this method is 32.44 dB in the test scenario, which is 4.80% and 3.13% higher than the generative adversarial network method and the residual network method, respectively. The average structural similarity is 0.92, which is 2.54 % and 1.01 % higher than the existing two algorithms, respectively.
- 参考文献/References:
-
[1] LEHTINEN J, KARRAS T, LAINE S, et al. Gradient-domain metropolis light transport[J]. ACM transactions on graphics, 2013, 32(4): 1–12.
[2] ROUSSELLE F, KNAUS C, ZWICKER M. Adaptive rendering with non-local means filtering[J]. ACM transactions on graphics, 2012, 31(6): 1–11.
[3] KALANTARI N K, BAKO S, SEN P. A machine learning approach for filtering Monte Carlo noise[J]. ACM transactions on graphics, 2015, 34(4): 1–12.
[4] MENG Xiaoxu, ZHENG Quan, Varshney A, et al. Real-time Monte Carlo denoising with the neural bilateral grid[C]//Eurographics Symposium on Rendering 2020. London: ESRL, 2020: 13-24.
[5] CHAITANYA C R A, KAPLANYAN A S, SCHIED C, et al. Interactive reconstruction of Monte Carlo image sequences using a recurrent denoising autoencoder[J]. ACM transactions on graphics, 2017, 36(4): 1–12.
[6] 谢川, 王勇超, 林志洁, 等. 基于对抗生成网络的蒙特卡罗噪声去除算法[J]. 模式识别与人工智能, 2018, 31(11): 1047–1060
XIE Chuan, WANG Yongchao, LIN Zhijie, et al. Monte Carlo noise removal algorithm based on adversarial generative network[J]. Pattern recognition and artificial intelligence, 2018, 31(11): 1047–1060
[7] VOGELS T, NOVáK J, ROUSSELLE F, et al. Kerneel predicting convolutional neural networks for denoising[P]. US20180293711: 15/814190, 2018-11-10.
[8] WONG K M, WONG T T. Deep residual learning for denoising Monte Carlo renderings[J]. Computational visual media, 2019, 5(3): 239–255.
[9] XU Bing, ZHANG Junfei, WANG Rui, et al. Adversarial Monte Carlo denoising with conditioned auxiliary feature modulation[J]. ACM transactions on graphics, 2019, 38(6): 1–12.
[10] VICINI D, ADLER D, NOVáK J, et al. Denoising deep Monte Carlo renderings[J]. Computer graphics forum, 2019, 38(1): 316–327.
[11] HUO Yuchi, YOON S E. A survey on deep learning-based Monte Carlo denoising[J]. Computational visual media, 2021, 7(2): 169–185.
[12] 王婕, 罗静蕊, 岳广德. 一种改进的多尺度融合并行稠密残差去噪网络[J]. 小型微型计算机系统, 2021, 42(4): 798–804
WANG Jie, LUO Jingrui, YUE Guangde. Improved parallel residual dense denoising network based on multi-scale fusion[J]. Journal of Chinese computer systems, 2021, 42(4): 798–804
[13] RONNEBERGER O. Invited talk: U-net convolutional networks for biomedical image segmentation[C]//Bildverarbeitung für die Medizin 2017. Berlin: Springer Vieweg, 2017: 3?3.
[14] SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-ResNet and the impact of residual connections on learning[C]//Proceedings of the AAAI Conference on Artificial Intelligence. San Francisco: AAAI Press, 2017, 31(1): 11231.
[15] WANG Tianyang, SUN Mingxuan, HU Kaoning. Dilated deep residual network for image denoising[C]//2017 IEEE 29th International Conference on Tools with Artificial Intelligence. Boston: IEEE, 2018: 1272?1279.
[16] BENEDIKT Bitterli. Rendering resources[EB/OL]. [2021?12?06].https://benedikt-bitterli.me/resources/.
[17] LIN Weiheng, WANG Beibei, WANG Lu, et al. A detail preserving neural network model for Monte Carlo denoising[J]. Computational visual media, 2020, 6(2): 157–168.
[18] SIMONYAN K, ZISSERMAN A. Two-stream convolutional networks for action recognition in videos[EB/OL]. (2014?06?09) [2021?12?06].https://arxiv.org/abs/1406.2199.
[19] KREINOVICH V, NGUYEN H, OUNCHAROEN R. How to estimate forecasting quality: a system- motivated derivation of symmetric mean absolute percentage error and other similar characteristics[EB/OL]. [2021?12?06]. http://www.cs.utep.edu/vladik/2014/tr14-53.pdf.
[20] ISOLA P, ZHU Junyan, ZHOU Tinghui, et al. Image-to-image translation with conditional adversarial networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 5967?5976.
[21] 许家荣. 核预测与GAN特征保持的蒙特卡罗渲染图像去噪[D]. 广州: 华南理工大学, 2020.
XU Jiarong. Denoising of Monte Carlo renderings based on kernel prediction and feature-preserved GAN[D]. Guangzhou: South China University of Technology, 2020.
[22] 佟雨兵, 张其善, 祁云平. 基于PSNR与SSIM联合的图像质量评价模型[J]. 中国图象图形学报, 2006, 11(12): 1758–1763
TONG Yubing, ZHANG Qishan, QI Yunping. Image quality assessing by combining PSNR and SSIM[J]. Journal of image and graphics, 2006, 11(12): 1758–1763
[23] 林椹尠, 张梦凯, 吴成茂. 利用残差密集网络的运动模糊复原方法[J]. 智能系统学报, 2021, 16(3): 442–448
LIN Zhenxian, ZHANG Mengkai, WU Chengmao. Image restoration with residual dense network[J]. CAAI transactions on intelligent systems, 2021, 16(3): 442–448
备注/Memo
收稿日期:2021-12-06。
基金项目:吉林省科技发展计划项目(20190302113GX).
作者简介:卢娟,硕士研究生,主要研究方向为深度学习、图像降噪;陈纯毅,教授,博士生导师,博士,中国图象图形学学会虚拟现实专委会委员、中国计算机学会会员。主要研究方向为虚拟现实、特种电影和光电系统仿真;承担过国家自然科学基金、国家重点研发计划、国家科技支撑计划、863计划、973计划等国家高层次项目,发表学术论文60余篇
通讯作者:陈纯毅.E-mail:2306915844@qq.com
更新日期/Last Update:
1900-01-01