A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses
Abstract
:1. Introduction
2. Materials and Methods
2.1. Overall Architecture
2.2. Architecture of the Proposed Network
2.2.1. Generator
2.2.2. Discriminator
2.3. GAN-Based Heterogeneous Losses Function
3. Simulation Results
3.1. Comparisons of Denoising Quality
3.2. Comparisons of Computational Complexity
4. Conclusions
Author Contributions
Funding
Conflicts of Interest
References
- Zhang, K.; Zuo, W.; Gu, S.; Zhang, L. Learning deep CNN denoiser prior for image restoration. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 22–25 July 2017; pp. 2808–2817. [Google Scholar]
- Perona, P.; Malik, J. Scale-space and edge detection using anisotropic diffusion. IEEE Trans. Pattern Anal. Mach. Intell. 1990, 12, 629–639. [Google Scholar] [CrossRef] [Green Version]
- Rudin, L.I.; Osher, S.; Fatemi, E. Nonlinear total variation based noise removal algorithms. Phys. D Nonlinear Phenom. 1992, 60, 259–268. [Google Scholar] [CrossRef]
- Tomasi, C.; Manduchi, R. Bilateral filtering for gray and color images. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), Bombay, India, 4–7 January 1998; pp. 839–846. [Google Scholar]
- Buades, A.; Coll, B.; Morel, J.M. A non-local algorithm for image denoising. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), San Diego, CA, USA, 20–26 June 2005; pp. 60–65. [Google Scholar]
- Dabov, K.; Foi, A.; Katkovnik, V.; Egiazarian, K. Image denoising by sparse 3-D transform-domain collaborative filtering. IEEE Trans. Image Process. 2007, 16, 2080–2095. [Google Scholar] [CrossRef] [PubMed]
- Gu, S.; Zhang, L.; Zuo, W.; Feng, X. Weighted nuclear norm minimization with application to image denoising. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Columbus, OH, USA, 24–27 June 2014; pp. 2862–2869. [Google Scholar]
- Xu, J.; Zhang, L.; Zhang, D.; Feng, X. Multi-channel Weighted Nuclear Norm Minimization for Real Color Image Denoising. In Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 22–29 October 2017; pp. 1105–1113. [Google Scholar]
- Burger, H.C.; Schuler, C.J.; Harmeling, S. Image denoising: Can plain neural networks compete with BM3D? In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Providence, RI, USA, 18–20 June 2012; pp. 2392–2399. [Google Scholar]
- Zhang, K.; Zuo, W.; Chen, Y.; Meng, D.; Zhang, L. Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising. IEEE Trans. Image Process. 2017, 26, 3142–3155. [Google Scholar] [CrossRef] [Green Version]
- Tai, Y.; Yang, J.; Liu, X.; Xu, C. MemNet: A Persistent Memory Network for Image Restoration. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 22–29 October 2017; pp. 4549–4557. [Google Scholar]
- Zhang, K.; Zuo, W.; Zhang, L. Ffdnet: Toward a fast and flexible solution for cnn-based image denoising. IEEE Trans. Image Process. 2018, 27, 4608–4622. [Google Scholar] [CrossRef] [Green Version]
- Guo, S.; Yan, Z.; Zhang, K.; Zuo, W.; Zhang, L. Toward convolutional blind denoising of real photographs. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, CA, USA, 16–20 June 2019; pp. 1712–1722. [Google Scholar]
- Chen, J.; Chen, J.; Chao, H.; Yang, M. Image blind denoising with generative adversarial network based noise modeling. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, USA, 19–21 June 2018; pp. 3155–3164. [Google Scholar]
- Yin, J.; Chen, B.; Li, Y. Highly Accurate Image Reconstruction for Multimodal Noise Suppression Using Semisupervised Learning on Big Data. IEEE Trans. Multimed. 2018, 20, 3045–3056. [Google Scholar] [CrossRef]
- Hou, X.; Luo, H.; Liu, J.; Xu, B.; Sun, K.; Gong, Y.; Liu, B.; Qiu, G. Learning Deep Image Priors for Blind Image Denoising. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Long Beach, CA, USA, 16–20 June 2019; pp. 1738–1747. [Google Scholar]
- Goodfellow, I.; Pouget-Abadie, J.; Mehdi, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative adversarial nets. In Proceedings of the Advances in Neural Information Processing Systems, Montreal, QC, Canada, 8–13 December 2014. [Google Scholar]
- Sifre, L. Rigid-Motion Scattering for Image Classification. Ph.D. Thesis, Ecole Polytechnique University, Palaiseau, France, 2014. [Google Scholar]
- Cho, S.I.; Kang, S.-J. Gradient Prior-aided CNN Denoiser with Separable Convolution-based Optimization of Feature Dimension. IEEE Trans. Multimed. 2019, 21, 484–493. [Google Scholar] [CrossRef]
- Ioffe, S.; Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In Proceedings of the International Conference on Machine Learning (ICML), Lille, France, 6–11 July 2015; pp. 448–456. [Google Scholar]
- Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolutional neural networks. In Advances in Neural Information Processing Systems; MIT Press: Lake Tahoe, NV, USA, 2012. [Google Scholar]
- Yu, F.; Koltun, V. Multi-scale context aggregation by dilated convolutions. In Proceedings of the International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, 2–4 May 2016. [Google Scholar]
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 20 June–1 July 2016; pp. 770–778. [Google Scholar]
- CIPR Database. Available online: http://www.cipr.rpi.edu/resource/stills/index.html (accessed on 19 January 2021).
- Jones, K. Methods of measurement for the power consumption of audio, video and related equipment. In ENERGY STAR Program. Requirements for Displays; ENERGY STAR Program Requirements for Displays: Washington, DC, USA, 2008. [Google Scholar]
- Football Sequences. Available online: http://media.xiph.org/video/derf/ (accessed on 19 January 2021).
- Cho, S.I.; Kang, S.-J. Geodesic path-based diffusion acceleration for image denoising. IEEE Trans. Multimed. 2018, 20, 1738–1750. [Google Scholar] [CrossRef]
- Kingma, D.; Ba, J. A method for stochastic optimization. In Proceedings of the International Conference on Learning Representations (ICLR), San Diego, CA, USA, 7–9 May 2015. [Google Scholar]
- Fowlkes, C.C.; Martin, D.R.; Malik, J. Local figure–ground cues are valid for natural images. J. Vis. 2007, 7, 1–9. [Google Scholar] [CrossRef] [Green Version]
- Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Miami, FL, USA, 20–25 June 2009; pp. 248–255. [Google Scholar]
- Ma, K.; Duanmu, Z.; Wu, Q.; Wang, Z.; Yong, H.; Li, H.; Zhang, L. Waterloo exploration database: New challenges for image quality assessment models. IEEE Trans. Image Process. 2017, 26, 1004–1016. [Google Scholar] [CrossRef]
- Abadi, M.; Agarwal, A.; Barham, P.; Brevdo, E.; Chen, Z.; Citro, C.; Corrado, G.S.; Davis, A.; Dean, J.; Devin, M.; et al. TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems. arXiv 2016, arXiv:1603.04467. [Google Scholar]
- Wang, Z.; Bovik, A.C.; Sheikh, H.R.; Simoncelli, E.P. Image quality assessment: From error visibility to structural similarity. IEEE Trans. Image Process. 2004, 13, 600–612. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Sheikh, H.R.; Bovik, A.C. Image information and visual quality. IEEE Trans. Image Process. 2006, 15, 430–444. [Google Scholar] [CrossRef] [PubMed]
- Zhang, L.; Zhang, L.; Mou, X.; Zhang, D. FSIM: A feature similarity index for image quality assessment. IEEE Trans. Image Process. 2011, 20, 2378–2386. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Li, T.H.; Li, Z.; Han, T.Y.; Rahardja, S.; Yeo, C. A perceptually relevant MSE-based image quality. IEEE Trans. Image Process. 2013, 22, 4447–4459. [Google Scholar]
- Shao, L.; Yan, R.; Li, X.; Liu, Y. From heuristic optimization to dictionary learning: A review and comprehensive comparison of image denoising algorithms. IEEE Trans. Cybern. 2014, 44, 1001–1013. [Google Scholar] [CrossRef]
- Kang, L.-W.; Lin, C.-W.; Fu, Y.-H. Automatic single-image-based rain streaks removal via image decomposition. IEEE Trans. Image Process. 2012, 21, 1742–1755. [Google Scholar] [CrossRef]
- Huang, T.-H. Enhancement of Backlight-Scaled Images. IEEE Trans. Image Process. 2013, 22, 4587–4597. [Google Scholar] [CrossRef]
- Li, S.; Kang, X. Fast multi-exposure image fusion with median filter and recursive filter. IEEE Trans. Consum. Electron. 2012, 58, 626–632. [Google Scholar] [CrossRef] [Green Version]
- Zhang, Y.; Zhang, Y.; Zhang, J.; Dai, Q. CCR: Clustering and collaborative representation for fast single image super-resolution. IEEE Trans. Multimed. 2016, 18, 405–417. [Google Scholar] [CrossRef]
- Jiang, J.; Ma, X.; Chen, C.; Lu, T.; Wang, Z.; Ma, J. Single image super-resolution via locally regularized anchored neighborhood regression and nonlocal means. IEEE Trans. Multimed. 2017, 19, 15–26. [Google Scholar] [CrossRef]
- Du, B.; Zhang, M.; Zhang, L.; Hu, R.; Tao, D. PLTD: Patch-based low-rank tensor decomposition for hyperspectral images. IEEE Trans. Multimed. 2017, 19, 67–79. [Google Scholar] [CrossRef]
- Liu, X.; Zhao, D.; Xiong, R.; Ma, S.; Gao, W.; Sun, H. Image interpolation via regularized local linear regression. IEEE Trans. Image Process. 2011, 20, 3455–3469. [Google Scholar] [PubMed] [Green Version]
Layer | Operations | Dimension [SR, SC, FD1, FD2] |
---|---|---|
1st layer | P_conv + D_conv + bias + ReLU | P_conv: 1 × 1 × 24 × 96 |
D_conv: 3 × 3 × 96 × 1 | ||
2nd~12th layers | P_conv + K dilated D_conv + BN + ReLU | P_conv: 1 × 1 × 96 × 96 |
D_conv:3 × 3 × 96 × 1 | ||
13th layer | Conv + bias | Conv: 3 × 3 × 96 × 3 |
Total number of weights | 116,640 (62% of the number of weights for IRCNN [1]) |
Layer | Operations | Dimension [SR, SC, FD1, FD2] |
---|---|---|
1st layer | P_conv + D_conv + bias + ReLU | P_conv: 1 × 1 × 24 × 32 |
D_conv: 3 × 3 × 32 × 1 | ||
2nd~12th layers | P_conv + K dilated D_conv + BN + ReLU | P_conv: 1 × 1 × 32 × 32 |
D_conv:3 × 3 × 32 × 1 | ||
13th layer | Conv + bias | Conv: 3 × 3 × 32 × 3 |
14th layer | 1024 dense (fully connected) | 1 N × 1024 |
15th layer | 1024 dense + sigmoid | 1024 × 1 |
Noise Level | σn = 15 | σn = 25 | σn = 35 | ||||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Image Set (Number of Image Set) | Kodak (24) | CIPR_M (14) | CIPR_C (18) | IEC (20) | Football (90) | CBSD (68) | AVG | Kodak (24) | CIPR_M (14) | CIPR_C (18) | IEC (20) | Football (90) | CBSD (68) | AVG | Kodak (24) | CIPR_M (14) | CIPR_C (18) | IEC (20) | Football (90) | CBSD (68) | AVG | ||
Noisy images | PSNR [dB] | 24.610 | 24.607 | 24.607 | 24.609 | 24.611 | 24.609 | 24.609 | 20.172 | 20.177 | 20.170 | 20.172 | 20.174 | 20.172 | 20.173 | 17.249 | 17.253 | 17.251 | 17.250 | 17.249 | 17.249 | 17.250 | |
SSIM | 0.682 | 0.654 | 0.627 | 0.606 | 0.663 | 0.726 | 0.660 | 0.485 | 0.455 | 0.418 | 0.391 | 0.449 | 0.544 | 0.457 | 0.360 | 0.334 | 0.298 | 0.270 | 0.319 | 0.421 | 0.334 | ||
VIF | 0.542 | 0.553 | 0.533 | 0.535 | 0.528 | 0.559 | 0.542 | 0.382 | 0.395 | 0.377 | 0.379 | 0.367 | 0.397 | 0.383 | 0.292 | 0.307 | 0.291 | 0.296 | 0.279 | 0.306 | 0.295 | ||
FSIM | 0.959 | 0.894 | 0.953 | 0.949 | 0.875 | 0.886 | 0.919 | 0.907 | 0.808 | 0.895 | 0.887 | 0.769 | 0.792 | 0.843 | 0.854 | 0.735 | 0.834 | 0.826 | 0.681 | 0.716 | 0.774 | ||
Model-based optimization methods | NLMC | PSNR [dB] | 31.568 | 32.507 | 33.724 | 34.808 | 32.693 | 30.472 | 32.629 | 28.933 | 30.087 | 31.141 | 31.979 | 30.325 | 27.961 | 30.071 | 27.231 | 28.303 | 29.220 | 29.955 | 28.683 | 26.394 | 28.297 |
SSIM | 0.887 | 0.916 | 0.932 | 0.935 | 0.880 | 0.886 | 0.906 | 0.812 | 0.870 | 0.889 | 0.891 | 0.819 | 0.805 | 0.847 | 0.760 | 0.828 | 0.847 | 0.850 | 0.779 | 0.750 | 0.802 | ||
VIF | 0.472 | 0.529 | 0.514 | 0.518 | 0.435 | 0.490 | 0.493 | 0.327 | 0.404 | 0.388 | 0.386 | 0.310 | 0.349 | 0.361 | 0.255 | 0.332 | 0.318 | 0.317 | 0.253 | 0.276 | 0.292 | ||
FSIM | 0.972 | 0.963 | 0.981 | 0.977 | 0.921 | 0.922 | 0.956 | 0.945 | 0.939 | 0.963 | 0.957 | 0.883 | 0.867 | 0.926 | 0.925 | 0.921 | 0.946 | 0.938 | 0.869 | 0.839 | 0.906 | ||
BM3DC | PSNR [dB] | 34.415 | 33.997 | 35.793 | 37.745 | 35.444 | 33.513 | 35.151 | 31.824 | 31.945 | 33.697 | 35.298 | 32.976 | 30.705 | 32.741 | 30.044 | 30.436 | 32.074 | 33.489 | 31.282 | 28.880 | 31.034 | |
SSIM | 0.934 | 0.940 | 0.958 | 0.964 | 0.943 | 0.937 | 0.946 | 0.893 | 0.914 | 0.938 | 0.942 | 0.904 | 0.890 | 0.913 | 0.853 | 0.889 | 0.917 | 0.920 | 0.866 | 0.845 | 0.882 | ||
VIF | 0.589 | 0.598 | 0.597 | 0.603 | 0.570 | 0.613 | 0.595 | 0.450 | 0.480 | 0.481 | 0.482 | 0.430 | 0.467 | 0.465 | 0.357 | 0.398 | 0.401 | 0.400 | 0.339 | 0.372 | 0.378 | ||
FSIM | 0.985 | 0.975 | 0.988 | 0.987 | 0.965 | 0.961 | 0.977 | 0.971 | 0.960 | 0.978 | 0.975 | 0.938 | 0.932 | 0.959 | 0.954 | 0.946 | 0.968 | 0.963 | 0.912 | 0.903 | 0.941 | ||
WNNM | PSNR [dB] | 32.484 | 33.657 | 34.932 | 36.286 | 33.847 | 31.272 | 33.746 | 30.117 | 31.446 | 32.647 | 33.858 | 31.454 | 28.772 | 31.383 | 28.653 | 29.997 | 31.026 | 32.172 | 29.953 | 27.276 | 29.846 | |
SSIM | 0.909 | 0.933 | 0.949 | 0.952 | 0.914 | 0.907 | 0.927 | 0.858 | 0.905 | 0.924 | 0.925 | 0.862 | 0.849 | 0.887 | 0.817 | 0.880 | 0.900 | 0.900 | 0.824 | 0.800 | 0.854 | ||
VIF | 0.527 | 0.577 | 0.543 | 0.537 | 0.487 | 0.552 | 0.537 | 0.391 | 0.457 | 0.429 | 0.418 | 0.350 | 0.408 | 0.409 | 0.310 | 0.383 | 0.358 | 0.346 | 0.273 | 0.323 | 0.332 | ||
FSIM | 0.978 | 0.971 | 0.983 | 0.980 | 0.943 | 0.940 | 0.966 | 0.957 | 0.954 | 0.970 | 0.964 | 0.904 | 0.899 | 0.941 | 0.934 | 0.938 | 0.957 | 0.948 | 0.874 | 0.864 | 0.919 | ||
MC-WNNM | PSNR [dB] | 33.943 | 34.022 | 35.716 | 37.120 | 34.872 | 32.918 | 34.765 | 31.367 | 31.890 | 33.497 | 34.620 | 32.231 | 30.245 | 32.308 | 29.726 | 30.484 | 31.917 | 32.879 | 30.593 | 28.564 | 30.694 | |
SSIM | 0.931 | 0.939 | 0.956 | 0.959 | 0.933 | 0.933 | 0.942 | 0.882 | 0.910 | 0.932 | 0.932 | 0.880 | 0.881 | 0.903 | 0.839 | 0.886 | 0.909 | 0.907 | 0.837 | 0.833 | 0.869 | ||
VIF | 0.579 | 0.597 | 0.586 | 0.577 | 0.538 | 0.604 | 0.580 | 0.434 | 0.477 | 0.467 | 0.452 | 0.386 | 0.455 | 0.445 | 0.345 | 0.402 | 0.392 | 0.374 | 0.300 | 0.363 | 0.363 | ||
FSIM | 0.983 | 0.974 | 0.987 | 0.984 | 0.958 | 0.959 | 0.974 | 0.965 | 0.957 | 0.975 | 0.970 | 0.921 | 0.924 | 0.952 | 0.946 | 0.942 | 0.964 | 0.955 | 0.890 | 0.892 | 0.931 | ||
Discriminativelearning methods (CNN-based methods | MLP | PSNR [dB] | - | - | - | - | - | - | - | 31.329 | 31.397 | 32.838 | 33.918 | 31.618 | 29.135 | 31.706 | 29.942 | 29.989 | 31.400 | 32.422 | 30.220 | 27.647 | 30.270 |
SSIM | - | - | - | - | - | - | - | 0.881 | 0.908 | 0.925 | 0.929 | 0.875 | 0.874 | 0.899 | 0.845 | 0.886 | 0.904 | 0.905 | 0.841 | 0.830 | 0.869 | ||
VIF | - | - | - | - | - | - | - | 0.378 | 0.446 | 0.392 | 0.368 | 0.366 | 0.415 | 0.394 | 0.304 | 0.373 | 0.323 | 0.299 | 0.288 | 0.331 | 0.320 | ||
FSIM | - | - | - | - | - | - | - | 0.909 | 0.937 | 0.935 | 0.930 | 0.904 | 0.913 | 0.921 | 0.881 | 0.920 | 0.919 | 0.912 | 0.877 | 0.882 | 0.899 | ||
Dn- CNNC | PSNR [dB] | 34.592 | 32.738 | 35.117 | 37.524 | 35.072 | 33.885 | 34.822 | 32.142 | 31.306 | 33.337 | 35.304 | 32.868 | 31.224 | 32.697 | 30.572 | 30.220 | 32.032 | 33.744 | 31.437 | 29.577 | 31.264 | |
SSIM | 0.939 | 0.936 | 0.956 | 0.963 | 0.939 | 0.942 | 0.946 | 0.901 | 0.913 | 0.936 | 0.942 | 0.898 | 0.902 | 0.915 | 0.867 | 0.894 | 0.918 | 0.922 | 0.865 | 0.865 | 0.888 | ||
VIF | 0.598 | 0.587 | 0.582 | 0.597 | 0.560 | 0.627 | 0.592 | 0.461 | 0.478 | 0.470 | 0.478 | 0.423 | 0.485 | 0.466 | 0.375 | 0.406 | 0.399 | 0.402 | 0.342 | 0.397 | 0.387 | ||
FSIM | 0.985 | 0.972 | 0.987 | 0.985 | 0.960 | 0.964 | 0.976 | 0.971 | 0.957 | 0.977 | 0.973 | 0.932 | 0.938 | 0.958 | 0.957 | 0.944 | 0.967 | 0.962 | 0.908 | 0.915 | 0.942 | ||
IR- CNNC | PSNR [dB] | 34.686 | 34.146 | 35.785 | 37.659 | 35.309 | 33.855 | 35.240 | 32.154 | 32.096 | 33.716 | 35.346 | 32.964 | 31.140 | 32.903 | 30.552 | 30.690 | 32.275 | 33.750 | 31.476 | 29.475 | 31.370 | |
SSIM | 0.939 | 0.940 | 0.958 | 0.964 | 0.942 | 0.942 | 0.947 | 0.902 | 0.917 | 0.938 | 0.943 | 0.902 | 0.900 | 0.917 | 0.868 | 0.896 | 0.920 | 0.923 | 0.867 | 0.863 | 0.889 | ||
VIF | 0.598 | 0.605 | 0.591 | 0.599 | 0.567 | 0.625 | 0.598 | 0.462 | 0.490 | 0.477 | 0.480 | 0.429 | 0.480 | 0.470 | 0.374 | 0.413 | 0.401 | 0.399 | 0.341 | 0.392 | 0.387 | ||
FSIM | 0.985 | 0.975 | 0.988 | 0.985 | 0.963 | 0.964 | 0.977 | 0.972 | 0.961 | 0.978 | 0.974 | 0.936 | 0.937 | 0.960 | 0.958 | 0.947 | 0.969 | 0.962 | 0.910 | 0.913 | 0.943 | ||
Mem- NetC | PSNR [dB] | 34.841 | 34.414 | 35.540 | 37.731 | 35.435 | 33.794 | 35.293 | 32.474 | 32.594 | 34.100 | 35.604 | 33.308 | 31.354 | 33.239 | 30.846 | 31.152 | 32.643 | 34.042 | 31.718 | 29.689 | 31.682 | |
SSIM | 0.940 | 0.943 | 0.959 | 0.964 | 0.943 | 0.943 | 0.949 | 0.906 | 0.921 | 0.942 | 0.945 | 0.910 | 0.905 | 0.922 | 0.872 | 0.900 | 0.923 | 0.925 | 0.872 | 0.868 | 0.893 | ||
VIF | 0.606 | 0.615 | 0.598 | 0.605 | 0.577 | 0.630 | 0.605 | 0.478 | 0.506 | 0.492 | 0.495 | 0.448 | 0.495 | 0.486 | 0.389 | 0.428 | 0.416 | 0.413 | 0.353 | 0.403 | 0.400 | ||
FSIM | 0.986 | 0.977 | 0.988 | 0.986 | 0.965 | 0.965 | 0.978 | 0.974 | 0.964 | 0.980 | 0.978 | 0.944 | 0.942 | 0.964 | 0.960 | 0.951 | 0.971 | 0.965 | 0.916 | 0.918 | 0.947 | ||
Pro_w/o_D (DSDC3) | PSNR [dB] | 34.870 | 34.405 | 35.995 | 37.886 | 35.514 | 34.004 | 35.446 | 32.319 | 32.337 | 33.913 | 35.526 | 33.138 | 31.290 | 33.087 | 30.785 | 31.070 | 32.549 | 34.052 | 31.679 | 29.666 | 31.634 | |
SSIM | 0.941 | 0.943 | 0.959 | 0.965 | 0.944 | 0.944 | 0.949 | 0.903 | 0.918 | 0.939 | 0.943 | 0.904 | 0.903 | 0.918 | 0.872 | 0.900 | 0.923 | 0.926 | 0.870 | 0.867 | 0.893 | ||
VIF | 0.607 | 0.615 | 0.601 | 0.607 | 0.581 | 0.632 | 0.607 | 0.467 | 0.491 | 0.480 | 0.481 | 0.432 | 0.488 | 0.473 | 0.382 | 0.421 | 0.409 | 0.405 | 0.344 | 0.400 | 0.394 | ||
FSIM | 0.986 | 0.977 | 0.988 | 0.986 | 0.966 | 0.966 | 0.978 | 0.972 | 0.961 | 0.979 | 0.975 | 0.938 | 0.940 | 0.961 | 0.959 | 0.950 | 0.970 | 0.963 | 0.911 | 0.916 | 0.945 | ||
Pro_wtih_D (DSDC3) | PSNR [dB] | 34.796 | 34.251 | 35.868 | 37.743 | 35.386 | 33.958 | 35.334 | 32.275 | 32.296 | 33.832 | 35.409 | 33.086 | 31.250 | 33.025 | 30.729 | 30.957 | 32.470 | 33.940 | 31.654 | 29.631 | 31.564 | |
SSIM | 0.940 | 0.942 | 0.958 | 0.964 | 0.942 | 0.943 | 0.948 | 0.902 | 0.917 | 0.938 | 0.941 | 0.902 | 0.902 | 0.917 | 0.870 | 0.898 | 0.921 | 0.924 | 0.870 | 0.867 | 0.892 | ||
VIF | 0.608 | 0.615 | 0.603 | 0.607 | 0.580 | 0.632 | 0.607 | 0.469 | 0.495 | 0.485 | 0.486 | 0.437 | 0.489 | 0.477 | 0.383 | 0.421 | 0.410 | 0.409 | 0.350 | 0.400 | 0.396 | ||
FSIM | 0.986 | 0.977 | 0.988 | 0.986 | 0.965 | 0.966 | 0.978 | 0.973 | 0.963 | 0.979 | 0.975 | 0.940 | 0.941 | 0.962 | 0.959 | 0.951 | 0.969 | 0.964 | 0.916 | 0.919 | 0.946 |
Image Set | Kodak | CIPR_M | CIPR_C | AVG | ||
---|---|---|---|---|---|---|
σn = 25 | MemNetC | PSNR [dB] | 32.474 | 32.594 | 34.100 | 33.06 |
SSIM | 0.906 | 0.921 | 0.942 | 0.92 | ||
Pro_w/o_D_ DSDC5 | PSNR [dB] | 32.424 | 32.502 | 34.052 | 32.99 | |
SSIM | 0.905 | 0.920 | 0.942 | 0.92 | ||
σn = 35 | MemNetC | PSNR [dB] | 30.846 | 31.152 | 32.643 | 31.55 |
SSIM | 0.872 | 0.900 | 0.923 | 0.90 | ||
Pro_w/o_D_ DSDC5 | PSNR [dB] | 30.860 | 31.168 | 32.654 | 31.56 | |
SSIM | 0.874 | 0.901 | 0.925 | 0.90 |
Image Set | Kodak | CIPR_M | CIPR_C | AVG | ||
---|---|---|---|---|---|---|
σn = 15 | Pro_wtih_D | PSNR [dB] | 34.796 | 34.251 | 35.868 | 34.972 |
SSIM | 0.940 | 0.942 | 0.958 | 0.947 | ||
VIF | 0.608 | 0.615 | 0.603 | 0.609 | ||
FSIM | 0.986 | 0.977 | 0.988 | 0.984 | ||
Pro_wtih_D without α | PSNR [dB] | 34.828 | 34.398 | 35.956 | 35.061 | |
SSIM | 0.940 | 0.942 | 0.959 | 0.947 | ||
VIF | 0.603 | 0.611 | 0.595 | 0.603 | ||
FSIM | 0.986 | 0.976 | 0.988 | 0.983 | ||
σn = 25 | Pro_wtih_D | PSNR [dB] | 32.275 | 32.296 | 33.832 | 32.801 |
SSIM | 0.902 | 0.917 | 0.938 | 0.919 | ||
VIF | 0.469 | 0.495 | 0.485 | 0.483 | ||
FSIM | 0.973 | 0.963 | 0.979 | 0.972 | ||
Pro_wtih_D without α | PSNR [dB] | 32.286 | 32.351 | 33.871 | 32.836 | |
SSIM | 0.903 | 0.918 | 0.939 | 0.920 | ||
VIF | 0.466 | 0.493 | 0.480 | 0.480 | ||
FSIM | 0.972 | 0.962 | 0.979 | 0.971 | ||
σn = 35 | Pro_wtih_D | PSNR [dB] | 30.726 | 30.957 | 32.473 | 31.385 |
SSIM | 0.871 | 0.898 | 0.921 | 0.897 | ||
VIF | 0.383 | 0.421 | 0.411 | 0.405 | ||
FSIM | 0.959 | 0.951 | 0.969 | 0.960 | ||
Pro_wtih_D without α | PSNR [dB] | 30.747 | 30.968 | 32.468 | 31.394 | |
SSIM | 0.871 | 0.899 | 0.922 | 0.897 | ||
VIF | 0.381 | 0.420 | 0.408 | 0.403 | ||
FSIM | 0.958 | 0.950 | 0.969 | 0.959 |
Parameter | DnCNNC | IRCNNC | MemNetC | |
---|---|---|---|---|
The number of weights | (3 × 3 × 3 × 64) + (3 × 3 × 64 × 64 × 15) + (3 × 3 × 64 × 3) = 556416 | (3 × 3 × 3 × 64) + (3 × 3 × 64 × 64 × 5) + (3 × 3 × 64 × 3) = 187776 | 3 × 3 × 3 × 64 + 3 × 3 × 64 × 64 × (2 × 6 × 6 + 6) + 3 × 3 × 64 × 64 × (7 + 8 + 9 + 10 + 11 + 12) = 4978368 | |
Parameter | Pro_w/o_D (DSDC3) | Pro_w/o_D (DSDC5) | ||
The number of weights | 1 × 1 × 24 × 96 + 12 × (3 × 3 × 96) + 11 × (1 × 1 × 96 × 96) + 3 × 3 × 96 × 3 = 116640 | 1 × 1 × 24 × 96 + 20 × (3 × 3 × 96) + 19 × (1 × 1 × 96 × 96) + 3 × 3 × 96 × 3 = 197280 |
Method | DnCNNC | IRCNNC | MemNetC | Pro_w/o_D (DSDC3) | Pro_w/o_D (DSDC5) |
---|---|---|---|---|---|
CT (MS) | 0.16 | 0.08 | 1.29 | 0.21 | 0.32 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Cho, S.I.; Park, J.H.; Kang, S.-J. A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses. Sensors 2021, 21, 1191. https://doi.org/10.3390/s21041191
Cho SI, Park JH, Kang S-J. A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses. Sensors. 2021; 21(4):1191. https://doi.org/10.3390/s21041191
Chicago/Turabian StyleCho, Sung In, Jae Hyeon Park, and Suk-Ju Kang. 2021. "A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses" Sensors 21, no. 4: 1191. https://doi.org/10.3390/s21041191
APA StyleCho, S. I., Park, J. H., & Kang, S.-J. (2021). A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses. Sensors, 21(4), 1191. https://doi.org/10.3390/s21041191