Over the past few years, deep learning-based approaches have come to dominate the areas of image processing, with no exception to the field of image denoising. Recently, self-supervised denoising networks have garnered extensive attention in the zero-shot domain. Nevertheless, training the denoising network solely with zero-shot data will lead to significant detail loss, thereby influencing the denoising quality. In this work, we proposed a novel dual mask mechanism based on Euclidean distance selection as a global masker for the blind spot network, thereby enhancing image quality and mitigating image detail loss. Furthermore, we also constructed a recurrent framework, named Blind2Grad, which integrated the gradient-regularized loss and a uniform loss for training the denoiser and enabled the double-mask mechanism to be continuously updated. This framework was capable of directly acquiring information from the original noisy image, giving priority to the key information that needed to be retained without succumbing to equivalent mapping. Moreover, we conducted a thorough theoretical analysis of the convergence properties of our Blind2Grad loss from both probabilistic and game-theoretic viewpoints, demonstrating its consistency with supervised methods. Our Blind2Grad not only demonstrated outstanding performance on both synthetic and real-world datasets, but also exhibited significant efficacy in processing images with high noise levels.
Citation: Bolin Song, Zhenhao Shuai, Yuanyuan Si, Ke Li. Blind2Grad: Blind detail-preserving denoising via zero-shot with gradient regularized loss[J]. AIMS Mathematics, 2025, 10(6): 14140-14166. doi: 10.3934/math.2025637
Over the past few years, deep learning-based approaches have come to dominate the areas of image processing, with no exception to the field of image denoising. Recently, self-supervised denoising networks have garnered extensive attention in the zero-shot domain. Nevertheless, training the denoising network solely with zero-shot data will lead to significant detail loss, thereby influencing the denoising quality. In this work, we proposed a novel dual mask mechanism based on Euclidean distance selection as a global masker for the blind spot network, thereby enhancing image quality and mitigating image detail loss. Furthermore, we also constructed a recurrent framework, named Blind2Grad, which integrated the gradient-regularized loss and a uniform loss for training the denoiser and enabled the double-mask mechanism to be continuously updated. This framework was capable of directly acquiring information from the original noisy image, giving priority to the key information that needed to be retained without succumbing to equivalent mapping. Moreover, we conducted a thorough theoretical analysis of the convergence properties of our Blind2Grad loss from both probabilistic and game-theoretic viewpoints, demonstrating its consistency with supervised methods. Our Blind2Grad not only demonstrated outstanding performance on both synthetic and real-world datasets, but also exhibited significant efficacy in processing images with high noise levels.
| [1] |
L. Lu, T. Deng, A method of self-supervised denoising and classification for sensor-based human activity recognition, IEEE Sens. J., 23 (2023), 27997–28011. https://doi.org/10.1109/JSEN.2023.3323314 doi: 10.1109/JSEN.2023.3323314
|
| [2] |
W. Dong, H. Wang, F. Wu, G. Shi, X. Li, Deep spatial-spectral representation learning for hyperspectral image denoising, IEEE T. Comput. Imag., 5 (2019), 635–648. https://doi.org/10.1109/TCI.2019.2911881 doi: 10.1109/TCI.2019.2911881
|
| [3] |
L. Zhuang, M. K. Ng, L. Gao, Z. Wang, Eigen-CNN: Eigenimages plus eigennoise level maps guided network for hyperspectral image denoising, IEEE T. Geosci. Remote, 62 (2024), 5512018. https://doi.org/10.1109/TGRS.2024.3379199 doi: 10.1109/TGRS.2024.3379199
|
| [4] |
Q. Jiang, H. Shi, S. Gao, J. Zhang, K. Yang, L. Sun, et al., Computational imaging for machine perception: Transferring semantic segmentation beyond aberrations, IEEE T. Comput. Imag., 10 (2024), 535–548. https://doi.org/10.1109/TCI.2024.3380363 doi: 10.1109/TCI.2024.3380363
|
| [5] |
F. Xiao, R. Liu, Y. Zhu, H. Zhang, J. Zhang, S. Chen, A dense multicross self-attention and adaptive gated perceptual unit method for few-shot semantic segmentation, IEEE T. Artif. Intell., 5 (2024), 2493–2504. https://doi.org/10.1109/TAI.2024.3369553 doi: 10.1109/TAI.2024.3369553
|
| [6] |
Y. Jin, S. Kwak, S. Ham, J. Kim, A fast and efficient numerical algorithm for image segmentation and denoising, AIMS Mathematics, 9 (2024), 5015–5027. https://doi.org/10.3934/math.2024243 doi: 10.3934/math.2024243
|
| [7] |
J. Liao, C. He, J. Li, J. Sun, S. Zhang, X. Zhang, Classifier-guided neural blind deconvolution: A physics-informed denoising module for bearing fault diagnosis under noisy conditions, Mech. Syst. Signal Pr., 222 (2025), 111750. https://doi.org/10.1016/j.ymssp.2024.111750 doi: 10.1016/j.ymssp.2024.111750
|
| [8] |
Y. Li, H. Chang, Y. Duan, CurvPnP: Plug-and-play blind image restoration with deep curvature denoiser, Signal Process., 233 (2025), 109951. https://doi.org/10.1016/j.sigpro.2025.109951 doi: 10.1016/j.sigpro.2025.109951
|
| [9] |
A. C. Bovik, T. S. Huang, D. C. Munson, The effect of median filtering on edge estimation and detection, IEEE T. Pattern Anal., PAMI-9 (1987), 181–194. https://doi.org/10.1109/TPAMI.1987.4767894 doi: 10.1109/TPAMI.1987.4767894
|
| [10] |
L. K. Choi, J. You, A. C. Bovik, Referenceless prediction of perceptual fog density and perceptual image defogging, IEEE T. Image Process., 24 (2015), 3888–3901. https://doi.org/10.1109/TIP.2015.2456502 doi: 10.1109/TIP.2015.2456502
|
| [11] |
H. Feng, L. Wang, Y. Wang, F. Han, H. Huang, Learnability enhancement for low-light raw image denoising: A data perspective, IEEE T. Pattern Anal., 46 (2024), 370–387. https://doi.org/10.1109/TPAMI.2023.3301502 doi: 10.1109/TPAMI.2023.3301502
|
| [12] |
Y. Luo, B. You, G. Yue, J. Ling, Pseudo-supervised low-light image enhancement with mutual learning, IEEE T. Circ. Syst. Vid., 34 (2024), 85–96. https://doi.org/10.1109/TCSVT.2023.3284856 doi: 10.1109/TCSVT.2023.3284856
|
| [13] |
Y. Zhao, Q. Zheng, P. Zhu, X. Zhang, M. Wenpeng, TUFusion: A transformer-based universal fusion algorithm for multimodal images, IEEE T. Circ. Syst. Vid., 34 (2024), 1712–1725. https://doi.org/10.1109/TCSVT.2023.3296745 doi: 10.1109/TCSVT.2023.3296745
|
| [14] |
M. Kang, M. Jung, Nonconvex fractional order total variation based image denoising model under mixed stripe and Gaussian noise, AIMS Mathematics, 9 (2024), 21094–21124. https://doi.org/10.3934/math.20241025 doi: 10.3934/math.20241025
|
| [15] |
J. Yin, X. Zhuang, W. Sui, Y. Sheng. J. Wang, R. Song, et al., A bearing signal adaptive denoising technique based on manifold learning and genetic algorithm, IEEE Sens. J., 24 (2024), 20758–20768. https://doi.org/10.1109/JSEN.2024.3403845 doi: 10.1109/JSEN.2024.3403845
|
| [16] |
X. Chen, L. Zhao, J. Xu, Z. Dai, L. Xu, N. Guo, et al., Feature-adaptive self-supervised leaning for microthrust measurement signals blind denoising, IEEE Sens. J., 25 (2025), 1015–1028. https://doi.org/10.1109/JSEN.2024.3493104 doi: 10.1109/JSEN.2024.3493104
|
| [17] |
S. Zhang, Y. Yang, Q. Qin, L. Feng, L. Jiao, Rapid blind denoising method for grating fringe images based on noise level estimation, IEEE Sens. J., 21 (2021), 8150–8160. https://doi.org/10.1109/JSEN.2021.3050237 doi: 10.1109/JSEN.2021.3050237
|
| [18] |
W. Wang, F. Wen, Z. Yan, P. Liu, Optimal transport for unsupervised denoising learning, IEEE T. Pattern Anal., 45 (2023), 2104–2118. https://doi.org/10.1109/TPAMI.2022.3170155 doi: 10.1109/TPAMI.2022.3170155
|
| [19] |
Y. Hou, J. Xu, M. Liu, G. Liu, L. Liu, F. Zhu, et al., NLH: A blind pixel-level non-local method for real-world image denoising, IEEE T. Image Process., 29 (2020), 5121–5135. https://doi.org/10.1109/TIP.2020.2980116 doi: 10.1109/TIP.2020.2980116
|
| [20] |
D. Gilton, G. Ongie, R. Willett, Deep equilibrium architectures for inverse problems in imaging, IEEE T. Comput. Imag., 7 (2021), 1123–1133. https://doi.org/10.1109/TCI.2021.3118944 doi: 10.1109/TCI.2021.3118944
|
| [21] |
B. Jiang, Y. Lu, J. Wang, G. Lu, D. Zhang, Deep image denoising with adaptive priors, IEEE T. Circ. Syst. Vid., 32 (2022), 5124–5136. https://doi.org/10.1109/TCSVT.2022.3149518 doi: 10.1109/TCSVT.2022.3149518
|
| [22] |
W. Wang, D. Yin, C. Fang, Q. Zhang, Q. Li, A novel multi-image stripe noise removal method based on wavelet and recurrent networks, IEEE Sens. J., 24 (2024), 26058–26069. https://doi.org/10.1109/JSEN.2024.3421337 doi: 10.1109/JSEN.2024.3421337
|
| [23] |
S. H. Chan, X. Wang, O. A. Elgendy, Plug-and-play ADMM for image restoration: Fixed-point convergence and applications, IEEE T. Comput. Imag., 3 (2017), 84–98. https://doi.org/10.1109/TCI.2016.2629286 doi: 10.1109/TCI.2016.2629286
|
| [24] |
Y. Sun, B. Wohlberg, U. S. Kamilov, An online plug-and-play algorithm for regularized image reconstruction, IEEE T. Comput. Imag., 5 (2019), 395–408. https://doi.org/10.1109/TCI.2019.2893568 doi: 10.1109/TCI.2019.2893568
|
| [25] |
K. Zhang, Y. Li, W. Zuo, L. Zhang, L. Van Gool, R. Timofte, Plug-and-Play Image Restoration With Deep Denoiser Prior, IEEE T. Pattern Anal., 44 (2021), 6360–6376. https://doi.org/10.1109/TPAMI.2021.3088914 doi: 10.1109/TPAMI.2021.3088914
|
| [26] | J. Lehtinen, J. Munkberg, J. Hasselgren, S. Laine, T. Karras, M. Aittala, et al., Noise2Noise: Learning image restoration without clean data, 2018, arXiv: 1803.04189 https://doi.org/10.48550/arXiv.1803.04189 |
| [27] | T. Huang, S. Li, X. Jia, H. Lu, J. Liu, Neighbor2Neighbor: Self-supervised denoising from single noisy images, In: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, 14776–14785. https://doi.org/10.1109/CVPR46437.2021.01454 |
| [28] |
J. Xu, Y. Huang, M. Cheng, L. Liu, F. Zhu, Z. Xu, et al., Noisy-as-clean: Learning self-supervised denoising from corrupted image, IEEE T. Image Process., 29 (2020), 9316–9329. https://doi.org/10.1109/TIP.2020.3026622 doi: 10.1109/TIP.2020.3026622
|
| [29] | N. Moran, D. Schmidt, Y. Zhong, P. Coady, Noisier2Noise: Learning to denoise from unpaired noisy data, In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, 12061–12069. https://doi.org/10.1109/CVPR42600.2020.01208 |
| [30] | V. Lempitsky, A. Vedaldi, D. Ulyanov, Deep image prior, In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, 9446–9454. https://doi.org/10.1109/CVPR.2018.00984 |
| [31] | A. Krull, T. Buchholz, F. Jug, Noise2Void-Learning denoising from single noisy images, In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, 2124–2132. https://doi.org/10.1109/CVPR.2019.00223 |
| [32] | J. Batson, L. Royer, Noise2Self: Blind denoising by self-supervision, 2019, arXiv: 1901.11365. https://doi.org/10.48550/arXiv.1901.11365 |
| [33] | Y. Quan, M. Chen, T. Pang, H. Ji, Self2Self with dropout: Learning self-supervised denoising from single image, In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, 1887–1895. https://doi.org/10.1109/CVPR42600.2020.00196 |
| [34] |
J. Lequyer, R. Philip, A. Sharma, W. Hsu, L. Pelletier, A fast blind zero-shot denoiser, Nat. Mach. Intell., 4 (2022), 953–963. https://doi.org/10.1038/s42256-022-00547-8 doi: 10.1038/s42256-022-00547-8
|
| [35] | Y. Mansour, R. Heckel, Zero-Shot Noise2Noise: Efficient image denoising without any data, In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, 14018–14027. https://doi.org/10.1109/CVPR52729.2023.01347 |
| [36] | Z. Wang, J. Liu, G. Li, H. Han, Blind2Unblind: Self-supervised image denoising with visible blind spots, In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, 2017–2026. https://doi.org/10.1109/CVPR52688.2022.00207 |
| [37] | X. Ma, Z. Wei, Y. Jin, P. Ling, T. Liu, B. Wang, et al., Masked pre-training enables universal zero-shot denoiser, 2024, arXiv: 2401.14966. https://doi.org/10.48550/arXiv.2401.14966 |
| [38] |
O. Ghahabi, J. Hernando, Restricted Boltzmann machines for vector representation of speech in speaker recognition, Comput. Speech Lang., 47 (2018), 16–29. https://doi.org/10.1016/j.csl.2017.06.007 doi: 10.1016/j.csl.2017.06.007
|
| [39] | Y. Xie, Z. Wang, S. Ji, Noise2Same: Optimizing a self-supervised bound for image denoising, 2020, arXiv: 2010.11971. https://doi.org/10.48550/arXiv.2010.11971 |
| [40] | C. R. Hauf, J. S. Houchin, The FlashPix™ image file format, In: Proc. IS & T 4th Color and Imaging Conf., 4 (1996), 234–234. https://doi.org/10.2352/CIC.1996.4.1.art00060 |
| [41] | R. Zeyde, M. Elad, M. Protter, On single image scale-up using sparse-representations, In: Curves and surfaces, Berlin, Heidelberg: Springer, 2012,711–730. https://doi.org/10.1007/978-3-642-27413-8_47 |
| [42] | D. Martin, C. Fowlkes, D. Tal, J. Malik, A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics, In: Proceedings Eighth IEEE International Conference on Computer Vision, 2001,416–423. https://doi.org/10.1109/ICCV.2001.937655 |
| [43] | J. Huang, A. Singh, N. Ahuja, Single image super-resolution from transformed self-exemplars, In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, 5197–5206. https://doi.org/10.1109/cvpr.2015.7299156 |
| [44] |
M. Weigert, U. Schmidt, T. Boothe, A. Müller, A. Dibrov, A. Jain, et al., Content-aware image restoration: Pushing the limits of fluorescence microscopy, Nat. Methods, 15 (2018), 1090–1097. https://doi.org/10.1038/s41592-018-0216-7 doi: 10.1038/s41592-018-0216-7
|
| [45] | A. Abdelhamed, S. Lin, M. S. Brown, A high-quality denoising dataset for smartphone cameras, In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, 1692–1700. https://doi.org/10.1109/CVPR.2018.00182 |
| [46] | J. Xu, H. Li, Z. Liang, D. Zhang, L. Zhang, Real-world Noisy Image Denoising: A New Benchmark, 2018, arXiv: 1804.02603. https://doi.org/10.48550/arXiv.1804.02603 |
| [47] | S. Nam, Y. Hwang, Y. Matsushita, S. J. Kim, A holistic approach to cross-channel image noise modeling and its application to image denoising, In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, 1683–1691. https://doi.org/10.1109/CVPR.2016.186 |
| [48] |
K. Dabov, A. Foi, V. Katkovnik, K. Egiazarian, Image denoising by sparse 3-D transform-domain collaborative filtering, IEEE T. Image Process., 16 (2007), 2080–2095. https://doi.org/10.1109/TIP.2007.901238 doi: 10.1109/TIP.2007.901238
|
| [49] | C. Mou, Q. Wang, J. Zhang, Deep generalized unfolding networks for image restoration, In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, 17378–17389. https://doi.org/10.1109/CVPR52688.2022.01688 |
| [50] | W. Lee, S. Son, K. M. Lee, AP-BSN: Self-supervised denoising for real-world images via asymmetric PD and blind-spot network, In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, 17704–17713. https://doi.org/10.1109/CVPR52688.2022.01720 |
| [51] | Z. Wang, Y. Fu, J. Liu, Y. Zhang, LG-BPN: Local and global blind-patch network for self-supervised real-world denoising, In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, 18156–18165. https://doi.org/10.1109/CVPR52729.2023.01741 |
| [52] | H. Chihaoui, P. Favaro, Masked and shuffled blind spot denoising for real-world images, In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, 3025–3034. https://doi.org/10.1109/CVPR52733.2024.00292 |
Figures(15) / Tables(5)
Bolin Song, Zhenhao Shuai, Yuanyuan Si, Ke Li. Blind2Grad: Blind detail-preserving denoising via zero-shot with gradient regularized loss[J]. AIMS Mathematics, 2025, 10(6): 14140-14166. doi: 10.3934/math.2025637
DownLoad: