Blind Denoising Research Articles

Dual convolutional neural network with attention for image blind denoising

Dual convolutional neural network with attention for image blind denoising

A high-quality self-supervised image denoising method based on SDDW-GAN and CHRNet

Image denoising remains a classic and crucial issue in the field of image processing, significantly impacting the outcomes of subsequent image processing tasks. For instance, the denoising network depends on “noise-clean” image pairs to train network effectively. However, it is often hampered by issues such as useful information loss, low training efficiency, and poor blind denoising. To address these challenges, this study proposes a novel image denoising network that integrates the complementary strengths of model-based and learning-based approaches, specifically leveraging the capabilities of both SDDW-GAN and CHRNet. Firstly, SDDW-GAN is designed to estimate the noise distribution on the input noisy images, and a fast-smoothing noisy block sampling algorithm is proposed to extract the noise blocks in noisy images in SDDW-GAN. Secondly, a network with dual generators and dual discriminators based on W-GAN is designed to estimate the noise distribution on the input noisy images and generate noise sample pairs with the same noise distribution, which solves the problem of relying on “noise-clean” image pairs. Thirdly, CHRNet is designed to compute the mapping relationship between the double-noise samples and the single-noise samples. In order to further improve the denoising effect, the dual-channel residual attention module is proposed for fusion learning of global and local features. Experimental results show that the proposed method has a better denoising effect in complex environments and outperforms existing denoising methods. Specifically, in comparison with the stand-alone denoising methods BM3D, DnCNN, Noise2Noise, and Blind2Unblind, the proposed method improves the average peak signal-to-noise ratio (PSNR) by 0.23 dB to 0.78 dB on two benchmarking datasets crossing different noise levels. Its denoising effect is also greater than other competitive stand-alone and combination methods. The proposed method can also extend to low-light image enhancement, deblurring, and super-resolution.

Speckle noise suppression of a reconstructed image in digital holography based on the BM3D improved convolutional neural network

In digital holographic measurement, when light waves pass through inhomogeneous media or surfaces, speckle noise is generated, resulting in random, granular light and dark spots in the hologram, which greatly reduces the image quality. Therefore, in order to improve the image quality of holographic reconstruction, a noise reduction method based on the BM3D improved convolutional neural network (CNN) is proposed in this paper. Firstly, the similarity and important statistical information between blocks can be obtained by using BM3D. Then, the denoising convolutional neural network (DnCNN) is used to learn the relationship between the noise of a large number of samples and the noise image, and further purify the image to retain the details for a better denoising effect. Finally, a reflective off-axis digital holographic optical path system is constructed to collect the holograms of the test samples, and the reconstructed images are obtained by the Fresnel diffraction method to constitute a dataset with the simulated holographic reconstructed images to validate the proposed method in this paper, compared to the other methods, such as DnCNN, convolutional blind denoising network (CBDNet), BM3D, and Wiener filtering. The experimental results of qualitative and quantitative analyses show that the proposed method combines the advantages of traditional algorithms and deep learning, significantly enhances the robustness of the system, optimizes the denoising performance, and preserves the details of the reconstructed image to the greatest extent.

Unsupervised Domain Adaptation for EM Image Denoising with Invertible Networks.

Electron microscopy (EM) image denoising is critical for visualization and subsequent analysis. Despite the remarkable achievements of deep learning-based non-blind denoising methods, their performance drops significantly when domain shifts exist between the training and testing data. To address this issue, unpaired blind denoising methods have been proposed. However, these methods heavily rely on image-to-image translation and neglect the inherent characteristics of EM images, limiting their overall denoising performance. In this paper, we propose the first unsupervised domain adaptive EM image denoising method, which is grounded in the observation that EM images from similar samples share common content characteristics. Specifically, we first disentangle the content representations and the noise components from noisy images and establish a shared domain-agnostic content space via domain alignment to bridge the synthetic images (source domain) and the real images (target domain). To ensure precise domain alignment, we further incorporate domain regularization by enforcing that: the pseudo-noisy images, reconstructed using both content representations and noise components, accurately capture the characteristics of the noisy images from which the noise components originate, all while maintaining semantic consistency with the noisy images from which the content representations originate. To guarantee lossless representation decomposition and image reconstruction, we introduce disentanglement-reconstruction invertible networks. Finally, the reconstructed pseudo-noisy images, paired with their corresponding clean counterparts, serve as valuable training data for the denoising network. Extensive experiments on synthetic and real EM datasets demonstrate the superiority of our method in terms of image restoration quality and downstream neuron segmentation accuracy. Our code is publicly available at https://github.com/sydeng99/DADn.

Denoising an X-ray image by exploring the power of its physical symmetry

Next-generation light source facilities offer extreme spatial and temporal resolving power, enabling multiscale, ultra-fast and dynamic characterizations. However, a trade-off between acquisition efficiency and data quality needs to be made to fully unleash the resolving potential, for which purpose powerful denoising algorithms to improve the signal-to-noise ratio of the acquired X-ray images are desirable. Yet, existing models based on machine learning mostly require massive and diverse labeled training data. Here we introduce a self-supervised pre-training algorithm with blind denoising capability by exploring the intrinsic physical symmetry of X-ray patterns without requiring high signal-to-noise ratio reference data. The algorithm is more efficient and effective than algorithms without symmetry involved, including an supervised algorithm. It allows us to recover physical information from spatially and temporally resolved data acquired in X-ray diffraction/scattering and pair distribution function experiments, where pattern symmetry is often well preserved. This study facilitates photon-hungry experiments as well as in situ experiments with dynamic loading.

Product of Gaussian Mixture Diffusion Models

In this work, we tackle the problem of estimating the density fX\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$ f_X $$\\end{document} of a random variable X\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$ X $$\\end{document} by successive smoothing, such that the smoothed random variable Y\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$ Y $$\\end{document} fulfills the diffusion partial differential equation (∂t-Δ1)fY(·,t)=0\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$ (\\partial _t - \\Delta _1)f_Y(\\,\\cdot \\,, t) = 0 $$\\end{document} with initial condition fY(·,0)=fX\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$ f_Y(\\,\\cdot \\,, 0) = f_X $$\\end{document}. We propose a product-of-experts-type model utilizing Gaussian mixture experts and study configurations that admit an analytic expression for fY(·,t)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$ f_Y (\\,\\cdot \\,, t) $$\\end{document}. In particular, with a focus on image processing, we derive conditions for models acting on filter, wavelet, and shearlet responses. Our construction naturally allows the model to be trained simultaneously over the entire diffusion horizon using empirical Bayes. We show numerical results for image denoising where our models are competitive while being tractable, interpretable, and having only a small number of learnable parameters. As a by-product, our models can be used for reliable noise level estimation, allowing blind denoising of images corrupted by heteroscedastic noise.

A dual fusion deep convolutional network for blind universal image denoising

Blind image denoising and edge-preserving are two primary challenges to recover an image from low-level vision to high-level vision. Blind denoising requires a single denoiser can denoise images with any intensity of noise, and it has practical utility since accurate noise levels cannot be acquired from realistic images. On the other hand, edge preservation can provide more image features for subsequent processing which is also important for the denoising. In this paper, we propose a novel blind universal image denoiser to remove synthesis and realistic noise while preserving the image texture. The denoiser consists of noise network and prior network parallelly, and then a fusion block is used to give the weight between these two networks to balance computation cost and denoising performance. We also use the Non-subsampled Shearlet Transform (NSST) to enlarge the size of receptive field to obtain more detailed information. Extensive denoising experiments on synthetic images and realistic images show the effectiveness of our denoiser.

On The Blind Denoising Efficiency of Image Denoising Algorithms Through Robustness, Image Quality and Computational Burden

The main goal of the image denoising is to recover the original image while attaining the structure of the image as much as possible. When the image denoising task is blind, we have no a priori information about the original image. Thus, we cannot measure the degradation level in the image directly; instead, noise variance can be estimated by the denoising algorithm. According to the estimated value, denoising is performed. Such algorithms are supposed to be robust to varying and high levels of noise interference. Moreover, in time-constrained real-world applications, they must balance the tradeoff between image quality and computation time. In this study, we assess the performance of the image denoising algorithms armored for these goals. We are aimed to determine the optimal performance yielded by such algorithms and the noise bounds wherein each algorithm is superior. After the experimental work, important conclusions are drawn.

Hybrid Dilated Convolution with Attention Mechanisms for Image Denoising

In the field of image denoising, convolutional neural networks (CNNs) have become increasingly popular due to their ability to learn effective feature representations from large amounts of data. In the field of image denoising, CNNs are widely used to improve performance. However, increasing network depth can weaken the influence of shallow layers on deep layers, especially for complex denoising tasks such as real denoising and blind denoising, where conventional networks fail to achieve high-quality results. To address this issue, this paper proposes a hybrid dilated convolution-based denoising network (AMDNet) that incorporates attention mechanisms. In specific, AMDNet consists of four modules: the sparse module (SM), the feature fusion module (FFM), the attention guidance module (AGM), and the image residual module (IRM). The SM employs hybrid dilated convolution to extract local features, while the FFM is used to integrate global and local features. The AGM accurately extracts noise information hidden in complex backgrounds. Finally, the IRM reconstructs images in a residual manner to obtain high-quality results after denoising. AMDNet has the following features: (1) The sparse mechanism in hybrid dilated convolution enables better extraction of local features, enhancing the network’s ability to capture noise information. (2) The feature fusion module, through long-range connections, fully integrates global and local features, improving the performance of the model; (3) the attention module is ingeniously designed to precisely extract features in complex backgrounds. The experimental results demonstrate that AMDNet achieves outstanding performance on three tasks (Gaussian noise, real noise, and blind denoising).

Single image noise level estimation by artificial noise

In this paper, we propose a new method for noise level estimation from a single noisy image contaminated by additive white Gaussian noise and gamma noise, which usually appear due to different physical imaging systems. Our method is motivated by the key observation that the curves of small eigenvalues of the patch covariance matrices corresponding to the original noisy image and the artificial noisy image are nearly parallel. Based on some assumptions, we derive a proportional relation equation between eigenvalues and variances of the noisy images in the redundant dimensions. The solution of this equation gives the basic noise estimation formula. However, the assumptions are not strictly satisfied for natural images with textures and fine details. To overcome this drawback, we modified the basic noise estimation formula by introducing a correction parameter. The proposed method is also extended to multiplicative Gamma noise estimation. Numerical results demonstrate that the proposed noise level estimation method outperforms the existing state-of-the-art approaches in accuracy for various scenes. In particular, the decreasing of Root Mean Square Error (RMSE) is an average of over 10% for the testing databases. We further demonstrate that the proposed noise estimation method can be used for blind denoising.

Blind Image Denoising via Deep Unfolding Network With Degradation Information Guidance

Blind denoising is an active research area in image processing. In recent years, denoising methods based on deep learning have achieved outstanding results. However, directly designing more complex networks is very challenging and often lacks interpretability. Besides, existing methods also often ignore the guidance of degradation information. Therefore, how to guide the design of deep neural networks by combining traditional algorithms while predicting and utilizing degradation information is an open problem. In this brief, based on the maximum a posterior (MAP) framework, we first estimate the degradation information, and design corresponding operators to obtain initial restoration estimation in high-dimensional mapping space. On this basis, the initial estimation is substituted into the denoising problem. According to the splitting algorithm and momentum based gradient descent method, the iterative optimization of the substituted problem is carried out, and the proposed algorithm framework is obtained. Furthermore, according to the proposed algorithm framework, we design the corresponding network structure, and a novel blind unfolding network named BDUNet is introduced. Experimental results show that our network not only outperforms blind methods but also has advantages over non-blind methods.

Variational Bayesian deep network for blind Poisson denoising

Deep learning-based approaches have recently achieved considerable results in Poisson denoising under low-light conditions. However, most existing methods mainly focus on the network architecture design, which lacks physical interpretability and thus unsuitable for blind denoising in real environments with unknown levels of noises. To address this issue, we propose a variational Bayesian deep network for blind Poisson denoising (VBDNet). We mainly consider an approximate posterior form for the noise variance in a variational Bayesian framework and utilize a neural network to parameterize the variance of Poisson noise. For network design, VBDNet is divided into two sub-networks. The noise estimation sub-network is responsible for the Bayesian inference. This network improves the blind denoising ability of the subsequent denoising sub-network by learning Poisson noise characteristics under different noise levels in the training process. A network of U-Net structures implements the denoising sub-network for noise removal. By combining the advantage of Bayesian inference (noise estimation sub-network) and deep learning (denoising sub-network), VBDNet outperforms other state-of-the-art methods on both synthetic and natural data. The code and details are available at https://github.com/HLImg/VBDNet.

A Self-Supervised Deep Learning Approach for Blind Denoising and Waveform Coherence Enhancement in Distributed Acoustic Sensing Data.

Fiber-optic distributed acoustic sensing (DAS) is an emerging technology for vibration measurements with numerous applications in seismic signal analysis, including microseismicity detection, ambient noise tomography, earthquake source characterization, and active source seismology. Using laser-pulse techniques, DAS turns (commercial) fiber-optic cables into seismic arrays with a spatial sampling density of the order of meters and a time sampling rate up to one thousand Hertz. The versatility of DAS enables dense instrumentation of traditionally inaccessible domains, such as urban, glaciated, and submarine environments. This in turn opens up novel applications such as traffic density monitoring and maritime vessel tracking. However, these new environments also introduce new challenges in handling various types of recorded noise, impeding the application of traditional data analysis workflows. In order to tackle the challenges posed by noise, new denoising techniques need to be explored that are tailored to DAS. In this work, we propose a Deep Learning approach that leverages the spatial density of DAS measurements to remove spatially incoherent noise with unknown characteristics. This approach is entirely self-supervised, so no noise-free ground truth is required, and it makes no assumptions regarding the noise characteristics other than that it is spatio-temporally incoherent. We apply our approach to both synthetic and real-world DAS data to demonstrate its excellent performance, even when the signals of interest are well below the noise level. Our proposed methods can be readily incorporated into conventional data processing workflows to facilitate subsequent seismological analyses.

Channel Estimation for Indoor Massive MIMO Visible Light Communication With Deep Residual Convolutional Blind Denoising Network

Massive multiple-input multiple-output (m-MIMO) visible light communication (VLC) has been considered as one of the promising components in the optical wireless communication system. However, the envisioned benefits may be limited due to the high computational complexity to estimate accurate channel state information (CSI). Besides, several propagation paths in the practical communication channels induces significant difficulties. In this paper, a deep residual convolutional blind denoising network (ResCBDNet) has been proposed to predict more realistic channels in indoor m-MIMO VLC communication system. Unlike the other convolutional networks which may overfit on the simplified additive white gaussian noise model, the ResCBDNet exploits noise level estimation subnetwork to improve the generalization ability to real noise as well as interactively reduce the noise in the channel matrix by adjusting the noise level map. More specifically, we have investigated to optimize the ResCBDNet over a large range of SNRs. During the simulation, the sparse channel matrix has been treated as a two dimensional natural image. Simulation results validate that the proposed network is very promising in practical channel estimation for the indoor m-MIMO VLC communication system and outperforms the state-of-the-art channel estimators in terms of normalized mean square error and peak signal-to-noise ratio.

Residual dense network with non-residual guidance for blind image denoising

Residual learning is one of the most effective components in blind image denoising. It learns to estimate the noise instead of the clean image itself. A shortcoming of residual learning is that it cannot capture hierarchical features efficiently because there are no connections between layers to extract multi-level features. Therefore, residual dense network (RDN) adds dense connections among layers in the residual block to facilitate extraction of hierarchical features. Although RDN is a superior non-blind denoiser, it is only better than state-of-the-art methods in blind denoising when noise level is high. In this paper, we employ the concept of Wiener filters and add a non-residual network as a guidance for RDN. The non-residual network predicts the signal instead of the noise to assist RDN in achieving satisfactory performance across all noise levels. Then, a guidance network combines the outputs from both residual and non-residual networks to generate the final denoised output. Experimental results show that the new architecture performs better than state-of-the-art blind denoisers quantitatively and image quality of the new method is also better than existing methods.

Contrastive blind denoising autoencoder for real time denoising of industrial IoT sensor data

In an industrial IoT setting, ensuring the quality of sensor data is a must when data-driven algorithms operate on the upper layers of the control system. Unfortunately, the common place in industrial facilities is to find sensor time series heavily corrupted by noise and outliers. In this work, a purely data-driven self-supervised learning-based approach based on a blind denoising autoencoder is proposed for real time denoising of industrial sensor data. The term \textit{blind} stresses that no prior knowledge about the noise is required for denoising, in contrast to typical denoising autoencoders where prior knowledge about the noise is required. Blind denoising is achieved by using a noise contrastive estimation (NCE) regularization on the latent space of the autoencoder, which not only helps to denoise but also induces a meaningful and smooth latent space. Experimental evaluation in both a simulated system and a real industrial process shows that the proposed technique outperforms classical denoising methods.

Research on microseismic denoising method based on CBDNet

Noise suppression is an important part of microseismic monitoring technology. Signal and noise can be separated by denoising and filtering to improve the subsequent analysis. In this paper, we propose a new denoising method based on convolutional blind denoising network (CBDNet). The method is partially modified for image denoising network CBDNet to make it suitable for one–dimensional data denoising. At present, most of the existing filtering methods are proposed for the Gaussian white noise denoising. In contrast, the proposed method also learns the wind noise, construction noise, traffic noise and mixed noise through the strategy of residual learning. The full convolution subnetwork is used to estimate the noise level, which significantly improves the signal-to-noise ratio and its performance of removing the correlated noise. The model is trained with different types of real noise and random noise. The denoising result is evaluated by corresponding indexes and compared with other denoising methods. The results show that the proposed method has better denoising performance than traditional methods, and it has a superior noise suppression level for oil well construction noise and mixed noise. The proposed method can suppress the interference of time–frequency overlapped end to end and still have noise suppression and event detection capability even if the signal is superimposed on other types of noise.

Hybrid loss and domain transform based seismic image blind denoising

Noise removal is a critical stage in the processing of seismic data. Recent data-driven approaches require pre-assumptions about the distribution of noise for training the model, but in the actual denoising work, the natural noise does not obey the assumed distribution and the ability of the network to deal with this kind of noise is limited. Therefore, this paper proposes an adversarial learning method, FDT, which based on hybrid loss and domain transform. It treats noisy seismic data and clean seismic data as two different domains, and performs adversarial training by cycling between the two domains. Then the network will automatically learn a mapping that converts noisy seismic data into clean seismic data, which can achieve promising denoising result despite the noisy distribution is unknown. Moreover, single pixel-level loss similar to L1 and L2 is not adequate to characterize the texture and structure of the seismic data, in this paper, we combined SSIM(Structure Similarity Index Measure) and L1 together as a reconstruction loss to ensure that seismic data remains the same geological and tectonic information after denoising. Experimental results demonstrate that the method in this paper can effectively remove most types of noise from seismic data and achieve advanced results on both synthetic and real datasets. And it outputs a PSNR(Peak Signal-to-Noise Ratio) value of 27.97 with a standard deviation of 0.2 for large scale noise, maintaining a highly reliable denoising performance.

Remote Sensing Image Recovery and Enhancement by Joint Blind Denoising and Dehazing

Due to the hazy weather and the long-distance imaging path, the captured remote sensing image (RSI) may suffer from detail loss and noise pollution. However, simply applying dehazing operation on a noisy hazy image may result in noise amplification. Therefore, in this paper, we propose joint blind denoising and dehazing for RSI recovery and enhancement to address this problem. Firstly, we propose an efficient and effective noise level estimation method based on quad-tree subdivision and integrate it into FFDNet for blind denoising. Secondly, a multi-scale guided filter decomposes the denoised hazy image into base and detailed layers, separating the initial details. Then, the dehazing procedure using the corrected boundary constraint is implemented in the base layer, while a nonlinear sigmoid mapping function enhances the detailed layers. The last step is to fuse the enhanced detailed layers and the dehazed base layer to get the final result. Using both synthetic remote sensing hazy image (RSHI) datasets and real-world RSHI, we perform comprehensive experiments to evaluate the proposed method. Results show that our method is superior to well-known methods in both dehazing and joint denoising and dehazing tasks.

Edge preserving image denoising with deep networks and ensemble convolutional neural network architectures

In the field of visual perception, the edges of images tend to be rich in effective visual stimuli, which contribute to the neural network's understanding of various scenes. Image smoothing is an image processing method used to highlight the wide area, low-frequency components, and main part of the image or to suppress image noise and high-frequency interference components, which can make the image's brightness smooth and gradual, reduce the abrupt gradient, and improve the image quality. Reducing noise is treated as one of the important problems in image processing. At the same time, preserving the edges of objects is of critical importance to protect the visual appearance of the objects. Deep networks have marked a trend in computer vision applications and this paper presents a customized Gaussian noise minimizing network with edge preserving filers. Here, ensemble architecture of convolutional neural network is used in minimizing the Gaussian noise and image denoising. The ensemble architecture is combined with VGG-19 and Xception design of CNN. The ensemble Convolutional Neural Networks (CNNs) are classified for Gaussian noisy images, real noisy images, blind denoising, and hybrid noisy images, representing the combination of noisy, blurred, and lowresolution images. Following the classification, motivations and principles of various deep learning methods are analyzed. Subsequently, a comparison of state-of-the-art methods on public denoising datasets is conducted, considering both quantitative and qualitative analyses. The experimental analysis is carried out in terms of PSNR, accuracy, precision, recall and F-measure.