Deep image prior-based self-supervised framework for noise reduction with structural preservation in neutron imaging

Noise reduction algorithms are widely used to mitigate noise effects on speech to improve the robustness of speech technology applications. However, they inevitably cause speech distortion. The tradeoff between noise reduction and speech distortion is a key concern in designing noise reduction algorithms. This study proposes a novel framework for noise reduction by considering this tradeoff. We regard speech estimation as a function approximation problem in a regularized reproducing kernel Hilbert space (RKHS). In the estimation, the objective function is formulated to find an approximation function by controlling the tradeoff between approximation accuracy and function complexity. For noisy observations, this is equivalent to controlling the tradeoff between noise reduction and speech distortion. Since the target function is approximated in an RKHS, either a linear or nonlinear mapping function can be naturally incorporated in the estimation by a “kernel trick”. Traditional signal subspace and Wiener filtering based noise reduction can be derived as special cases when a linear kernel function is applied in this framework. We first provided a theoretical analysis of the tradeoff property of the framework in noise reduction. Then we applied our proposed noise reduction method in speech enhancement and noisy robust speech recognition experiments. Compared to several classical noise reduction methods, our proposed method showed promising advantages.

Periodic structures that may exist within the neutron imaging detector can introduce periodic noise into the imaging results, directly degrading image quality and further affecting the performance of deconvolution. This periodic noise appears as four-pointed star-shaped peaks in the amplitude spectrum of the frequency domain. However, the distribution of honeycomb-like noise structures in neutron imaging results makes it difficult to detect using conventional thresholding methods. We propose a method that applies a dilation operation before threshold detection to enhance the contrast between peaks and the surrounding areas. Then, a notch filter is used to smooth the peaks containing noise information, thereby removing the periodic noise structure. This approach effectively eliminates honeycomb structures of approximately 40 micrometers and improves the image quality after deconvolution processing.

We address computed tomography (CT) metal artifacts reduction (MAR) using a generative deep-learning model in the imaging physics framework. Existing deep learning-based MAR methods, though promising, generally lack explicit physical modeling of artifact formation and rely heavily on data-driven mappings. The absence of physics priors not only limits scalability, as they often require paired or task-specific datasets, but also makes such methods prone to hallucination, anatomical distortion, and unstable artifact suppression. We propose a novel self-supervised framework for CT MAR, integrating a lightweight multi-layer perceptron (MLP)-based beam-hardening correction with a conditional latent diffusion model (LDM). By incorporating a physics-informed correction step and an artifact-reproducing simulation technique, the framework aims to enhance scalability across diverse scenarios, reduce hallucination effects, and improve structural fidelity in the reconstructed images. The proposed MLP performs physics-driven polynomial correction, serving as a simplified but efficient alternative to existing approaches. Also, the proposed MLP implicitly incorporates sinogram consistency into its optimization objective, allowing case-specific adaptation and convergence toward the desired solution. Additionally, the learned MLP parameters are reused to simulate artifact-contaminated images from artifact-free scans, generating pseudo paired data for self-supervised training without requiring real paired datasets. A conditional LDM is trained on these synthetic pairs to remove residual artifacts. By operating in a low-dimensional latent space, the LDM significantly reduces inference time while maintaining high-quality reconstructions. The proposed method is evaluated on both the SynDeepLesion dataset and real clinical data, demonstrating superior artifact removal and structural preservation compared to the existing state-of-the-art MAR techniques. We particularly highlight the robustness, generalizability, and clinical applicability of the proposed framework. We proposed a self-supervised metal artifact reduction framework that combines MLP-based beam-hardening correction with a conditional latent diffusion model in the imaging physics framework. The MLP module provides physics motivated beam-hardening corrected CT images, while the residual artifact simulation strategy enables fully self-supervised training without the need for paired data. The proposed method demonstrated superior artifact suppression and structural preservation on both synthetic and clinical datasets, outperforming existing approaches.

IoT has opened the floodgates the importance use of image processing in areas like surveillance, auto-navigating vehicles, and smart healthcare. Nonetheless, noise in IoT-generated images largely distorts image quality as well as subsequent analytical operations that may be performed. The fundamental goal is to derive and test stable mathematical models that improve noise reduction while maintaining real-time computation that would be viable for limited IoT devices. The paper uses both first order Linear filtering techniques like Gaussian filters and Median and second order filters which are in the transform domain like Wavelet transforms. These techniques are then systematically invoked on images contaminated by Gaussian, Salt-and-Pepper and Speckle noise types for performance evaluation. Simulation studies using MATLAB with camera man image ‘cameraman.tif’ for standard prove the Wavelet transforms outperforms in noise reduction attaining up to 15.3 dB PSNR and 0.92 SSIM for better structural preservation. On the other hand, Gaussian and Median filters take 0.05 – 0.07 seconds to process and hence well suited for real-time applications.

This paper proposes a deep convolutional neural network-based low-light image enhancement method. In order to adaptively enhance the image brightness, a convolutional neural network with convolutional modules is designed. Lowlight image is firstly down-sampled into sub-images. Then an illumination map is obtained from the input image to provide additional information to the network. The network works on a tensor that consists of sub-images and illumination map, achieving a good performance in brightness increasing and structure preservation. The enhanced result is reconstructed from the output sub-images. Experimental results demonstrate the effectiveness of the proposed method in low-light image enhancement.

Deep learning-based image fusion techniques have been widely applied in multi-modal medical image fusion. However, existing methods often face challenges such as insufficient preservation of image structure and quality, loss of multi-scale structural information, and noise and artifacts during the fusion process. To address these issues, this paper proposes an uncertainty and gradient-guided deep medical image fusion network (UAGFusion), which includes an uncertainty-guided fusion network (UAFN) and a gradient-guided modulation gate refinement network (GGMGFN), with the aim of exploring methods to improve medical image fusion performance. The UAFN incorporates an uncertainty-guided feature enhancement module, cross-scale attention mechanism, and multi-level feature fusion (FF) module to enhance the integration and expression capabilities of multi-modal features. The GGMGFN employs a gradient-guided detail extractor to capture edge and texture information and uses a modulated gated fusion module to integrate multi-source feature streams. To guide network training, this paper designs a hybrid loss function comprising content loss, edge loss, multi-scale structural similarity loss, perceptual loss, and uncertainty perceptual loss to achieve a balance between fusion quality and uncertainty. Experimental results demonstrate that the proposed method achieves competitive performance compared to existing approaches across several objective metrics, while offering improvements in structural information preservation and noise robustness.

The purpose of the study was to present a sequence of development of an algorithm for improving the quality of images obtained using side-scan sonar. Mathematical image processing techniques such as contrast enhancement, edge processing, and colorimetric techniques were used to achieve this goal. Steps to improve image contrast included normalising signal intensity, adaptive contrast enhancement using limited contrast alignment of histograms, and correction of uneven lighting. The sonar radiation pattern and image intensity normalisation scheme were demonstrated. The contrast limited adaptive histogram equalisation filter showed higher values of the peak signal-to-noise ratio and structural similarity index compared to conventional histogram alignment, indicating better preservation of detail, image structure, and noise reduction. Analysis of edge processing, in particular by Canny and Sobel, has shown their potential effectiveness in improving the detail of underwater structures. In addition, the use of Gaussian smoothing allowed reducing the level of high-frequency noise and make textures smoother. As a result, there was a decrease in graininess, softness of object contours, and overall smoothing of the scene. In addition, cubic spline regression showed normalised image data. In turn, colorimetric analysis focused on converting images between greyscale and colour spaces, which made it easier to identify underwater objects and structures. An example of Hue-Saturation-Value components was given, which demonstrated different effects on the quality of sonar image visualisation. The Value component provided the most expressive distinction between the object and the background, while the Hue component was ineffective for structure analysis. The combination of Value and Saturation allowed for improved contour detail. Optimisation of the pseudo-colour gamut allowed adapting the image to different tasks, contributing to more accurate object recognition. The results obtained confirm the feasibility of using the presented methods in a wide range of applied tasks related to visualisation and analysis of underwater environments

We present a nonparametric regression method for denoising 3-D image sequences acquired via fluorescence microscopy. The proposed method exploits the redundancy of the 3-D+time information to improve the signal-to-noise ratio of images corrupted by Poisson-Gaussian noise. A variance stabilization transform is first applied to the image-data to remove the dependence between the mean and variance of intensity values. This preprocessing requires the knowledge of parameters related to the acquisition system, also estimated in our approach. In a second step, we propose an original statistical patch-based framework for noise reduction and preservation of space-time discontinuities. In our study, discontinuities are related to small moving spots with high velocity observed in fluorescence video-microscopy. The idea is to minimize an objective nonlocal energy functional involving spatio-temporal image patches. The minimizer has a simple form and is defined as the weighted average of input data taken in spatially-varying neighborhoods. The size of each neighborhood is optimized to improve the performance of the pointwise estimator. The performance of the algorithm (which requires no motion estimation) is then evaluated on both synthetic and real image sequences using qualitative and quantitative criteria.

Noise robust single image super-resolution using a multiscale image pyramid

One of the major factors plaguing the performance of synthetic aperture radar (SAR) imagery is the presence of signal-dependent speckle noise. Grainy in appearance, speckle noise is primarily due to the phase fluctuations of the electromagnetic return signals. Since inherent spatial-correlation characteristics of speckle in SAR images are not exploited in existing multiplicative models for speckle noise, a new approach is proposed here that provides a new mathematical framework for modeling and reduction of speckle noise. The contribution of this paper is twofold. First, a novel model for speckled SAR imagery is introduced based on Markov random fields (MRFs) in conjunction with statistical optics. Second, utilizing the model, a global energy-minimization algorithm, based on simulated annealing (SA), is introduced for speckle reduction. In particular, the joint conditional probability density function (cpdf) of the intensity of any two points in the speckled image and the associated correlation function are used to derive the cpdf of any center pixel intensity given its four neighbors. The Hammersley- Clifford theorem is then used to derive the energy function associated with the MRF. The SA, built on the Metropolis sampler, is employed for speckle reduction. Four metrics are used to assess the quality of the speckle reduction: the mean-square error, SNR, an edge-preservation parameter and the equivalent number of looks. A comparative study using both simulated speckled images as well as real SAR images indicates that the proposed approach performs favorably in comparison to existing filtering techniques such as the modified-Lee and the enhanced Frost-algorithms.

We propose a new algorithm for despeckling multiframe Optical Coherence Tomography (OCT) data based on wavelet shrinkage using anisotropic diffusion and similarity comparison between frames. In this algorithm detail coefficients are weighted for noise reduction, where these weights are calculated based on similarity comparison between approximation coefficients. This comparison is based on the assumption that frames have similar structural content while noise is temporally uncorrelated. Approximation coefficients are denoised using Perona Malik anisotropic diffusion. Finally these processed coefficients are averaged to get a denoised image. Experimental results show that the proposed method performs better than the previously formulated denoising algorithms both in terms of noise reduction and structural content preservation.

This paper presents an efficient adjoint-based shape optimization framework for airframe noise reduction in which algorithmic differentiation (AD) is applied to the open-source multiphysics solver SU2 to obtain design sensitivities. An AD-based consistent discrete adjoint solver is developed that directly inherits the convergence properties of the primal flow solver due to the differentiation of the entire nonlinear fixed-point iterator. In addition, a 2D and 3D far-field noise prediction framework coupling the Computational Fluid Dynamics (CFD) and Computational Aeroacoustics (CAA) using a permeable surface Ffowcs Williams-Hawkings (FWH) approach in the time domain is also developed. The resultant AD-based discrete adjoint solver is applied to the reduction of noise radiation from 2D and 3D rod–airfoil configurations. The results suggest that the unsteady adjoint information provided by this AD-based discrete adjoint framework is accurate and robust, due to the AD of the entire design chain including the dynamic mesh movement routine and various turbulence models, as well as the hybrid CFD-CAA model. This framework is shown to be highly effective in reducing the tonal noise component due to the turbulent wake impingement on the airfoil leading edge. This study also compares the optimization results obtained on the basis of 3D unsteady Reynolds-averaged Navier–Stokes equation simulations and analyzes them using scale-resolving simulations of higher fidelity.

The development of jet noise reduction (JNR) techniques is critical to the integration of supersonic aircraft into commercial travel and transportation. Prescribing JNR methods remains inherently difficult as the reduction of sound is frequently accompanied by a reduction in nozzle performance. This paper utilizes resolvent analysis as a reduced-order modeling technique for the purpose of assessing jet noise reduction concepts on a low bypass ratio coannular nozzle design. An automated framework is built to facilitate the computation of resolvent gain values for varying nozzle conditions and geometries. Specifically, parametric studies are conducted on the mixing duct length and extraction ratio parameters of the nozzle design. RANS mean flow data, used as input for the resolvent computation, are validated against experimental data for the same nozzle configuration. The resolvent analysis is verified by comparing sets of singular values and forcing/response modes against corresponding plots reported in the literature. Tailoring the resolvent framework with approaches using both near-field and far-field acoustics operators has been confirmed to reach consistent conclusions. Maximal gain profiles computed for a range of nozzle mixing duct lengths are found to be insensitive to this design parameter. Likewise, dominant gain profiles are also found to be insensitive to variations in extraction ratio at constant thrust. Both of these behaviors are found to be consistent with experimental data suggesting an insubstantial change in noise when varying these nozzle operating parameters. The findings of these parametric studies suggest that the maximal resolvent gain can be a viable proxy for jet noise. Applying resolvent analysis as a means to assess JNR potential shows promise for comparative design studies, though extending the study to more diverse nozzle conditions is necessary to affirm this.

In this letter, we propose an unsupervised framework for speech noise reduction based on the recent development of low-rank and sparse matrix decomposition. The proposed framework directly separates the speech signal from noisy speech by decomposing the noisy speech spectrogram into three submatrices: the noise structure matrix, the clean speech structure matrix, and the residual noise matrix. Evaluations on the Noisex-92 dataset show that the proposed method achieves a signal-to-distortion ratio approximately 2.48 dB and 3.23 dB higher than that of the robust principal component analysis method and the non-negative matrix factorization method, respectively, when the input SNR is -5 dB.

X-ray backlighting is a crucial diagnostic technique in inertial confinement fusion (ICF) experiments, enabling the observation of the geometrical structure of an imploding capsule. However, x-ray emission from a short-pulse laser illuminating a metallic strip often proves insufficient and unstable, resulting in high noise levels in observed images that complicate data analysis. In this study, we propose a comprehensive framework for multi-level noise reduction in x-ray backlighting images, integrating data synthetic techniques with a novel multi-level noise removal method based on Frequency Residual U-Net (FR-UNet). The observed x-ray images are decomposed into distinct signal and noise components. The signal is simulated based on the underlying backlighting geometry, incorporating a range of relevant parameters to accurately model the setup variance. The noise model is constructed through a comprehensive analysis of both shot noise and periodic speckle noise. The FR-UNet model is trained on synthetically generated images and subsequently applied to experimental data. Our framework demonstrates a significant improvement in image quality, with the signal-to-noise ratio increasing from ∼10–20 dB to ∼30 dB. Further evaluation on an open dataset shows favorable results, achieving a peak signal-to-noise ratio of ∼35 and a structural similarity index measure of ∼0.96. This approach offers substantial potential for enhancing image quality and diagnostic accuracy in inertial confinement fusion (ICF) experiments and can be adapted to similar applications involving x-ray backlighting systems in other high-energy physics contexts.