Denoising Research Articles

RIRGAN: An end-to-end lightweight multi-task learning method for brain MRI super-resolution and denoising

A common problem in the field of deep-learning-based low-level vision medical images is that most of the research is based on single task learning (STL), which is dedicated to solving one of the situations of low resolution or high noise. Our motivation is to design a model that can perform both SR and DN tasks simultaneously, in order to cope with the actual situation of low resolution and high noise in low-level vision medical images. By improving the existing single image super-resolution (SISR) network and introducing the idea of multi-task learning (MTL), we propose an end-to-end lightweight MTL generative adversarial network (GAN) based network using residual-in-residual-blocks (RIR-Blocks) for feature extraction, RIRGAN, which can concurrently accomplish super-resolution (SR) and denoising (DN) tasks. The generator in RIRGAN is composed of several residual groups with a long skip connection (LSC), which can help form a very deep network and enable the network to focus on learning high-frequency (HF) information. The introduction of a discriminator based on relativistic average discriminator (RaD) greatly improves the discriminator’s ability and makes the generated image have more realistic details. Meanwhile, the use of hybrid loss function not only ensures that RIRGAN has the ability of MTL, but also enables RIRGAN to give a more balanced attention between quantitative evaluation of metrics and qualitative evaluation of human vision. The experimental results show that the quality of the restoration image of RIRGAN is superior to the SR and DN methods based on STL in both subjective perception and objective evaluation metrics when processing medical images with low-level vision. Our RIRGAN is more in line with the practical requirements of medical practice.

Research on Bi–ISAR Sparse Aperture High Resolution Imaging Algorithm under Low SNR

In the imaging process of bistatic inverse synthetic aperture radar (Bi–ISAR), the echo is easily affected by the internal interference and external environment of the radar system, resulting in the problems of sparse aperture and low echo signal noise ratio. The efficiency of conventional sparse aperture imaging methods is greatly reduced. To solve the above problems, a Bi–ISAR sparse aperture imaging algorithm based on Complex Variational Modal Decomposition (CVMD) and wavelet threshold de–noising is proposed. Firstly, the Bi–ISAR sparse aperture echo signal model is established, the sparse basis matching the echo is constructed, the echo signal is decomposed into different intrinsic mode functions (IMF) by CVMD, and the IMFs belonging to the signal are separated by the energy relationship criterion. Then, after the IMFs is de–noising by the improved wavelet threshold de–noising algorithm, the original signal is synthesized, and the signal after de–noising is reconstructed by using the orthogonal matching pursuing algorithm. Simulation results show that the proposed algorithm can achieve Bi–ISAR sparse aperture high−resolution imaging under low signal noise ratio and has fine anti−noise performance.

Deep learning-based super-resolution and de-noising for XMM-newton images

ABSTRACT The field of artificial intelligence based image enhancement has been rapidly evolving over the last few years and is able to produce impressive results on non-astronomical images. In this work, we present the first application of Machine Learning based super-resolution (SR) and de-noising (DN) to enhance X-ray images from the European Space Agency’s XMM-Newton telescope. Using XMM-Newton images in band [0.5, 2] keV from the European Photon Imaging Camera pn detector (EPIC-pn), we develop XMM-SuperRes and XMM-DeNoise – deep learning-based models that can generate enhanced SR and DN images from real observations. The models are trained on realistic XMM-Newton simulations such that XMM-SuperRes will output images with two times smaller point-spread function and with improved noise characteristics. The XMM-DeNoise model is trained to produce images with 2.5× the input exposure time from 20 to 50 ks. When tested on real images, DN improves the image quality by 8.2 per cent, as quantified by the global peak-signal-to-noise ratio. These enhanced images allow identification of features that are otherwise hard or impossible to perceive in the original or in filtered/smoothed images with traditional methods. We demonstrate the feasibility of using our deep learning models to enhance XMM-Newton X-ray images to increase their scientific value in a way that could benefit the legacy of the XMM-Newton archive.

Development of Convolutional Network-based Denoising Technique using Deep Reinforcement Learning in Computed Tomography

Development of Convolutional Network-based Denoising Technique using Deep Reinforcement Learning in Computed Tomography

Filter-based Denoising Methods for AWGN corrupted images

Filter-based Denoising Methods for AWGN corrupted images

Frequency Domain Denoising and Regularity Study on XRF Spectra of Soils With Different Lead Concentrations

Frequency Domain Denoising and Regularity Study on XRF Spectra of Soils With Different Lead Concentrations

Research on sparsity of frequency modulated signal in fractional Fourier transform domain

Frequency modulated (FM) signal is extensively applied in sonar, radar, laser and emerging optical cross-research, its sparsity is a common basic issue for the sampling, denoising and compression of FM signal. This paper mainly studies the sparsity of FM signal in the fractional Fourier transform (FRFT) domain, and a maximum singular value method (MSVM) is proposed to estimate the compact FRFT domain of FM signal. This method uses the maximum singular value of amplitude spectrum of FM signal to measure the compact domain, and WOA is used to search the compact domain, which effectively improves the shortcomings of the existing methods. Compared with MNM and MACF, this method gives a sparser representation of FM signal in the FRFT domain, which has less number of significant amplitudes. Finally, the primary application of this method in the FM signal filtering is given.

Evaluation of Denoising Filters Based on Edge Locations

This paper presents a method to evaluate denoising filters based on edge locations in their denoised images. Image quality assessment has often been performed by using structural similarity (SSIM). However, SSIM does not provide clearly the geometric accuracy of features in denoised images. Thus, in this paper, a method to localize edge locations with subpixel accuracy based on adaptive weighting of gradients is used for obtaining the subpixel locations of edges in ground truth image, noisy images, and denoised images. Then, this paper proposes a method to evaluate the geometric accuracy of edge locations based on root mean squares error (RMSE) and jaggedness with reference to ground truth locations. Jaggedness is a measure proposed in this study to measure the stability of the distribution of edge locations. Tested denoising filters are anisotropic diffusion (AF), bilateral filter, guided filter, weighted guided filter, weighted mean of patches filter, and smoothing filter (SF). SF is a simple filter that smooths images by applying a Gaussian blurring to a noisy image. Experiments were performed with a set of simulated images and natural images. The experimental results show that AF and SF recovered edge locations more accurately than the other tested filters in terms of SSIM, RMSE, and jaggedness and that SF produced better results than AF in terms of jaggedness.

Texture Based Segmentation of Remotely Sensed Images

Texture based segmentation approach extracts homogeneous regions that have similar texture properties. The texture patterns make use of the gray scale relationship between the center pixel and its surrounding neighbors. In this paper, a novel method is proposed that encodes the spatial relationship between adjacent pairs of neighbors on either side of the center pixel including itself along the given directions in an image. The novelty of the proposed method is that the texture operator is applied on multiple windows of the image to produce a stack of images from which the texture pattern is obtained. The proposed operator is applied on SAR images of ocean surface to detect regions of interest like ship or slick. Keywords: Binary pattern, De-speckle, Gradient, Segmentation, Watershed.

Despeckling for side-scan sonar images based on adaptive block-matching and 3D filtering

Side-scan sonar (SSS) is an electronic device that utilizes the propagation characteristics of sound waves under water to complete underwater detection. Because the SSS produces images and maps according to the intensity of acoustic echo, speckle noise will be inevitably involved. A speckle denoising method based on block-matching and 3D filtering (BM3D) is proposed to filter the multiplicative speckle noise in SSS images. First, the SSS image is transformed by power and logarithm. The wavelet transform is used to estimate the general noisy level of the polluted image. Second, the parameters of the BM3D algorithm are updated according to the noise estimation results of each local patch. At last, after comparing the general noise estimation and the local noise estimation, the proposed algorithm chooses the best estimation to filter every patch separately to solve the problem that the noise is not evenly distributed. The experimental results show that the improved BM3D algorithm can effectively reduce the speckle noise in SSS images and obtain good visual effects. The Equivalent Number of Looks of the proposed algorithm is at least 6.83% higher, the Speckle Suppression Index is lower than traditional algorithm, and the Speckle Suppression and Mean Preservation Index is reduced by at least 3.30%. This method is mainly used for sonar image noise reduction, and has certain practical values for ultrasonic, radar or OCT images polluted by speckle noise.

Design of Cascaded Adaptive Filter for ECG Denoising Applications

Design of Cascaded Adaptive Filter for ECG Denoising Applications

Development of a coupled aerosol lidar data quality assurance and control scheme with Monte Carlo analysis and bilateral filtering

Mie-scatter lidar can capture the vertical distribution of aerosols, and a high degree of quantification of lidar data would be capable of coupling with a chemical transport model (CTM). Thus, we develop a data quality assurance and control scheme for aerosol lidar (TRANSFER) that mainly includes a Monte Carlo uncertainty analysis (MCA) and bilateral filtering (BF). The AErosol RObotic NETwork (AERONET) aerosol optical depth (AOD) is utilized as the ground truth to evaluate the validity of TRANSFER, and the result exhibits a sharp 41% (0.36) decrease in root mean square error (RMSE), elucidating an acceptable overall performance of TRANSFER. The maximum removal of uncertainties appears in MCA with an RMSE of 0.08 km−1, followed by denoising (DN) with 50% of MCA in RMSE. BF can smooth interior data without destroying the edge of the structure. The most noteworthy correction occurs in summer with an RMSE of 0.15 km−1 and Pearson correlation coefficient of 0.8, and the least correction occurs in winter with values of 0.07 km−1 and 0.93, respectively. Overestimations of raw data are mostly identified, and representative values occur with weak southerly winds, low visibility, high relative humidity (RH) and high concentrations of both ground fine particulate matter (PM2.5) and ozone. Apart from long-term variations, the intuitional variation in a typical overestimated pollution episode, especially represented by vertical profiles, shows a favorable performance of TRANSFER during stages of transport and local accumulation, as verified by backward trajectories. Few underestimation cases are mainly attributed to BF smoothing data with a sudden decrease. The main limitation of TRANSFER is the zigzag profiles found in a few cases with very small extinction coefficients. As a supplement to the research community of aerosol lidar and an exploration under complicated pollution in China, TRANSFER can aid in the preprocessing of lidar data-powered applications.

End-to-End Unpaired Image Denoising with Conditional Adversarial Networks

Image denoising is a classic low level vision problem that attempts to recover a noise-free image from a noisy observation. Recent advances in deep neural networks have outperformed traditional prior based methods for image denoising. However, the existing methods either require paired noisy and clean images for training or impose certain assumptions on the noise distribution and data types. In this paper, we present an end-to-end unpaired image denoising framework (UIDNet) that denoises images with only unpaired clean and noisy training images. The critical component of our model is a noise learning module based on a conditional Generative Adversarial Network (cGAN). The model learns the noise distribution from the input noisy images and uses it to transform the input clean images to noisy ones without any assumption on the noise distribution and data types. This process results in pairs of clean and pseudo-noisy images. Such pairs are then used to train another denoising network similar to the existing denoising methods based on paired images. The noise learning and denoising components are integrated together so that they can be trained end-to-end. Extensive experimental evaluation has been performed on both synthetic and real data including real photographs and computer tomography (CT) images. The results demonstrate that our model outperforms the previous models trained on unpaired images as well as the state-of-the-art methods based on paired training data when proper training pairs are unavailable.

Artefact Removal from EEG Signals u sing Total Variation De noising

Artefacts removing (de-noising) from EEG signals has been an important aspect for medical practitioners for diagnosis of health issues related to brain. Several methods have been used in last few decades. Wavelet and total variation based de-noising have attracted the attention of engineers and scientists due to their de-noising efficiency. In this article, EEG signals have been de-noised using total variation based method and results obtained have been compared with the results obtained from the celebrated wavelet based methods . The performance of methods is measured using two parameters: signal-to-noise ratio and root mean square error. It has been observed that total variation based de-noising methods produce better results than the wavelet based methods.

Image denoising algorithm based on boosting high order non-convex total variation model

Image denoising algorithm based on boosting high order non-convex total variation model

融合l1-TV正则化约束RPCA模型的视频去噪和目标检测

融合l1-TV正则化约束RPCA模型的视频去噪和目标检测

A review of image denoising methods

A review of image denoising methods

Review on CNN Based Image Denoising

Review on CNN Based Image Denoising

Image denoising algorithm based on dual-branch modified codec

针对传统图像去噪算法多噪声去除难，深层卷积神经网络去噪模型网络复杂、训练时间长等问题，提出一种基于自编码器结构的双分支改良编解码网络，实现高效图像去噪。双分支结构之一采用降-升采样实现点噪声消除，另一分支专注于宏观的图像修复和伪像去除，后端利用残差结构进行整合，实现数字图像混合噪声去噪。实验结果显示：对于含有标准差为15，均值为0的高斯噪声、噪声密度为5%的椒盐噪声和散粒噪声的混合噪声图像测试集，实验去噪效果相较于输入混合噪声图像峰值信噪比，平均提升了5.3%。与12层全卷积神经网络相比，去噪效果相当，训练速度提升了25.4%，体现了其“轻量级”的优点。实验表明：该方法相较于深层卷积神经网络，训练速度快，网络简单；相较于传统图像去噪算法，噪声去除效果也较为明显。该算法可应用于轻量级视觉平台后端去噪。

Seismic response analysis of embankment dams under decomposed earthquakes

In this study, the seismic response analysis of embankment dams was investigated through numerical modeling. The seismic behavior of dams under main earthquake records and wavelet-based records were studied. Earthquake records were decomposed using de-noising method (DNM) and down-sampling method (DSM) up to five levels. In decomposition process, low and high frequencies of the main earthquake record were separated into two signals. Acceleration response, spectral acceleration, and Fourier amplitude spectrum at the crest of embankment dams under different decomposition levels were evaluated. The seismic behavior under main and decomposed earthquake records was compared. The results indicate an acceptable agreement between the seismic responses of embankment dams under wavelet-based decomposed records and main earthquake motions. Dynamic analyses show that the DNM-based decomposed earthquake records have a better performance compared to DSM-based records. DNM-based records up to level 4 and DSM-based records up to level 2 have a high accuracy in assessment of seismic behavior of embankment dams. The periods corresponding to the maximum values of acceleration spectra and the frequencies corresponding to the maximum values of Fourier amplitude spectra of embankment dam crest under main and decomposed records are in good agreement. The results demonstrate that the main earthquake records can be replaced by wavelet-based decomposed records in seismic analysis of embankment dams.