Image Denoising Research Articles

Low-light Image Enhancement Method Based on Illumination Curve Adjustment

This paper presents an advanced low-light image enhancement approach based on Zero-Reference Deep Curve Estimation. The technique enhances images through: (1) the integration of the Dynamic Stochastic Resonance equation for enhanced contrast and brightness, leveraging adjustable parameters for optimal image quality, and (2) application of an Anisotropic Diffusion function for image denoising and smoothing, preserving edges and texture details while effectively removing noise. This adaptive algorithm tailors diffusion intensity to local image features, further refining the resulting image quality. Experimental results confirm the efficacy of our approach in improving visual quality and noise removal in low-light images.

A Neural Network for Hyperspectral Image Denoising by Combining Spatial–Spectral Information

Hyperspectral imaging often suffers from various types of noise, including sensor non-uniformity and atmospheric disturbances. Removing multiple types of complex noise in hyperspectral images (HSIs) while preserving high fidelity in spectral dimensions is a challenging task in hyperspectral data processing. Existing methods typically focus on specific types of noise, resulting in limited applicability and an inadequate ability to handle complex noise scenarios. This paper proposes a denoising method based on a network that considers both the spatial structure and spectral differences of noise in an image data cube. The proposed network takes into account the DN value of the current band, as well as the horizontal, vertical, and spectral gradients as inputs. A multi-resolution convolutional module is employed to accurately extract spatial and spectral noise features, which are then aggregated through residual connections at different levels. Finally, the residual mixed noise is approximated. Both simulated and real case studies confirm the effectiveness of the proposed denoising method. In the simulation experiment, the average PSNR value of the denoised results reached 31.47 at a signal-to-noise ratio of 8 dB, and the experimental results on the real data set Indian Pines show that the classification accuracy of the denoised hyperspectral image (HSI) is improved by 16.31% compared to the original noisy version.

Neural Network Classification in Javanese Handwriting Recognition using Projection Profile Histogram and Local Binary Pattern Histogram

Indonesia consists of various regional tribes, where each tribe has cultural diversity and some even have their own regional letters, like Javanese tribe has Javanese characters. Javanese letters consist of 20 basic letters called Nglegena script. Subject about Javanese language is delivered to elementary student until now aims to preserve Indonesian culture especially the Javanese. In this study, we present two feature extraction methods are Local Binary Pattern (LBP) and Profile Projection (PP). Neural Network (NN) chosen as classifiers for classifying 20 javanese letters Nglegena. Some digital image processing processes are carried out, are image inversion, dilation, denoising and skeletoning. The Javanese script dataset is taken from the Kaggle database with the name Aksara Jawa: Aksara Jawa Custom Dataset, consists of 2154 train images and 480 test images. The experiment were carried out in two models, Projection Profile Histogram - Neural Network (PPH-NN) and Local Binary Pattern Histogram - Neural Network (LBPH-NN). The experiment show that both feature extraction methods have very good performance, 99.98% PPH-NN and 89.6% LBPH-NN on average.

RestorNet: An efficient network for multiple degradation image restoration

Even though Single Degradation Image Restoration (SDIR) has made significant progress and achieved remarkable performance, Multiple Degradation Image Restoration (MDIR) remains a long-term and arduous challenge to achieve the similar levels of success. To further improve the performance and efficiency of MDIR, we propose a novel MDIR method named RestorNet, which comprises an unsupervised degradation encoder for the learning of multi-scale degradation representations and a Multi-scale Degradation-assisted Restoration Module (MDRM) for image reconstruction. Our RestorNet aims to remove noise, rain, and haze in a unified network from the following three aspects. Firstly, to better distinguish among different degradations and learn the corruption information more accurately, we introduce a degradation-specific contrastive loss based on contrastive learning. Next, we develop a multi-scale degradation representation learning method to improve preservation of the spatial structure and distribution of inputs, and to extract multi-scale information to satisfy the diverse requirements of restoring different degraded images. Finally, to make a more reasonable use of degradation representation, we present a novel semi-guided strategy for effective feature transformation, where the multi-scale degradation representations are only incorporated into the MDRM encoder. For image denoising, deraining, and dehazing, by integrating the approaches above, RestorNet not only outperforms the recent state-of-the-art MDIR algorithms with lower computational complexity, but also achieves impressive performance in SDIR. Extensive experiments demonstrate the effectiveness and superiority of our proposed method.

Weighted guided image filtering with entropy evaluation weighting

Although the guided image filter (GIF) is an excellent edge-preserving filter, it generally suffers from halo artifacts due to the local property and the fixed regularization parameter. To address the problem, a weighted guided image filter (WGIF) was proposed by incorporating an edge-aware weighting into the GIF. In the filtering process, WGIF employs an averaging strategy for edge-aware weighting. Although the averaging strategy is a highly efficient method, it is susceptible to extreme values and tends to obscure critical factors, so it often leads to inaccurate results. Consequently, the output results quality of the WGIF is often degraded. To remedy the deficiency, a weighted guided image filter with entropy evaluation weighting (EEW-WGIF) is proposed in this paper. EEW-WGIF employs an edge-aware weighting strategy based on entropy evaluation method to detect edges more accurately, and incorporates an explicit constraint based on the gradient variation to better preserve edges. To verify the filtering effectiveness of the EEW-WGIF, it was applied to edge-preserving smoothing filtering, exposure images fusion, single image detail enhancement, structure-transferring filtering and image denoising. Experimental results show that the proposed filter can achieve excellent performance in both visual quality and objective evaluation.

Deep Learning-Based Denoising of CEST MR Data: A Feasibility Study on Applying Synthetic Phantoms in Medical Imaging.

Chemical Exchange Saturation Transfer (CEST) magnetic resonance imaging (MRI) provides a novel method for analyzing biomolecule concentrations in tissues without exogenous contrast agents. Despite its potential, achieving a high signal-to-noise ratio (SNR) is imperative for detecting small CEST effects. Traditional metrics such as Magnetization Transfer Ratio Asymmetry (MTRasym) and Lorentzian analyses are vulnerable to image noise, hampering their precision in quantitative concentration estimations. Recent noise-reduction algorithms like principal component analysis (PCA), nonlocal mean filtering (NLM), and block matching combined with 3D filtering (BM3D) have shown promise, as there is a burgeoning interest in the utilization of neural networks (NNs), particularly autoencoders, for imaging denoising. This study uses the Bloch-McConnell equations, which allow for the synthetic generation of CEST images and explores NNs efficacy in denoising these images. Using synthetically generated phantoms, autoencoders were created, and their performance was compared with traditional denoising methods using various datasets. The results underscored the superior performance of NNs, notably the ResUNet architectures, in noise identification and abatement compared to analytical approaches across a wide noise gamut. This superiority was particularly pronounced at elevated noise intensities in the in vitro data. Notably, the neural architectures significantly improved the PSNR values, achieving up to 35.0, while some traditional methods struggled, especially in low-noise reduction scenarios. However, the application to the in vivo data presented challenges due to varying noise profiles. This study accentuates the potential of NNs as robust denoising tools, but their translation to clinical settings warrants further investigation.

Image Denoising of Low-Dose PET Mouse Scans with Deep Learning: Validation Study for Preclinical Imaging Applicability.

Positron emission tomography (PET) image quality can be improved by higher injected activity and/or longer acquisition time, but both may often not be practical in preclinical imaging. Common preclinical radioactive doses (10MBq) have been shown to cause deterministic changes in biological pathways. Reducing the injected tracer activity and/or shortening the scan time inevitably results in low-count acquisitions which poses a challenge because of the inherent noise introduction. We present an image-based deep learning (DL) framework for denoising lower count micro-PET images. For 36 mice, a 15-min [18F]FDG (8.15 ± 1.34MBq) PET scan was acquired at 40min post-injection on the Molecubes β-CUBE (in list mode). The 15-min acquisition (high-count) was parsed into smaller time fractions of 7.50, 3.75, 1.50, and 0.75min to emulate images reconstructed at 50, 25, 10, and 5% of the full counts, respectively. A 2D U-Net was trained with mean-squared-error loss on 28 high-low count image pairs. The DL algorithms were visually and quantitatively compared to spatial and edge-preserving denoising filters; the DL-based methods effectively removed image noise and recovered image details much better while keeping quantitative (SUV) accuracy. The largest improvement in image quality was seen in the images reconstructed with 10 and 5% of the counts (equivalent to sub-1MBq or sub-1min mouse imaging). The DL-based denoising framework was also successfully applied on the NEMA-NU4 phantom and different tracer studies ([18F]PSMA, [18F]FAPI, and [68Ga]FAPI). Visual and quantitative results support the superior performance and robustness in image denoising of the implemented DL models for low statistics micro-PET. This offers much more flexibility in optimizing preclinical, longitudinal imaging protocols with reduced tracer doses or shorter durations.

Deep-Learning-Based Multitask Ultrasound Beamforming

In this paper, we present a new method for multitask learning applied to ultrasound beamforming. Beamforming is a critical component in the ultrasound image formation pipeline. Ultrasound images are constructed using sensor readings from multiple transducer elements, with each element typically capturing multiple acquisitions per frame. Hence, the beamformer is crucial for framerate performance and overall image quality. Furthermore, post-processing, such as image denoising, is usually applied to the beamformed image to achieve high clarity for diagnosis. This work shows a fully convolutional neural network that can learn different tasks by applying a new weight normalization scheme. We adapt our model to both high frame rate requirements by fitting weight normalization parameters for the sub-sampling task and image denoising by optimizing the normalization parameters for the speckle reduction task. Our model outperforms single-angle delay and sum on pixel-level measures for speckle noise reduction, subsampling, and single-angle reconstruction.

Denoising convolutional autoencoder for improving the classification performance based on noisy galaxy images

Images of galaxy objects are of great importance to the work of astronomers. Nowadays, the task of galaxy classification has been aided by Neural-Network-based classification models, who are powerful yet vulnerable to the attack of noisy images. In this research, RGB noises and bright spots simulating stars were generated, and a Convolutional-Neural-Network (CNN) based lightweight denoising image autoencoder was proposed. Firstly, a benchmark CNN classifier using DenseNet structure was trained on the Galaxy 10 DECals dataset, which consists of over 17, 000 RGB color galaxy images. Then, noisy images were generated by adding bright spots of different size and color simulating stars and applying gaussian RGB noises over the original images. The CNN autoencoder that consists of Convolutional layer in its encoder and Convolution Transpose layers in decoder was trained on the raw and noisy training data to learn effective galaxy image denoising. Finally, the effect of the autoencoder was evaluated by contrasting the performance of the CNN classifier over the noisy and denoised images. In contrast with being evaluated on the raw testing set, the CNN classifiers accuracy dropped by 0.41 when tested on the generated noisy testing images, indicating the effectiveness of the attack of image noises. While after denoising with the proposed autoencoder, the classifiers accuracy increased significantly by 0.37. Output denoised images also suggest that the autoencoder can effectively remove the applied bright spot and gaussian RGB noise, recreating the original shape of the galaxy.

Adaptive block level bilateral filtering algorithm

During the acquisition or transmission process, video images are subject to random signal interference and generate noise, which can hinder people's understanding of the image and subsequent processing work. Therefore, it is necessary to study video image denoising and filtering algorithms. Bilateral filter is one of many typical video image filtering algorithms. However, the traditional bilateral filter algorithm does not consider the differences in the contents of different regions of the image. It is difficult to obtain the optimal filtering effect by using a fixed filtering weight to filter the entire image, which leads to problems such as blurred image edges and inadequate details processing. This paper studies the influence of different filter block sizes on the bilateral filter effect, and proposes an algorithm to adaptively update the bilateral filter weight according to the block's variance. The experimental result shows that the performance of adaptive bilateral filter with different block sizes is expected to be better than that of traditional algorithms with fixed filter weights.

Impact of Traditional and Embedded Image Denoising on CNN-Based Deep Learning

In digital image processing, filtering noise is an important step for reconstructing a high-quality image for further processing such as object segmentation, object detection, and object recognition. Various image-denoising approaches, including median, Gaussian, and bilateral filters, are available in the literature. Since convolutional neural networks (CNN) are able to directly learn complex patterns and features from data, they have become a popular choice for image-denoising tasks. As a result of their ability to learn and adapt to various denoising scenarios, CNNs are powerful tools for image denoising. Some deep learning techniques such as CNN incorporate denoising strategies directly into the CNN model layers. A primary limitation of these methods is their necessity to resize images to a consistent size. This resizing can result in a loss of vital image details, which might compromise CNN’s effectiveness. Because of this issue, we utilize a traditional denoising method as a preliminary step for noise reduction before applying CNN. To our knowledge, a comparative performance study of CNN using traditional and embedded denoising against a baseline approach (without denoising) is yet to be performed. To analyze the impact of denoising on the CNN performance, in this paper, firstly, we filter the noise from the images using traditional means of denoising method before their use in the CNN model. Secondly, we embed a denoising layer in the CNN model. To validate the performance of image denoising, we performed extensive experiments for both traffic sign and object recognition datasets. To decide whether denoising will be adopted and to decide on the type of filter to be used, we also present an approach exploiting the peak-signal-to-noise-ratio (PSNRs) distribution of images. Both CNN accuracy and PSNRs distribution are used to evaluate the effectiveness of the denoising approaches. As expected, the results vary with the type of filter, impact, and dataset used in both traditional and embedded denoising approaches. However, traditional denoising shows better accuracy, while embedded denoising shows lower computational time for most of the cases. Overall, this comparative study gives insights into whether denoising will be adopted in various CNN-based image analyses, including autonomous driving, animal detection, and facial recognition.

Comparative Analysis of the Prox Penalty and Bregman Algorithms for Image Denoising

Image restoration is an interesting ill-posed problem. It plays a critical role in the concept of image processing. We are looking for an image that is as near to the original as possible among images that have been skewed by Gaussian and additive noise. Image deconstruction is a technique for restoring a noisy image after it has been captured. The numerical results achieved by the prox-penalty method and the split Bregman algorithm for anisotropic and isotropic TV denoising problems in terms of image quality, convergence, and signal noise rate (SNR) are compared in this paper. It should be mentioned that isotropic TV denoising is faster than anisotropic. Experimental results indicate that the prox algorithm produces the best high-quality output (clean, not smooth, and textures are preserved). In particular, we obtained (21.4, 21) the SNR of the denoising image by the prox for sigma 0.08 and 0.501, such as we obtained (10.0884, 10.1155) the SNR of the denoising image by the anisotropic TV and the isotropic TV for sigma 0.08 and (-1.4635, -1.4733) for sigma 0.501.

Anti-scattering medium computational ghost imaging with modified Hadamard patterns

Illumination patterns of computational ghost imaging (CGI) systems suffer from reduced contrast when passing through a scattering medium, which causes the effective information in the reconstruction result to be drowned out by noise. A two-dimensional (2D) Gaussian filter performs linear smoothing operation on the whole image for image denoising. It can be combined with linear reconstruction algorithms of CGI to obtain the noise-reduced results directly, without post-processing. However, it results in blurred image edges while performing denoising and, in addition, a suitable standard deviation is difficult to choose in advance, especially in an unknown scattering environment. In this work, we subtly exploit the characteristics of CGI to solve these two problems very well. A kind of modified Hadamard pattern based on the 2D Gaussian filter and the differential operation features of Hadamard-based CGI is developed. We analyze and demonstrate experimentally that using Hadamard patterns for illumination but using our developed modified Hadamard patterns for reconstruction (MHCGI) can enhance the robustness of CGI against turbid scattering medium. Our method not only helps directly obtain noise-reduced results without blurred edges but also requires only an approximate standard deviation, i.e., it can be set in advance. The experimental results on transmitted and reflected targets demonstrate the feasibility of our method. Our method helps to promote the practical application of CGI in the scattering environment.

Wavelet analysis model inspired convolutional neural networks for image denoising

In recent years, many image denoising methods have been proposed based on convolutional neural networks (CNNs). While these methods have shown continuous performance improvement by introducing various mechanisms and structures, their computational cost tends to become increasingly expensive, owing to the resulting complex network architectures. This paper aims at winning the trade-off between computational efficiency and denoising performance for CNN-based image denoisers. Towards this end, we draw inspirations from traditional variational models with wavelet analysis operators for CNN architecture design. A model-inspired CNN is proposed with four key modules: iterative encoding-decoding units inspired by the iterative denoising process, directional convolutions inspired by the separable wavelet filters, inception modules inspired by the multi-scale analysis of wavelets, and stage-wise connections inspired by the adding noise back operation. In experiments, our CNN shows high computational efficiency in both training and test, with competitive results to state-of-the-art approaches.

Pest Detection Using Image Denoising and Cascaded Unet Segmentation for Pest Images

This study proposes a novel approach for pest detection in pest images using image denoising and cascaded UNET segmentation. The proposed approach involves the use of a hybrid neural network RESNET50 with CNN for image dataset training, optimized using Believed Adam Optimization. The images are then preprocessed using a superior MLP model for image denoising, which enhances the image quality and reduces the noise present in the image. The images are then passed through an adaptive UNET architecture for image segmentation, which is based on domain adaptation and semantic segmentation. The cascaded UNET segmentation improves the segmentation accuracy, and the domain adaptation ensures that the model can be applied to new datasets without requiring additional training. The proposed approach achieves a high accuracy rate of 98.5% in detecting pests images. This approach can be used in various applications related to pest detection and management, including agriculture and pest control.

Improving Adversarial Robustness via Distillation-Based Purification

Despite the impressive performance of deep neural networks on many different vision tasks, they have been known to be vulnerable to intentionally added noise to input images. To combat these adversarial examples (AEs), improving the adversarial robustness of models has emerged as an important research topic, and research has been conducted in various directions including adversarial training, image denoising, and adversarial purification. Among them, this paper focuses on adversarial purification, which is a kind of pre-processing that removes noise before AEs enter a classification model. The advantage of adversarial purification is that it can improve robustness without affecting the model’s nature, while another defense techniques like adversarial training suffer from a decrease in model accuracy. Our proposed purification framework utilizes a Convolutional Autoencoder as a base model to capture the features of images and their spatial structure.We further aim to improve the adversarial robustness of our purification model by distilling the knowledge from teacher models. To this end, we train two Convolutional Autoencoders (teachers), one with adversarial training and the other with normal training. Then, through ensemble knowledge distillation, we transfer the ability of denoising and restoring of original images to the student model (purification model). Our extensive experiments confirm that our student model achieves high purification performance(i.e., how accurately a pre-trained classification model classifies purified images). The ablation study confirms the positive effect of our idea of ensemble knowledge distillation from two teachers on performance.

A Comprehensive Review of GAN-Based Denoising Models for Low-Dose Computed Tomography Images

Computed Tomography (CT) offers great visualization of the intricate internal body structures. To protect a patient from the potential radiation-related health risks, the acquisition of CT images should adhere to the “as low as reasonably allowed” (ALARA) standard. However, the acquired Low-dose CT (LDCT) images are inadvertently corrupted by artifacts and noise during the processes of acquisition, storage, and transmission, degrading the visual quality of the image and also causing the loss of image features and relevant information. Most recently, generative adversarial network (GAN) models based on deep learning (DL) have demonstrated ground-breaking performance to minimize image noise while maintaining high image quality. These models’ ability to adapt to uncertain noise distributions and representation-learning ability makes them highly desirable for the denoising of CT images. The state-of-the-art GANs used for LDCT image denoising have been comprehensively reviewed in this research paper. The aim of this paper is to highlight the potential of DL-based GAN for CT dose optimization and present future scope of research in the domain of LDCT image denoising.

A differential evolution with autonomous strategy selection and its application in remote sensing image denoising

Differential evolution (DE) is an efficient and effective global optimization algorithm, and many DE variants with adaptive strategy selection have been proposed. However, among the existing DE variants with adaptive strategy selection, only a few last generations of population information are utilized to conduct statistical analysis for strategy selection, and the population cumulative historical experience information does not be fully used to assist population to effectively search. In this paper, a DE with autonomous strategy selection (ASS-DE) is proposed to make full use of the population cumulative historical experience information. In ASS-DE, an individual can autonomously choose the better mutation strategy guided by the cumulative historical experience. As the same time, a parameter updating mechanism with archive is introduced to assign appropriate control parameters to the strategy. Additionally, an evolutionary learning mechanism based on individual similarity is proposed to guide the learning of historical experience information. To verify the performance of ASS-DE, CEC2015 and CEC2017 benchmark functions and the real-world optimization problem of wavelet parameter optimization in remote sensing image denoising is compared with some state-of-the-art intelligence algorithms. The experimental results show that ASS-DE obtains the promising performance compared with other DE variants. The source code of ASS-DE can be found in GitHub at https://github.com/Ramessis/ASS-DE_denoising.git.

Image cascade matching method based on an improved random sampling consistency algorithm

In the process of image matching, image quality will have a certain impact on the extraction of image features, directly affecting the matching accuracy of subsequent images. Therefore, an image cascade matching method based on improved random sampling consistency algorithm is proposed. This method first needs to improve the quality of the matched target image through preprocessing results such as image denoising and enhancement, and prepare for feature point extraction during subsequent image cascade matching; Then extract the feature points of the target to be matched in the image, and use the improved random sampling consistency method to quickly estimate and simplify the registration model based on the extraction results. Use similar feature triangles to perform model pre testing, and use the maximum Euclidean distance method to improve the accuracy of image cascade registration, achieving image cascade matching. The experimental results show that the proposed method can achieve image enhancement while removing image noise, improve the quality of the image, comprehensively extract the feature points to be matched in the image, and achieve accurate matching of image feature points.

V-shaped neural network structure based on multi-scale features for image denoising

Due to the good performance of deep learning, more and more image denoising methods incorporating deep learning are implemented, including fixed-scale-based methods and multi-scale-based methods. It is easy for fixed-scale-based neural networks to achieve a better denoising performance when their depth is increased, whereas feature extraction of multiple scales from images can be obtained using multi-scale-based neural networks. In this work, a multi-scale image denoising method has been proposed which is mainly based on fixed-scale but combines the method of obtaining feature information from shallow image sampling. We propose a diamond-shaped module and a V-shaped subnetwork for extracting the features of images obtained from shallow sampling and improve the up-sampling and down-sampling units to obtain a better sampling effect. Densely connecting images of same size in MSDN can solve the problem of gradient vanishing since increasing the depth of the network and shallow image sampling prevent the loss of image details due to excessive sampling. Experiments show that our method can efficiently retain the image details during image denoising, and this method is especially helpful for preserving minute image details as compared to the other state-of-the-art denoising algorithms.