Search

Blind noise-level estimation (NLE) is a fundamental issue in digital image processing. This paper provides a noise-level estimator for additive white Gaussian noise and multiplicative Gaussian noise using principal texture patches (PTPs). The proposed algorithm first identifies the principal texture of the noisy image by using the principal component analysis, and then, it chooses PTPs to calculate the noise level. The major contributions of this paper toward addressing the challenges in the NLE literature include: 1) analyzing the nonlinear relationship between the smallest eigenvalue, true noise level, number of chosen patches, and block size; 2) extracting PTPs; and 3) computing noise level using PTPs. The abundant experimental results show that the proposed method works well for various natural images over a large range of noise levels and it performs well for multiplicative noise. Compared with some state-of-the-art approaches, our method has the best performance and a faster running speed. [ABSTRACT FROM AUTHOR]

Image denoising methods generally remove not only noise but also fine-scale textures and thus degrade the subjective image quality. In this paper, we propose a method of recovering the texture component that is lost under a state-of-the-art denoising method called weighted nuclear norm minimization (WNNM). We recover the image texture with a linear minimum mean squared error estimator (LMMSE filter), which requires statistical information about the texture and noise. This requirement is the key problem preventing the application of the LMMSE filter for texture recovery because such information is not easily obtained. We propose a new method of estimating the necessary statistical information using Stein’s lemma and several assumptions and show that our estimated information is more accurate than the simple estimation in terms of the Fréchet distance. Experimental results show that our proposed method can improve the objective quality of denoised images. Moreover, we show that our proposed method can also improve the subjective quality when an additional parameter is chosen for the texture to be added.

In this study, the authors propose a method to calculate the consistency of alpha masking to assess the robustness of the matting algorithm. This study evaluates consistent alpha masks based on the Gaussian–Hermite moment in combination with gradient amplitude and gradient direction. The gradient direction describes the appearance and shape of local objects in the image, and the gradient amplitude accurately reflects the contrast and texture changes of small details in the image. They selected Gaussian blur, pretzel noise, and combined noise to destroy the image, and then evaluated the consistency of the original alpha mask and noise alpha mask. To determine the robustness of the matting algorithm, they assessed the degree of consistency of the alpha mask using three different evaluation levels. The experimental results show that noise has a greater impact on the performance of the matting algorithm, which shows a decreasing trend as the noise level in the image deepens. In noisy images, the traditional matting algorithm exhibits better robustness compared to the recently proposed trap matting algorithm. Different matting algorithms present different adaptations to different noises.

Since the signal-to-noise ratio of synthetic aperture radar (SAR) image is low, it is difficult to match SAR images with optical images accurately. A new method of SAR village targets extraction and heterogonous image registration is proposed in this study. Median filter and partial differential method are first used to remove speckle noise; next, the grey level co-occurrence matrix algorithm is used by calculating four typical texture parameters of it to obtain preliminary regions of interest (ROI) of village targets and false alarm areas are cleared by proposed local-grey-level congruency method; then proper numbers of Fourier descriptors and morphological algorithms are used to describe ROI again to maintain the boundary features of targets better. Finally, HU moment invariants, which is combined with Euclidean distance and cosine similarity measurements, are utilised to achieve heterogonous image registration. The experimental results show that the root-mean-square-error of the registration image is small and the heterogonous registration image is accurate, demonstrating that the authors’ method is stable and accurate, and owns certain research value and broad application prospect as well.

To overcome the staircase effects and simultaneously avoid edge blurring, here the authors proposed a novel algorithm of adaptive fourth-order partial differential equation (PDE) for image smoothing. The proposed method is based on You and Kaveh (YK) model that first performs a pre-processing using coherence-enhancing diffusion filter before the smoothing. Then, the authors select an adaptive edge-stopping function founded on the Tansig function to effectively remove the noise and insignificant details without destruction of important edges and textures in the image. MATLAB simulation results validate the efficiency of the proposed model quantitatively and visually, when compared with other state-of-the-art methods.

Objective. While simulated low-dose CT images and phantom studies cannot fully approximate subjective and objective effects of deep learning (DL) denoising on image quality, live animal models may afford this assessment. This study is to investigate the potential of DL in CT dose reduction on image quality compared to iterative reconstruction (IR). Approach. The upper abdomen of a live 4 year old sheep was scanned on a CT scanner at different exposure levels. Images were reconstructed using FBP and ADMIRE with 5 strengths. A modularized DL network with 5 modules was used for image reconstruction via progressive denoising. Radiomic features were extracted from a region over the liver. Concordance correlation coefficient (CCC) was applied to quantify agreement between any two sets of radiomic features. Coefficient of variation was calculated to measure variation in a radiomic feature series. Structural similarity index (SSIM) was used to measure the similarity between any two images. Diagnostic quality, low-contrast detectability, and image texture were qualitatively evaluated by two radiologists. Pearson correlation coefficient was computed across all dose-reconstruction/denoising combinations. Results. A total of 66 image sets, with 405 radiomic features extracted from each, are analyzed. IR and DL can improve diagnostic quality and low-contrast detectability and similarly modulate image texture features. In terms of SSIM, DL has higher potential in preserving image structure. There is strong correlation between SSIM and radiologists’ evaluations for diagnostic quality (0.559) and low-contrast detectability (0.635) but moderate correlation for texture (0.313). There is moderate correlation between CCC of radiomic features and radiologists’ evaluation for diagnostic quality (0.397), low-contrast detectability (0.417), and texture (0.326), implying that improvement of image features may not relate to improvement of diagnostic quality. Conclusion. DL shows potential to further reduce radiation dose while preserving structural similarity, while IR is favored by radiologists and more predictably alters radiomic features. [ABSTRACT FROM AUTHOR]

Images of titanium alloy welds detected by time-of-flight diffraction (TOFD) have problems, including large noise signals and many interference streaks around the defects, all of which seriously limit the accuracy and effectiveness of defect recognition. Existing image denoising methods lack the knowledge of the noise characteristics of TOFD images of titanium alloy weld and the preprocessing experience of technicians in the field. In addition, it is difficult to select the parameters of the preprocessing methods, and they are easily influenced by the level of technical personnel, resulting in low efficiency and poor consistency in preprocessing. To address these problems, we proposed a denoising method based on the combination of wavelet band features and deep-learning theory for TOFD images of titanium alloy weld. First, based on the wavelet preprocessing method and the experience of nondestructive testing (NDT) technicians, we constructed an image pair dataset consisting of the original TOFD images of titanium alloy weld and the desired target images to realize the accumulation of engineers' preprocessing knowledge. Second, we constructed a multiband wavelet feature fusion U-net image denoising model (WU-net) and designed a loss function under three constraints of image consistency, image texture information consistency, and structural similarity. This model was able to learn to achieve end-to-end adaptive denoising for TOFD images of titanium alloy weld. Third, we illustrated and validated the effectiveness of TOFD image preprocessing for titanium alloy weld. The results showed that the proposed method effectively eliminated TOFD image noise and improved the accuracy of defect recognition. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Image denoising, as an essential component of image pre-processing, effectively reduces noise interference to enhance image quality, a factor of considerable research importance. Traditional denoising methods often lead to the blurring of image details and a lack of realism at the image edges. To deal with these issues, we propose an image denoising algorithm named Residual structure and Cooperative Attention mechanism based on Generative Adversarial Networks (RCA-GAN). This algorithm proficiently reduces noise while focusing on preserving image texture details. To maximize feature extraction, this model first employs residual learning within a portion of the generator's backbone, conducting extensive multi-dimensional feature extraction to preserve a greater amount of image details. Secondly, it introduces a simple yet efficient cooperative attention module to enhance the representation capacity of edge and texture features, further enhancing the preservation of intricate image details. Finally, this paper constructs a novel loss function—the Multimodal Loss Function—for the network training process. The experimental results were evaluated using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) as evaluation metrics. The experimental results demonstrate that the proposed RCA-GAN image denoising algorithm has increased the average PSNR from 24.71 dB to 33.76 dB, achieving a 36.6% improvement. Additionally, the average SSIM value has risen from 0.8451 to 0.9503, indicating a 12.4% enhancement. It achieves superior visual outcomes, showcasing the ability to preserve image texture details to a greater extent and excel in edge preservation and noise suppression. [ABSTRACT FROM AUTHOR]

Eliminating mixed noise from images is a challenging task because accurately describing the attenuation of noise distribution is difficult. However, most existing algorithms for mixed noise removal solely rely on the local information of the image and neglect the global information, resulting in suboptimal denoising performance when dealing with complex mixed noise. In this paper, we propose a nested UNet based on multi-scale feature extraction (MSNUNet) for mixed noise removal. In MSNUNet, we introduce a U-shaped subnetwork called MSU-Subnet for multi-scale feature extraction. These multi-scale features contain abundant local and global features, aiding the model in estimating noise more accurately and improving its robustness. Furthermore, we introduce a multi-scale feature fusion channel attention module (MSCAM) to effectively aggregate feature information from different scales while preserving intricate image texture details. Our experimental results demonstrate that MSNUNet achieves leading performance in terms of quality metrics and the visual appearance of images. [ABSTRACT FROM AUTHOR]

Intensified complementary metal-oxide semiconductor (ICMOS) sensors can capture images under extremely low-light conditions (≤0.01 lux illumination), but the results exhibit spatially clustered noise that seriously damages the structural information. Existing image-denoising methods mainly focus on simulated noise and real noise from normal CMOS sensors, which can easily mistake the ICMOS noise for the latent image texture. To solve this problem, we propose a low-light cross-scale transformer (LL-CSFormer) that adopts multi-scale and multi-range learning to better distinguish between the noise and signal in ICMOS sensing images. For multi-scale aspects, the proposed LL-CSFormer designs parallel multi-scale streams and ensures information exchange across different scales to maintain high-resolution spatial information and low-resolution contextual information. For multi-range learning, the network contains both convolutions and transformer blocks, which are able to extract noise-wise local features and signal-wise global features. To enable this, we establish a novel ICMOS image dataset of still noisy bursts under different illumination levels. We also designed a two-stream noise-to-noise training strategy for interactive learning and data augmentation. Experiments were conducted on our proposed ICMOS image dataset, and the results demonstrate that our method is able to effectively remove ICMOS image noise compared with other image-denoising methods using objective and subjective metrics. [ABSTRACT FROM AUTHOR]

Noise in an image can affect one's extraction of image information, therefore, image denoising is an important image pre-processing process. Many of the existing models have a large number of estimated parameters, which increases the time complexity of the model solution and the achieved denoising effect is less than ideal. As a result, in this paper, an improved image-denoising algorithm is proposed based on the TV model, which effectively solves the above problems. The L1 regularization term can make the solution generated by the model sparser, thus facilitating the recovery of high-quality images. Reducing the number of estimated parameters, while using the inverse gradient to estimate the regularization parameters, enables the parameters to achieve global adaption and improves the denoising effect of the model in combination with the TV regularization term. The split Bregman iteration method is used to decouple the model into several related subproblems, and the solutions of the coordinated subproblems are derived as optimal solutions. It is also shown that the solution of the model converges to a Karush–Kuhn–Tucker point. Experimental results show that the algorithm in this paper is more effective in both preserving image texture structure and suppressing image noise. [ABSTRACT FROM AUTHOR]

Deep learning has been widely used in removing image noise, but the blind denoising problem in agriculture limits the accuracy of citrus recognition by picking robots, and most denoising methods do not preserve the texture details of the image well. To solve these problems, this study proposes a multi‐module image denoising method based on feature pyramid network. The algorithm is based on the pyramid network structure, adding modules such as attention mechanism, dilated convolution, and feature fusion to denoise images. The attention mechanism and dilated convolution enhance the extraction of noisy features. The feature fusion module fuses the feature maps of each layer to solve the problem of incomplete image detail texture preservation. The experimental results of this algorithm on different datasets show that the performance of our method on various datasets with different noise levels is superior, and it is solving the problem of real image denoising. It is effective to perform both on and save the image texture details, which proves the practicability of the method in this study. After using this method to remove the real noise in the image, the average score of citrus detection accuracy is 52.4% higher than that before denoising. Practical Applications: Citrus picking is a labor‐intensive work. Traditionally, citrus picking depends on manpower. This harvesting method has high cost and low efficiency, which seriously hinders the development of citrus industry. In this study, the pyramid network structure is used to denoise the citrus image by adding modules such as attention mechanism, dilated convolution, and feature fusion, so as to guide the robot to recognize the citrus and improve the production efficiency. It is of great value to the recognition technology of industrial citrus picking robot. [ABSTRACT FROM AUTHOR]

When the art images are restored by the virtual restoration method, there are problems such as insufficient clarity and more noise in the reference image. An improved cartoon-texture decomposition method for art image fusion is proposed. The nonlinear local total variation component is used as the indicator function of image decomposition to obtain the image cartoon structure component and texture oscillation component. According to the oscillation component's strong repetitiveness and structural directionality, the image texture part is filtered by combining the improved directional diffusion algorithm. Using the sparse coefficients of the fused cartoon component and the sparse coefficients of the texture component, the cartoon and texture of the image is inverse transformed and weighted and summed to obtain the recovered image after fusion. The experimental results show that this paper has a good effect after image fusion, and the recovered clarity is higher, which can better express the basic information of the source image; compared with several decomposition fusion methods commonly used at present, this paper has better recovery performance and detail processing ability and preserves the edge information of essential details in the image while filtering and denoising and is more excellent in objective performance evaluation indexes such as PSNR and SSIM. It can be used as a reference basis in the restoration process of art images. [ABSTRACT FROM AUTHOR]

Purpose: The low‐dose computed tomography (CT) imaging can reduce the damage caused by x‐ray radiation to the human body. However, low‐dose CT images have a different degree of artifacts than conventional CT images, and their resolution is lower than that of conventional CT images, which can affect disease diagnosis by clinicians. Therefore, methods for noise‐level reduction and resolution improvement in low‐dose CT images have inevitably become a research hotspot in the field of low‐dose CT imaging. Methods: In this paper, residual attention modules (RAMs) are incorporated into the residual encoder–decoder convolutional neural network (RED‐CNN) and generative adversarial network with Wasserstein distance (WGAN) to learn features that are beneficial to improving the performances of denoising networks, and developed models are denoted as RED‐CNN‐RAM and WGAN‐RAM, respectively. In detail, RAM is composed of a multi‐scale convolution module and an attention module built on the residual network architecture, where the attention module consists of a channel attention module and a spatial attention module. The residual network architecture solves the problem of network degradation with increased network depth. The function of the attention module is to learn which features are beneficial to reduce the noise level of low‐dose CT images to reduce the loss of detail in the final denoising images, which is also the key point of the proposed algorithms. Results: To develop a robust network for low‐dose CT image denoising, multidose‐level torso phantom images provided by a cooperating equipment vendor are used to train the network, which can improve the network's adaptability to clinical application. In addition, a clinical dataset is used to test the network's migration capabilities and clinical applicability. The experimental results demonstrate that these proposed networks can effectively remove noise and artifacts from multidose CT scans. Subjective and objective analyses of multiple groups of comparison experiments show that the proposed networks achieve good noise suppression performance while preserving the image texture details. Conclusion: In this study, two deep learning network models are developed using multidose‐level CT images acquired from a commercial spiral CT scanner. The two network models can reduce and even remove streaking artifacts, and noise from low‐dose CT images confirms the effectiveness of the proposed algorithms. [ABSTRACT FROM AUTHOR]

Remote sensing image denoising has important applications in aerospace, geophysical exploration and communication engineering. Traditional wavelet transform cannot represent the details and contours of the image texture effectively, because pseudo-Gibbs effect, ringing and blurring of edge details will be produced during de-noising. In this paper, in view of the shortcomings of the existing directional filter design, an eight-direction filter bank that directly decomposes each direction of the image using each directional subband filter is proposed, which is combined with Laplace pyramid transformation to form an optimized contour transformation. A denoising threshold processing algorithm for remote sensing images based on optimized contourlet transformation is proposed. Compared with the traditional denoising method, the improved algorithm has the characteristics of multi-scale and multi-direction, and can better capture the detailed information of the image. The improved semi-soft threshold function is used to process the transformed coefficients, which can better restore the edge information in the image. The results of a series of simulation experiments show that compared with the traditional wavelet threshold function, the contourlet hard threshold function and contourlet soft threshold function, this method achieves better visual effects and higher PSNR (peak signal to noise ratio) while the image is denoised, and the denoising effect is improved by 0.11%. The proposed new threshold function is feasible and can handle the texture information and edge information of remote sensing image well. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have performed well in image denoising. Although some CNNs enlarge convolutional kernels and increase stacked convolutional layers to overcome the locality defect of convolutional operations, they may increase computational costs. In this paper, we propose a parallel and serial denoising network (PSDNet) for image denoising to preserve image texture. Specifically, the proposed PSDNet contains a parallel block (PB), a serial block (SB), and a reconstruction block (RB). A PB uses two heterogeneous sub-networks with a deformable convolution in a parallel way to extract comparative information for better-recovering image texture. A SB utilizes an enhanced residual dense architecture via combinations of a batch normalization, ReLU, and convolutional layer in a serial way to refine obtained features for obtaining more accurate noise information. A RB is responsible for reconstructing images. Experimental results reveal that our PSDNet is very effective in image denoising, according to quantitative analysis and visual analysis. Codes can be obtained at https://github.com/hellloxiaotian/PSDNet. • Heterogeneous architecture with deformable convolution can better filter noise. • An enhanced residual architecture is used to remove redundant features. • Combining a parallel and serial way can improve effects of images denoising. • Proposed network is effective for synthesized and real noisy image denoising. [ABSTRACT FROM AUTHOR]

Many current image denoising works focus on Gaussian noise removal. However, in many real-life applications, the noise in an image can follow non-Gaussian distributions, such as compressed speckle noise, salt-and-pepper impulse noise (SPIN), and random-valued impulse noise (RVIN). In such scenarios, image denoising models with mean squared error loss will fail to obtain the best noise estimation results. Thus, we propose a non-Gaussian denoising convolutional neural network (DTNet) with truncation loss, which can address compressed speckle noise, SPIN, and RVIN with the same denoising framework. The model is composed of Res2Net blocks with the ability to extract multiscale information. Experiments show that our proposed DTNet method can effectively remove compressed speckle noise and impulse noise, including SPIN and RVIN. In particular, DTNet can recover the original images from images with 90% SPIN or images with 70% RVIN. [ABSTRACT FROM AUTHOR]

In the field of image processing, common noise types include Gaussian noise, salt and pepper noise, speckle noise, uniform noise and pulse noise. Different types of noise require different denoising algorithms and techniques to maintain image quality and fidelity. Traditional image denoising methods not only remove image noise, but also result in the detail loss in the image. It cannot guarantee the clean removal of noise information while preserving the true signal of the image. To address the aforementioned issues, an image denoising method combining an improved threshold function and wavelet transform is proposed in the experiment. Unlike traditional threshold functions, the improved threshold function is a continuous function that can avoid the pseudo Gibbs effect after image denoising and improve image quality. During the process, the output image of the finite ridge wave transform is first combined with the wavelet transform to improve the denoising performance. Then, an improved threshold function is introduced to enhance the quality of the reconstructed image. In addition, to evaluate the performance of different algorithms, different densities of Gaussian noise are added to Lena images of black, white, and color in the experiment. The results showed that when adding 0.010.01 variance Gaussian noise to black and white images, the peak signal-to-noise ratio of the research method increased by 2.58dB in a positive direction. The mean square error decreased by 0.10dB. When using the algorithm for denoising, the research method had a minimum denoising time of only 13ms, which saved 9ms and 3ms compared to the hard threshold algorithm (Hard TA) and soft threshold algorithm (Soft TA), respectively. The research method exhibited higher stability, with an average similarity error fluctuating within 0.89%. The above results indicate that the research method has smaller errors and better system stability in image denoising. It can be applied in the field of digital image denoising, which can effectively promote the positive development of image denoising technology to a certain extent. [ABSTRACT FROM AUTHOR]

With the continuous advancement of deep learning technology, U-Net–based algorithms for image denoising play a crucial role in medical image processing. However, most U-Net-based medical image denoising algorithms typically have large parameter sizes, which poses significant limitations in practical applications where computational resources are limited or large-scale patient data processing are required. In this paper, we propose a medical image denoising algorithm called AsymUNet, developed using an asymmetric U-Net framework and a spatially rearranged multilayer perceptron (MLP). AsymUNet utilizes an asymmetric U-Net to reduce the computational burden, while a multiscale feature fusion module enhances the feature interaction between the encoder and decoder. To better preserve the image details, spatially rearranged MLP blocks serve as the core building blocks of AsymUNet. These blocks effectively extract both the local and global features of the image, reducing the model's reliance on prior knowledge of the image and further accelerating the training and inference processes. Experimental results demonstrate that AsymUNet achieves superior performance metrics and visual results compared with other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

In the present era, transfer of digital images via communication channel is affected by noise. Received images need to be processed so that noise is eliminated, which requires information about the noise and the image. Noise is an unwanted signal that interferes with the original signal and reduces the quality of the digital image. In this article, a noise elimination algorithm is applied where the image is preserved by preservation of the edges and image texture details, without fading during the image smoothing process. Eliminating of the noise from the image by preserving the edges is becoming one of the researched after topic in digital image transfer. Many noise elimination techniques have been developed for image processing and computer visual communities. Although these methods appear to be very different by using compound filter banks to create an uncompromising feature to maintain meaningful edges, we find better results than other traditional methods. In this article, we have removed the noise filter relatively and efficiently using the Dmeyer Wavelets Transform, Hybrid Wavelets Directional and Wavelet transforms by applying the Caesar window, and improved the edges of the image. [ABSTRACT FROM AUTHOR]

Noise is one of the main sources of quality deterioration not only for visual inspection but also in computerized processing in brain magnetic resonance (MR) image analysis such as tissue classification, segmentation and registration. Accordingly, noise removal in brain MR images is important for a wide variety of subsequent processing applications. However, most existing denoising algorithms require laborious tuning of parameters that are often sensitive to specific image features and textures. Automation of these parameters through artificial intelligence techniques will be highly beneficial. In the present study, an artificial neural network associated with image texture feature analysis is proposed to establish a predictable parameter model and automate the denoising procedure. In the proposed approach, a total of 83 image attributes were extracted based on four categories: 1) Basic image statistics. 2) Gray-level co-occurrence matrix (GLCM). 3) Gray-level run-length matrix (GLRLM) and 4) Tamura texture features. To obtain the ranking of discrimination in these texture features, a paired-samples t-test was applied to each individual image feature computed in every image. Subsequently, the sequential forward selection (SFS) method was used to select the best texture features according to the ranking of discrimination. The selected optimal features were further incorporated into a back propagation neural network to establish a predictable parameter model. A wide variety of MR images with various scenarios were adopted to evaluate the performance of the proposed framework. Experimental results indicated that this new automation system accurately predicted the bilateral filtering parameters and effectively removed the noise in a number of MR images. Comparing to the manually tuned filtering process, our approach not only produced better denoised results but also saved significant processing time. [ABSTRACT FROM AUTHOR]

In order to improve the image quality, in this paper, we propose an improved PM model. In the proposed model, we introduce two novel diffusion coefficients and a residual error term and replace the integer differential operator with the fractional differential operator in the PM model. The diffusion coefficients can be used effectively for edge detection and noise removal. The residual error term can help to prevent image distortion. Fractional order differential operator has a good characteristic that it can enhance image texture information while removing image noise. Additionally, in the two new diffusion coefficients, a novel method is proposed for automatically setting parameter k, and it does not need to do any experiments to get the value of k. For the computing fractional order diffusion coefficient, we employ the discrete Fourier transform, and an iterative scheme is carried out in the frequency domain. In the proposed model, not only is the integer differential operator replaced with the fractional differential operator, but also the order of the fractional differentiation is determined adaptively with the local variance. Comparing with some existing models, the experimental results show that the proposed algorithm can not only better suppress noise, but also better preserve edge and texture information. Moreover, the running time is greatly reduced. [ABSTRACT FROM AUTHOR]

Infrared images are susceptible to interference from a variety of factors during acquisition and transmission, resulting in the inclusion of mixed noise, which seriously affects the accuracy of subsequent vision tasks. To solve this problem, we designed a mixed noise removal algorithm for infrared images based on improved CycleGAN. First, we proposed a ResNet-E Block that incorporates the EMA (Efficient Multi-Scale Attention Module) and build a generator based on it using the skip-connection structure to improve the network's ability to remove mixed noise of different strengths. Second, we added the PSNR (Peak Signal-to-Noise Ratio) as an extra calculation item of cycle consistency loss, so that the network can effectively retain the detailed information of infrared images while denoising. Finally, we conducted experimental validation on both synthetic noisy images and real noisy images, which proved that our algorithm can effectively remove the mixed noise in infrared images and the denoising effect is better than other similar methods. [ABSTRACT FROM AUTHOR]