580 results about "Image denoising" patented technology

The invention relates to the field of image data processing, and discloses a texture preserving type image denoising and enhancing method based on a generative adversarial network, which comprises the following steps: acquiring an original image signal, and calculating low-frequency sub-band data and high-frequency sub-band data by using discrete wavelet transform; calculating the gradient magnitude of the low-frequency sub-band data to generate a structural significance gradient map; establishing a reverse mapping relation based on the structure saliency gradient map, and generating a spatial self-adaptive dynamic gating threshold; performing statistical gating on the high-frequency sub-band data by using the dynamic gating threshold to generate a high-pass gain coefficient and a low-pass suppression coefficient; according to the method, cross-band modulation logic of the structure flow to the texture flow is established, so that the problem that weak texture signals are easy to lose under non-uniform illumination is solved, and non-structured noise filtering and structured micro texture restoration are realized on the premise of not depending on semantic tags.

The invention discloses a wafer probe mark accurate detection method based on deep learning, and belongs to the field of wafer probe mark detection, and the method comprises the steps: constructing a pin mark image denoising preprocessing network, employing a small target feature protection and enhancement strategy based on HSV color space and local contrast joint adjustment, and carrying out the recognition of a small target feature; denoising and contrast optimization are carried out on the needle mark image; a multi-scene training sample is generated through mosaic splicing and mix fusion; a dense small target enhancement module is introduced into the backbone network to enhance needle mark feature expression, and a multi-scale feature fusion module is arranged in the neck network to extract full-scale features; and establishing an anchor frame optimization system adaptive to the minimum needle mark target, and adopting an optimizer and learning rate collaborative optimization training strategy to realize model adaptive convergence. According to the method, high-precision detection and robust identification of the wafer probe mark can be realized under a complex background, and the detection accuracy and stability are remarkably improved.

The invention discloses an artificial intelligence denoising method for a speckle shearing interference image based on self-supervised learning. The method is a self-supervised image denoising method based on blind spot learning. According to the method, an improved self-supervised network architecture is adopted, pixel points of an image are input through a random mask and are replaced by neighborhood pixel values, and a blind spot training sample is constructed. The network only learns and predicts an original noise value of a masked pixel point in a training process, and does not depend on a clean image as a supervision signal, so that real self-supervised learning is realized. The de-noising method comprises the steps of constructing a data set meeting training requirements, performing de-noising model training by using a blind spot learning method and a self-supervised network, performing de-noising calculation on a speckle interferogram through a fully trained model, and finally obtaining a de-noised interference image. The method has the characteristics of simplicity and convenience in calculation and high processing speed, and can meet the requirement of large-data-volume image processing.

The invention provides a medical image denoising and segmentation integrated model trained based on a conductible diffusion method, and belongs to the field of medical image processing, and the model comprises an improved denoising and diffusion module which is used for receiving an initial medical image, predicting an original clean signal of the initial medical image based on an improved denoising and diffusion probability model, and obtaining a final denoised medical image; the joint motion and segmentation module is used for receiving the initial medical image, extracting features through a shared encoder, and performing joint learning through a motion estimation branch and a segmentation branch to obtain segmentation data; the cascade training mechanism carries out end-to-end training on the improved de-noising diffusion module and the joint motion and segmentation module through a derivable connection, gradient back propagation is realized by using a joint loss function, and cascade training is completed. According to the method, the problem that the quality of segmented medical images is reduced due to the fact that denoising and segmentation tasks cannot be collaboratively optimized due to the fact that a traditional denoising method cannot be guided in sampling and is difficult to carry out cascade training with a segmentation model is solved.

The invention discloses a post-stroke depression risk prediction system based on cerebral small vascular diseases and inflammatory markers, and relates to the technical field of computer-aided engineering, the post-stroke depression risk prediction system comprises: a data acquisition module acquires CSVD image data, serum inflammatory markers and clinical baseline information; the data preprocessing module processes image denoising registration, fills up marker missing values and encodes clinical information; the CSVD feature extraction module extracts image omics features and quantifies severity; the inflammation marker module calculates statistical characteristics and inflammation intensity; the multi-dimensional fusion module integrates features and eliminates redundancy; the risk prediction module is trained by using an improved attention CNN-LSTM model; the result output module visualizes risk and intervention suggestions; the model dynamic optimization module updates parameters by incremental learning. According to the method, multi-source key data are integrated, the prediction accuracy and generalization ability are improved, clinical interpretation and dynamic adaptability are achieved, early recognition of post-stroke depression is assisted, and patient prognosis is improved.

The invention relates to an adaptive BM3D terahertz medical image denoising method based on KSVD and SSIM optimization, and belongs to the technical field of image processing. In the basic estimation stage, firstly, an 8 * 8 reference block is selected in a noise image, similar blocks are searched in a neighborhood with the reference block as the center to form a similar block group, and noise is removed and image details are recovered through hard filter value filtering processing. And finally, the BM3D algorithm performs weighted average processing on the estimation blocks in all the groups to generate a final de-noised image. And carrying out KSVD noise reduction on the image subjected to basic estimation, inputting the image subjected to noise reduction and an original noise image into final estimation, and carrying out block matching grouping, Wiener filtering and aggregation to form a final result. According to the method, local sparsity and non-local similarity of pictures can be fully utilized, cross-regional redundant information is utilized to suppress noise, the noise reduction capability is improved, meanwhile, more structural similarity is reserved, and a better effect is achieved in the face of high noise.

The invention relates to an automatic milling cutter setting method and system based on machine vision, and belongs to the technical field of milling cutter setting. The method comprises the following steps: firstly, positioning initial position coordinates of a milling cutter, and planning an initial tool setting path of the milling cutter by combining target tool setting position coordinates; then obtaining an image of the milling cutter in the initial cutter setting path, carrying out image denoising and image deblurring processing, carrying out edge detection after obtaining a second image, extracting an edge contour of the milling cutter, calculating sub-pixel coordinates of edge points of the contour, and carrying out parametric fitting to obtain a current milling cutter position and a current milling cutter posture; inputting the initial tool setting path, the wear degree of the milling cutter, the current position of the milling cutter and the posture of the milling cutter into an error prediction model to predict the current motion error of the milling cutter; calculating the path compensation amount according to the current motion error of the milling cutter, and adjusting the tool setting path of the milling cutter according to the compensation amount. The method can reduce the interference of the motion blur of the milling cutter and environmental factors, and realizes the quantitative adjustment and correction of the tool setting of the milling cutter.

The invention discloses a three-dimensional visualization head and neck tumor preoperative imaging system based on deep learning. The method comprises the following steps: image analysis and preprocessing; carrying out image denoising and spatial-temporal feature extraction based on a deep learning convolutional neural network architecture, and introducing a u-Net model to realize image segmentation and semantic annotation; performing three-dimensional reconstruction and surface rendering by adopting a Marking Cubes algorithm, and realizing visual interaction image presentation and personalized operation by combining volume rendering and surface rendering; and data is output through multiple ports, and data viewing and interactive operation of a web terminal and a mobile terminal are supported. The method has the advantages that according to the technical scheme, through fusion of a three-dimensional attention mechanism and multi-phase data, the tumor segmentation Dice coefficient reaches 0.92 + / -0.03, and the method is superior to a traditional threshold segmentation method.

The invention discloses an ultrasonic image denoising method based on variational mode decomposition and local space sparse fusion, which comprises the following steps of: firstly, adaptively optimizing key parameters of variational mode decomposition by using a grey wolf optimization algorithm to realize stable and efficient decomposition of an ultrasonic image; then, classifying the modal components according to the structural features of the modal components, and implementing differentiated denoising strategies for different types of modals to separate noise and reserve useful information; after modal reconstruction, a sparse expression method based on local space information is further introduced, according to the method, accurate boundary detection is carried out through gradient vector flow and gray scale proportion analysis, self-adaptive partitioning is carried out on an image according to boundary information, and finally sparse reconstruction is carried out through double dictionaries trained for different areas. According to the method, speckle noise in the ultrasonic image can be effectively suppressed, and meanwhile, the capability of keeping the edge and detail information of a tissue structure is remarkably improved, so that the ultrasonic image with higher quality is obtained.

The invention provides a brain image denoising method based on a U-shaped dynamic convolution multi-scale multi-branch network. The brain image denoising method comprises the steps that brain noise images to be denoised are input into three branch networks formed by U-netAM, DSHFN and MSDSRN in parallel; in the U-netAM branch, multi-level features are extracted through an encoder in sequence, after weighting is conducted through a channel and a space attention mechanism, the spatial resolution is recovered through a decoder, and a first feature map is obtained; in the DSHFN branch, the dynamic convolution kernel generates a corresponding convolution kernel in real time according to input image features, the image is decomposed into a low-frequency part and a high-frequency part, and the low-frequency part and the high-frequency part are subjected to weighted fusion after being processed by a low-pass filter and a high-pass filter respectively to obtain a second feature map; in the MSDSRN branch, adopting multi-scale depth separable convolution to extract multi-scale features in parallel, and obtaining a third feature map through residual connection and fusion; inputting the three feature maps into an FPB block for fusion to obtain a noise feature map; and performing pixel-by-pixel subtraction on the original brain noise image and the noise feature map, and outputting a denoised brain image.

The invention discloses a real scene-oriented double-branch self-supervised image denoising method, which comprises the following steps of: decomposing a noise image into a low-frequency sub-band and a high-frequency sub-band, processing noise in the high-frequency sub-band by adaptively adjusting a soft threshold and a hard threshold, reconstructing the noise image, and extracting frequency domain denoising features through a residual block network; the noise image is masked, the masked image is input into the U-Net backbone network, and the U-Net backbone network outputs spatial domain denoising features; and fusing the frequency domain denoising features and the spatial domain denoising features, and inputting the fused features into a convolutional network to obtain a denoised noise image. A U-Net structure is adopted in a spatial domain branch, the detail retention capability is enhanced through multi-scale feature extraction and dynamic mask generation, meanwhile, a frequency domain branch is introduced, and noise suppression is carried out through Haar wavelet transform, so that the limitation that U-Net only depends on spatial domain processing is made up.

The invention discloses a concrete defect image generation method based on mask guidance. The method is suitable for data enhancement and structured image synthesis tasks of defect images such as concrete cracks under the small sample condition. According to the method, a space mask mechanism is introduced, a defect area is accurately positioned and extracted from a reference image, and the injection accuracy and control granularity of defect features are effectively improved. The method comprises three main stages: a multi-scale theme-background feature extraction stage, an adaptive time feature generation stage and a multi-scale cross attention image denoising stage. In the feature extraction stage, spatial attention enhancement and fine-grained semantic alignment are carried out on a reference image and a subject text, so that unified representation of multi-modal features is realized; in the time feature generation stage, feature adaptive time is introduced, and time step related weights are generated to dynamically regulate and control the fusion proportion of theme and background features; in the image denoising stage, guidance denoising and image generation are completed in combination with UNet and cross attention.

The invention belongs to the technical field of image processing, and particularly relates to an image denoising processing system based on multistage filtering collaboration, which comprises a noise feature analysis module, an image semantic analysis module, a collaboration strategy generation module, a multi-path parallel filtering module, a self-adaptive weighted fusion module and a perception quality refining module. According to the invention, blind noise evaluation is carried out on the lightweight convolutional neural network, the input image is preprocessed, the image is divided into different semantic regions, and then noise information and the semantic regions are matched through a filter library and a plurality of filters optimized for different features. And dynamically selecting the most suitable filter for each semantic partition, calling an adaptive median filter to process a smooth region containing impulse noise, and ensuring that each filter strictly performs denoising operation according to a region and parameters specified by a strategy set, so that the filtering strategy is changed from blindness to intelligence, and the filtering accuracy is improved. And the visual bolster degree and the intelligent level of the denoising effect are improved.

The invention provides a noise modeling and denoising method for a low-light-level image intensifier. The noise modeling and denoising method comprises the following steps: step 1, establishing an imaging noise model for the low-light-level image intensifier; step 2, collecting output data of the low-light image intensifier under different illumination conditions, calibrating parameters of an imaging noise model, and superposing the calibrated imaging noise model on a noise-free image sample to construct a training data set conforming to actual imaging characteristics; step 3, designing an image denoising network based on the convolutional neural network, the input of the network being a noise-containing image, and the output being a noise-suppressed clean image; and step 4, training and optimizing the convolutional neural network based on the training data set, and verifying synthetic data and real data. According to the method, the adaptive capacity and generalization performance of the neural network to complex noise can be effectively enhanced, and finally, more accurate and stable image enhancement and denoising effects can be realized in a low-light imaging scene.

The invention discloses a method for performing visual inspection on surface forming quality characteristics of a welding seam by utilizing three-dimensional reconstruction, which comprises the following steps of: S1, performing uniform-speed scanning on the welding seam along the direction of a welding bead by using linear structured light emitted by a linear laser, and performing image acquisition on linear structured light stripes formed on the surface of the welding seam by using an industrial camera; s2, converting the acquired RGB image into a gray level image frame by frame, and carrying out image distortion correction and image denoising processing; s3, carrying out ROI (Region of Interest) positioning on the formed line structured light stripe image on the surface of the welding seam by adopting a pixel point gray level distribution calculation method; s4, carrying out image segmentation on the ROI region of the weld line structured light stripe image; and S5, carrying out sub-pixel-level center line extraction on the ROI of the weld line structured light stripe image by adopting a gray extreme value weighted centroid method. Automatic detection of welding seam forming quality characteristics such as laser welding and electric arc welding can be achieved, and the operation efficiency of a production line and the welding seam quality detection precision can be greatly improved.

The invention discloses a frequency modulation and wavelet sub-band guided double-domain collaborative Transformer X-ray image denoising method, which comprises the following steps of: acquiring a noise-containing digital ray original image and a corresponding clear reference image, and constructing a data set after preprocessing the noise-containing digital ray original image and the corresponding clear reference image; constructing a network model of a double-domain collaborative coding-decoding architecture; performing 3 * 3 deep convolution on an input image to extract shallow layer features; in the encoding stage, ETB and AFMB are alternately stacked to represent local and global information, a WB-LKED module is embedded to strengthen fine-grained features, and WDB executes down-sampling and transmits high-frequency features to a decoding end; in the decoding stage, the WUB recovers the resolution through double-path up-sampling, integrates the same-scale features of an encoder, enhances details by using high-frequency features, splices the features, then carries out ETB and AFMB refining, obtains output features through 3 * 3 deep convolution, and combines a global residual error connection optimization result; and training the model by using the data set, inputting a to-be-denoised image, and outputting a final result. According to the method, the problems of weak complex noise interference resistance, poor detail retention effect and limited CNR improvement can be solved.

The invention discloses a construction safety monitoring system for house construction, which belongs to the technical field of safety monitoring and comprises a construction monitoring data integration module, a construction monitoring image denoising module, a construction safety monitoring model building module and a real-time construction safety monitoring module. The method specifically comprises the following steps: respectively extracting three types of noise features and obtaining a weight coefficient, cascading a double-branch enhanced feature map and connecting the feature map with an initial feature map residual error, fusing a security mask, constructing a three-scale encoder and an asymmetric decoder, mapping a deep fusion feature into a noise feature and removing the noise feature, and providing a clear and complete image data support; dividing personnel feature map windows based on double scales, performing weighted fusion on same-scale features and aligning resolutions, extracting local, mesoscale and global features in parallel, weighting query vectors element by element by using importance weights of all positions, screening effective spatial features through convolution and gating weights, and obtaining a query result; and the real-time performance and the accuracy of house construction safety monitoring are improved.

The invention discloses a two-stage Raman hyperspectral imaging method, which relates to the technical field of spectral signal analysis and comprises the following steps: S0, preprocessing Raman hyperspectral data to remove abnormal values; the method comprises the following steps: S1, taking a wave number where a target characteristic peak is located as a target wave number, and based on a neighborhood of the target wave number in Raman hyperspectral data, performing spectrum denoising on the Raman hyperspectral data by using an adaptive low-rank matrix approximation algorithm to obtain the Raman hyperspectral data after spectrum denoising; and S2, using an improved BM3D method based on a rotating block to carry out image denoising on a Raman image after Raman imaging is carried out on a target wave number in the Raman hyperspectral data after spectrum denoising. According to the method, the neighborhood of the target wave number is used as the target area, the adaptive low-rank matrix approximation algorithm is used for spectrum denoising, then the improved BM3D method based on the rotating block is used for image denoising, Raman hyperspectral fast and efficient imaging is achieved, and the method has the advantages of being good in denoising effect, high in speed and light in weight.

This invention discloses a method and system for hyperspectral image denoising based on a spatial-spectral joint self-attention mechanism, belonging to the field of image processing technology. First, based on the characteristics of hyperspectral images, a spatial-spectral joint self-attention mechanism network is constructed. The noisy hyperspectral image is used as input to the network to extract spatial-spectral features. Then, a global spectral self-attention mechanism is used to extract the band correlations of the hyperspectral image. Finally, the extracted spatial-spectral features are reconstructed using a multiple perceptron and residual connections to reconstruct a clean, noise-free hyperspectral image. The system includes a feature extraction subsystem, a non-local spatial self-attention subsystem, a global spectral self-attention subsystem, and an image reconstruction subsystem. This invention can effectively restore noisy hyperspectral images to obtain noise-free hyperspectral images. Compared to convolutional networks, it can better model long-range dependency information and has better adaptability to target hyperspectral images.

The embodiment of the invention provides an image denoising method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring a to-be-processed image, and extracting low-frequency features and high-frequency features of the to-be-processed image; determining a first parameter corresponding to a preset scale according to the high-frequency feature and the preset scale; the first parameter represents the position of an object in the to-be-processed image represented in the high-frequency feature; determining a second parameter and a third parameter corresponding to a preset scale according to the low-frequency feature and the preset scale; the second parameter represents the position of an object in the to-be-processed image represented in the low-frequency feature; the third parameter represents image details of an object in the to-be-processed image represented in the low-frequency feature; determining a target image according to the high-frequency features and the first parameter, the second parameter and the third parameter corresponding to each preset scale; the target image is an image obtained after the to-be-processed image is subjected to noise removal. The method is used for improving the image denoising precision.

Methods and apparatus for end-to-end unsupervised multi-document blind image denoising is presented. The multi-document blind image denoiser removes various noise types from noisy documents without paired target cleaned documents and preserves the contents for optical character recognition. The end-to-end unsupervised multi-document blind image denoiser integrates a Mixture of Experts with a cycle-consistent GAN as the base network that effectively removes multiple types of noise, including salt & pepper noise, blurred and / or faded text, as well as watermarks from documents at various levels of intensity.

The invention discloses a retina OCT (Optical Coherence Tomography) image denoising method based on zero sample learning. The method comprises the following steps: generating a noise independent image pair for a single noisy image by adopting a CDIS (Coherent Discrete Identifier) The method comprises the following steps of: carrying out de-noising by using an MLFSnet network (AF-MSDSConv, Aap-LSM, SA, a reconstruction module); using RMSE symmetry and consistency joint loss zero sample training; in the reasoning stage, an original image is directly input, and a same-resolution de-noising result is output. The speckle noise can be significantly suppressed and the layered structure of the retina can be retained without noise-free true values or pairwise data, so that the retina can be used while being shot.

The invention discloses a visible light image and sound wave data-based visual fusion display method and system, and relates to the technical field of image and sound wave fusion display, and the method comprises the steps: data collection: collecting image frames through an industrial visible light camera, collecting sound wave data through an eight-channel microphone array, synchronizing timestamps, and enabling a dynamic scene to be high in frame rate; a preprocessing step: denoising and sharpening the image, filtering interference by sound waves, and carrying out space-time registration; a feature extraction step: extracting image edge texture and sound wave frequency energy features; a fusion calculation step: dynamically adjusting the weight and generating a fusion feature matrix in combination with an attention mechanism; and a visual display step: constructing 2D, 3D and AR interfaces, superposing sound wave information and supporting multi-modal interaction and parameter control. According to the method, the image and sound wave fusion accuracy is improved, scene adjustment parameters are dynamically adapted, multi-dimensional display is visual, interaction is convenient and fast, the exception recognition and response capability is enhanced, and the method adapts to multi-scene requirements.

The invention relates to the technical field of computer vision and image processing, and provides a tiny motion trail tracking method based on an optical flow method and spatial filtering amplification. The objective of the invention is to solve the problems of low trajectory tracking precision, poor real-time performance and insufficient environmental adaptability caused by weak tiny motion signals in the existing shooting training video analysis technology, and the method comprises the following steps: receiving a video frame sequence, and carrying out brightness normalization and image denoising preprocessing on each frame; based on a predefined reference template, dynamically positioning a gun taste region of interest ROI through a matching method; the ROI is subjected to spatial filtering amplification processing, an image frequency layer is decomposed on the spatial scale, a change signal within a specific frequency range is amplified on the time scale, and an enhanced image is reconstructed to highlight tiny motion; applying an optical flow method, including a dense optical flow algorithm or a sparse optical flow algorithm, on the enhanced ROI region to estimate a motion vector; and constructing a trajectory point sequence based on the motion vector, carrying out smoothing processing, and outputting smoothed trajectory data.

The invention provides an image noise reduction method and system, computer equipment and a storage medium, and belongs to the field of image processing, and the method comprises the steps: firstly converting an original strip mine coal rock image into a gray level image, and carrying out the image processing of the gray level image for the problem of edge blurring caused by mixed noise in the strip mine coal rock image; the method comprises the following steps: firstly, extracting a noise model, analyzing the noise type and distribution characteristics of the noise model to implement preliminary noise reduction, filtering out small-scale noise of an image subjected to preliminary noise reduction through Gaussian-Laplacian joint transformation to obtain an enhanced image, then calculating a gradient magnitude image of the enhanced image, dynamically generating a parameter matrix in combination with a normalized gradient value, and constructing an adaptive generalized overall variation noise reduction model; and taking the gradient amplitude image and the self-adaptive parameter matrix as input, updating the optimal estimation image through iterative optimization, and obtaining an optimal noise reduction image when the iteration energy variation is smaller than a threshold value. According to the method, while mixed noise is removed, coal rock texture details are remarkably reserved, and high-quality data support is provided for mining decisions of strip mines.

The invention provides a multi-module fusion self-supervision denoising method for an industrial CT image, and belongs to the field of digital image processing and industrial nondestructive testing. In order to solve the problems of detail loss, poor denoising effect and the like in an industrial CT image denoising process, the invention provides a self-supervised denoising model which takes U-net as a trunk network and fuses an edge enhancement module, a convolution block attention module and a convolution attention fusion module. Meanwhile, by means of a mixed loss function composed of structural similarity index loss, mean square error loss and ResNet perception loss, the noise level of the industrial CT image can be remarkably reduced, meanwhile, the integrity of the object structure and the contour edge is accurately kept, and the image quality is effectively improved.

The present disclosure relates to method and apparatuses for denoising an image. One example method includes receiving an input image captured by an image sensor, implement a trained artificial intelligence model to form an estimate of a noise pattern in the input image, to form an estimate of at least one noise statistic for the image sensor, and to refine the estimate of the noise pattern based on the estimate of the at least one noise statistic, and form an output image by subtracting the refined estimate of the noise pattern from the input image.

This application discloses a computing system to obtain a wafer image of an electronic device having physical structures manufactured using one or more lithographic masks associated with a layout design describing the electronic design. The computing system can implement an unsupervised deep learning algorithm to process the wafer image to remove at least some noise from the wafer image, which generates a denoised wafer image. The computing system can extract contours corresponding to the physical structures of the electronic device from the denoised wafer image of the electronic device without use of the layout design or a mask design. The computing system can calibrate the layout design or the mask design describing the one or more lithographic masks based, at least in part, on the contours extracted from the denoised wafer image.

The invention relates to a video coding image noise reduction processing method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring an original video frame image and a step length parameter; dividing the original video frame image into a plurality of pixel units according to the step length parameter; calculating local gradient data for each pixel unit to obtain a local gradient matrix; generating a filtering template according to the local gradient matrix; wherein the filtering template is used for representing a filtering weight of a pixel in the pixel unit; and performing weighted filtering processing on the pixel units according to the filtering template. By adopting the method, the quality of the video image to be coded can be improved.

The invention relates to an image denoising method and device, equipment and a medium, and the method comprises the steps: obtaining a training data set which comprises a plurality of image sample pairs, and each image sample pair comprises a noise sample image and a noise-free supervision image corresponding to the noise sample image; dividing the image sample pair according to a preset size to determine image blocks corresponding to the noisy sample image and the noiseless supervision image; the method comprises the following steps of: embedding a plurality of improved Transform Blocks in each coding layer and each decoding layer of a preset U-Net trunk structure, and introducing a mixed feature compensation module at an input end and an output end in the U-Net trunk structure to construct an image denoising model; and inputting a to-be-denoised image into the image denoising model trained to the convergence state to determine a clean image corresponding to the to-be-denoised image so as to complete image denoising. According to the invention, the generalization ability and stability of image denoising in a complex real scene can be greatly improved.