Image noise reduction method and system and readable storage medium

Abstract

The invention relates to an image noise reduction method and system and a readable storage medium, and the method comprises the steps: carrying out the Gaussian filtering of a current frame image, carrying out the down-sampling, obtaining a first down-sampling image, and carrying out the Gaussian pyramid decomposition, and obtaining a current frame image group; performing up-sampling to obtain an up-sampling image, and subtracting the up-sampling image from the current frame image to obtain high-frequency information; motion estimation and texture estimation under different scales are carried out on the current frame image group and the reference frame image group, and guided filtering correction is carried out; performing time-space domain filtering on the first down-sampling image and the reference frame image based on the motion estimation weight; performing up-sampling on the down-sampling image after noise reduction to obtain an up-sampling image after noise reduction, and performing weighted fusion on the up-sampling image after noise reduction and the high-frequency information according to the texture estimation weight to obtain a final image; and performing down-sampling on the final image and performing Gaussian pyramid decomposition to obtain a reference frame image group for noise reduction of the next frame image. According to the method, more texture details can be reserved while the signal-to-noise ratio of the image is increased.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image denoising method, system and readable storage medium. Background Art

[0002] Various cameras based on CMOS sensors will inevitably be affected by noise during the imaging, storage and transmission process of video recording, thereby reducing the image quality. Image noise reduction is to remove noise through the current signal, improve the image signal-to-noise ratio, and restore the information in the real scene. For video sequences, it can be regarded as noise reduction of multiple frames. Since there is a correlation between different frames, time domain noise reduction is also used to improve the image signal-to-noise ratio. Generally, video noise reduction can be divided into two parts: time domain noise reduction and spatial domain noise reduction.

[0003] Spatial domain denoising algorithms are generally divided into filter, local similarity, transform domain, and data-driven methods. Filter-based methods generally use low-pass filters to filter images, filter out high-frequency noise in the image, and obtain images with less noise, such as Gaussian filters and mean filters. Local similarity-based methods use the fact that there are many similar textures in the neighborhood of the image, which have certain correlations. With the help of similarity judgment, weighted averaging is performed to improve the image signal-to-noise ratio. Transform domain-based methods convert image signals to other signal domains such as frequency, filter out the high-frequency characteristics of noise signals in this domain, and then convert them back to the spatial domain to obtain images with higher signal-to-noise ratios. Data-driven methods are represented by deep learning, which allows neural networks to learn the fitting expression of noise based on a large amount of data, and then forward infer the input online to obtain a noisy image or a clean image with noise removed, thereby improving the signal-to-noise ratio.

[0004] The time domain noise reduction algorithm is generally implemented by superimposing multiple frames. For example, the IIR filter is used to average the past multiple frames to obtain a better signal-to-noise ratio. There are also caches of the previous and next frames of the current frame, and the coherence is used to predict the motion vector of the current frame image, thereby assisting in better time domain noise reduction.

[0005] Generally, it is difficult to completely suppress the noise of video data only through temporal noise reduction or spatial noise reduction. Therefore, it is necessary to combine temporal noise reduction and spatial noise reduction at the same time to better improve the signal-to-noise ratio while ensuring image resolution. However, this temporal and spatial noise reduction method requires caching at least one full frame of reference frame images to achieve temporal noise reduction. In the era of increasing video resolution, the bandwidth pressure for actual hardware devices is increasing day by day, which poses a serious challenge to the power consumption performance of the device. At the same time, the accuracy of motion estimation also affects the performance of temporal and spatial noise reduction. Some weak texture areas with poor signal-to-noise ratio are difficult to be screened out, resulting in the loss of texture details. Summary of the invention

[0006] The object of the present invention is to provide an image denoising method, system and readable storage medium, aiming to reduce the bandwidth required for reference frame image caching, while improving the image signal-to-noise ratio and removing large particle noise without reducing image resolution.

[0007] To achieve the above object, the present invention provides an image denoising method, comprising:

[0008] Performing Gaussian filtering on the current frame image and performing a first downsampling to obtain a first downsampling image, and performing Gaussian pyramid decomposition on the first downsampling image to obtain a current frame image group including a plurality of scale images;

[0009] Upsampling the first downsampled image to obtain an upsampled image, and subtracting the current frame image from the upsampled image to obtain high frequency information of the current frame image;

[0010] Performing motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and correcting the motion estimation weights and texture estimation weights at each scale by guided filtering;

[0011] Performing spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a downsampled image after noise reduction;

[0012] Upsampling the denoised downsampled image to obtain a denoised upsampled image, and weightedly fusing the denoised upsampled image with the high-frequency information according to a texture estimation weight at a corresponding scale to obtain a final image;

[0013] The final image is downsampled and decomposed using a Gaussian pyramid to obtain a reference frame image group for denoising the next frame image.

[0014] Optionally, the maximum size of the reference frame image group is the image size after the reference frame image is down-sampled for the first time.

[0015] Optionally, the motion estimation is performed using an inter-frame difference method, an optical flow method, or a density-based clustering algorithm.

[0016] Optionally, an edge extraction operator is used to perform the texture estimation.

[0017] Optionally, when correcting the motion estimation weights and texture estimation weights at each scale by guided filtering, the weight information is calculated and corrected layer by layer from the smallest scale to the largest scale.

[0018] Based on the same technical concept, the present invention also provides an image noise reduction system, comprising:

[0019] A first decomposition module is configured to perform Gaussian filtering on the current frame image and perform a first downsampling to obtain a first downsampled image, and perform Gaussian pyramid decomposition on the first downsampled image to obtain a current frame image group including a plurality of scale images;

[0020] an upsampling module, configured to upsample the first downsampled image to obtain an upsampled image, and to perform a difference between the current frame image and the upsampled image to obtain high frequency information of the current frame image;

[0021] an estimation module, configured to perform motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and to correct the motion estimation weights and texture estimation weights at each scale by guided filtering;

[0022] a filtering module configured to perform spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a downsampled image after noise reduction;

[0023] a fusion module configured to upsample the denoised downsampled image to obtain a denoised upsampled image, and weightedly fuse the denoised upsampled image with the high-frequency information according to a texture estimation weight at a corresponding scale to obtain a final image;

[0024] The second decomposition module downsamples the final image and performs Gaussian pyramid decomposition to obtain a reference frame image group for denoising the next frame image.

[0025] Optionally, the maximum size of the reference frame image group is the image size after the reference frame image is down-sampled for the first time.

[0026] Optionally, the motion estimation is performed using an inter-frame difference method, an optical flow method, or a density-based clustering algorithm.

[0027] Optionally, an edge extraction operator is used to perform the texture estimation.

[0028] Based on the same technical concept, the present invention also provides a readable storage medium on which a computer program is stored. When the computer program is executed, the image noise reduction method as described above can be implemented.

[0029] In an image denoising method, system and readable storage medium provided by the present invention, by performing motion estimation and texture estimation on multi-scale images and correcting the weights at each scale through guided filtering, more accurate motion estimation weights and texture estimation weights can be obtained, thereby performing spatiotemporal filtering and fusing high-frequency information, while increasing the image signal-to-noise ratio and retaining more texture details, thereby obtaining a denoising result with a better signal-to-noise ratio and more texture. In addition, the reference frame image group used for denoising the next frame of the image is obtained by performing Gaussian pyramid decomposition after downsampling the current image, which can effectively save transmission bandwidth. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Those skilled in the art will appreciate that the accompanying drawings are provided for a better understanding of the present invention and do not constitute any limitation on the scope of the present invention.

[0031] Figure 1 The present invention provides a flowchart of an image noise reduction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, advantages and features of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the accompanying drawings are in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structure, proportion, size, etc. illustrated in the drawings of this specification are only used to match the content disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention. Any modification of the structure, change in the proportional relationship or adjustment of the size, under the same or similar conditions as the effects that can be produced by the present invention and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention.

[0033] As used in the present invention, the singular forms "a", "an", and "the" include plural referents unless the context clearly indicates otherwise. As used in the present invention, the term "or" is generally used in a sense that includes "and / or", unless the context clearly indicates otherwise. As used in the present invention, the term "several" is generally used in a sense that includes "at least one", unless the context clearly indicates otherwise. As used in the present invention, the term "at least two" is generally used in a sense that includes "two or more", unless the context clearly indicates otherwise.

[0034] Please refer to Figure 1 The embodiment of the present invention provides an image noise reduction method, comprising the following steps:

[0035] S1, performing Gaussian filtering on the current frame image and performing a first downsampling to obtain a first downsampling image, and performing Gaussian pyramid decomposition on the first downsampling image to obtain a current frame image group including multiple scale images;

[0036] S2, upsampling the first downsampled image to obtain an upsampled image, and subtracting the current frame image from the upsampled image to obtain high frequency information of the current frame image;

[0037] S3, performing motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and correcting the motion estimation weights and texture estimation weights at each scale by guided filtering;

[0038] S4, performing spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a downsampled image after noise reduction;

[0039] S5, upsampling the denoised downsampled image to obtain a denoised upsampled image, and weightedly fusing the denoised upsampled image with the high-frequency information according to a texture estimation weight at a corresponding scale to obtain a final image;

[0040] S6. Down-sample the final image and perform Gaussian pyramid decomposition to obtain a reference frame image group for denoising the next frame image.

[0041] This embodiment performs motion estimation and texture estimation on multi-scale images, and corrects the weights at each scale through guided filtering, so as to obtain more accurate motion estimation weights and texture estimation weights, thereby performing spatiotemporal filtering and fusing high-frequency information, thereby increasing the image signal-to-noise ratio while retaining more texture details, thereby obtaining a denoising result with a better signal-to-noise ratio and more texture. In addition, the reference frame image group used for denoising the next frame of the image is obtained by downsampling the current image and performing Gaussian pyramid decomposition, which can effectively save transmission bandwidth.

[0042] First, S1 is executed to perform Gaussian filtering on the current frame image and perform the first downsampling to obtain a first downsampling image, and the first downsampling image is decomposed by Gaussian pyramid to obtain a current frame image group including multiple scale images. Then S2 is executed to upsample the first downsampling image to obtain an upsampling image, and the current frame image is subtracted from the upsampling image to obtain high-frequency information of the current frame image.

[0043] Specifically, first obtain the current frame image, denoted as Q, with a size of HxW. Then downsample the current frame image Q to obtain Q1, and the downsampling method includes but is not limited to average pooling, selecting a single point and other downsampling operations, and its size is H / 2xW / 2. Then upsample Q1 to obtain Q', and the upsampling method includes but is not limited to upsampling operations such as interpolation and padding. The difference between Q and Q' can be calculated to obtain the high-frequency information H.

[0044] In this embodiment, the downsampled image Q1 is decomposed into k layers by Gaussian pyramid decomposition. The decomposition process of the i-th layer (0≤i≤k) is to first use a radius r0 and a variance δ 2 The image is low-pass filtered by a Gaussian kernel, and then downsampled to obtain the i+1th layer with a resolution of 1 / 4 of the previous layer, which can be expressed as:

[0045]

[0046] Then, S3 is executed to perform motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and correct the motion estimation weights and texture estimation weights at each scale through guided filtering.

[0047] In this embodiment, the pre-cached reference frame image group is obtained by performing Gaussian pyramid decomposition after downsampling the previous frame image, so the maximum size of the reference frame image group is the image size after the reference frame image is downsampled for the first time. i The reference frame image of the same scale is denoted as P i , the “performing motion estimation and texture estimation at different scales” mentioned in this application refers to performing motion estimation and texture estimation on the current frame image and the reference frame image at the same scale, and repeating the process multiple times to obtain the motion estimation and texture estimation results of the current frame image and the reference frame image at different scales. For example, at the i-th layer, the current frame image Q i and the reference frame image P i Perform motion estimation and texture estimation to obtain motion estimation weights M at different scales i and texture estimation weight D i .

[0048] Optionally, the motion estimation is performed using an inter-frame difference method, an optical flow method, or a density-based clustering algorithm.

[0049] Optionally, an edge extraction operator such as a sobel operator, a laplace operator, etc. is used to perform the texture estimation.

[0050] Get the motion estimation weight M at different scales i and texture estimation weight D iAfter that, the motion estimation weights and texture estimation weights at each scale are corrected by guided filtering. Specifically, taking the correction of motion estimation weight as an example, the corrected motion estimation weight of the i-th layer is denoted as M′ i , the modified estimated weight M′ of the minimum scale kth layer k =M k The calculation process of the modified weight is to upsample the modified estimated weight of the i+1 layer as the guide map, and then perform guided filtering on the uncorrected estimated weight of the i-th layer to obtain M′ i The motion weight correction is finally passed to the first layer, and the corrected motion estimation weight of size H / 2xW / 2 is obtained, which is recorded as M1′. The texture weight correction is similar, and finally passed to the 0th layer, and the corrected texture estimation weight of size HxW is obtained, which is recorded as D0′.

[0051] Then, S4 is executed to perform spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a denoised downsampled image. Specifically, at the first level scale, spatiotemporal filtering is performed on the current frame image of the same scale based on the modified motion estimation weight M1′, which is divided into spatial filtering and temporal filtering fusion.

[0052] In the first downsampled image Q i Spatial filtering is performed at the corresponding scale using non-local mean filtering. In the NxN window centered at the current point, nxn block matching is performed to calculate the similarity between blocks. The similarity is then mapped into fusion weights for weighted fusion, as follows:

[0053] Assume that the coordinates of the center point of the central block are (x0, y0), and the coordinates of the center point of the block to be calculated for similarity are (x1, y1), then the similarity is expressed by the formula:

[0054]

[0055] The spatial domain filtering result is recorded as S, and the weighted fusion process is expressed as

[0056]

[0057] Among them, k1 and k2 are hyperparameters, and the temporal filtering is the weighted fusion of the spatial filtering result S of the current frame image and the reference frame image P1 at the corresponding scale according to the weight α. The temporal fusion result is recorded as T, and the formula is as follows:

[0058] T=αS+(1-α)P1

[0059]

[0060] Among them, k3 and k4 are hyperparameters.

[0061] Then, S5 is executed to upsample the denoised downsampled image to obtain a denoised upsampled image, which has the same size as the current frame image, and weightedly fuse the denoised upsampled image with the high-frequency information according to the texture estimation weight at the corresponding scale to obtain a final image. In this embodiment, the time domain fusion result T is upsampled to obtain T', and at the 0th scale, based on the texture weight W D The high-frequency information H obtained by S2 is integrated, and the final image after denoising in the spatiotemporal domain is recorded as O, which can be expressed as follows:

[0062]

[0063] O=T′+W D H

[0064] Among them, k5 and k6 are hyper parameters.

[0065] Finally, S6 is executed to downsample the final image and perform Gaussian pyramid decomposition to obtain a reference frame image group for denoising the next frame image, so as to save bandwidth consumption.

[0066] Based on the same inventive concept, an embodiment of the present invention further provides an image denoising system, which is used to implement the above-mentioned image denoising method, including:

[0067] A first decomposition module is configured to perform Gaussian filtering on the current frame image and perform a first downsampling to obtain a first downsampled image, and perform Gaussian pyramid decomposition on the first downsampled image to obtain a current frame image group including a plurality of scale images;

[0068] an upsampling module, configured to upsample the first downsampled image to obtain an upsampled image, and to perform a difference between the current frame image and the upsampled image to obtain high frequency information of the current frame image;

[0069] an estimation module, configured to perform motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and to correct the motion estimation weights and texture estimation weights at each scale by guided filtering;

[0070] a filtering module configured to perform spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a downsampled image after noise reduction;

[0071] a fusion module configured to upsample the denoised downsampled image to obtain a denoised upsampled image, and weightedly fuse the denoised upsampled image with the high-frequency information according to a texture estimation weight at a corresponding scale to obtain a final image;

[0072] The second decomposition module downsamples the final image and performs Gaussian pyramid decomposition to obtain a reference frame image group for denoising the next frame image.

[0073] Optionally, the maximum size of the reference frame image group is the image size after the reference frame image is down-sampled for the first time.

[0074] Optionally, the motion estimation is performed using an inter-frame difference method, an optical flow method, or a density-based clustering algorithm.

[0075] Optionally, an edge extraction operator is used to perform the texture estimation.

[0076] Based on the same inventive concept, an embodiment of the present invention further provides a readable storage medium on which a computer program is stored. When the computer program is executed, the image noise reduction method as described above can be implemented.

[0077] The readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device, such as but not limited to an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device or any suitable combination thereof. The more specific example (non-exhaustive list) of the readable storage medium includes: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer program described herein can be downloaded from the readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network and / or a wireless network. The network can include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer program from the network and forwards the computer program for storage in the readable storage medium in each computing / processing device. The computer program for performing the operation of the present invention can be a run instruction, an instruction set architecture (ISA) instruction, a machine instruction, a machine-related instruction, a microcode, a firmware instruction, a state setting data, or a source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer program can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet). In some embodiments, by utilizing the state information of a computer program to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute computer-readable program instructions to implement various aspects of the present invention.

[0078] Here, various aspects of the present invention are described with reference to the flowchart and / or block diagram of the method, system and computer program product according to the embodiment of the present invention. It should be understood that each square frame of the flowchart and / or block diagram and the combination of the square frames in the flowchart and / or block diagram can be realized by a computer program. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so as to produce a machine, so that these programs, when executed by the processor of a computer or other programmable data processing device, produce a device for realizing the function / action specified in one or more square frames in the flowchart and / or block diagram. These computer programs can also be stored in a readable storage medium, and these computer programs make the computer, programmable data processing device and / or other equipment work in a specific way, so that the readable storage medium storing the computer program includes a manufactured product, which includes instructions for realizing various aspects of the function / action specified in one or more square frames in the flowchart and / or block diagram.

[0079] The computer program may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are executed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the computer program executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0080] In summary, the embodiments of the present invention provide an image denoising method, system and readable storage medium. By performing motion estimation and texture estimation on multi-scale images and correcting the weights at each scale through guided filtering, more accurate motion estimation weights and texture estimation weights can be obtained, thereby performing spatiotemporal filtering and fusing high-frequency information, increasing the image signal-to-noise ratio while retaining more texture details, thereby obtaining a denoising result with a better signal-to-noise ratio and more texture. In addition, the reference frame image group used for denoising the next frame of the image is obtained by downsampling the current image and performing Gaussian pyramid decomposition, which can effectively save transmission bandwidth.

[0081] The above description is only a description of the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any changes and modifications made by a person skilled in the art in the field of the present invention based on the above disclosure are within the scope of protection of the present invention. Obviously, a person skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations fall within the scope of the present invention and its equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An image denoising method, characterized in that: include: Performing Gaussian filtering on the current frame image and performing a first downsampling to obtain a first downsampling image, and performing Gaussian pyramid decomposition on the first downsampling image to obtain a current frame image group including a plurality of scale images; Upsampling the first downsampled image to obtain an upsampled image, and subtracting the current frame image from the upsampled image to obtain high frequency information of the current frame image; Performing motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and correcting the motion estimation weights and texture estimation weights at each scale by guided filtering; Performing spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a downsampled image after noise reduction; Upsampling the denoised downsampled image to obtain a denoised upsampled image, and weightedly fusing the denoised upsampled image with the high-frequency information according to a texture estimation weight at a corresponding scale to obtain a final image; The final image is downsampled and decomposed using a Gaussian pyramid to obtain a reference frame image group for denoising the next frame image.

2. The image denoising method according to claim 1, characterized in that: The maximum size of the reference frame image group is the image size after the reference frame image is down-sampled for the first time.

3. The image denoising method according to claim 1, characterized in that: The motion estimation is performed using an inter-frame difference method, an optical flow method, or a density-based clustering algorithm.

4. The image denoising method according to claim 1, characterized in that: The texture estimation is performed using an edge extraction operator.

5. The image denoising method according to claim 1, characterized in that: When the motion estimation weights and texture estimation weights at each scale are corrected by guided filtering, the weight information is calculated and corrected layer by layer from the smallest scale to the largest scale.

6. An image noise reduction system, characterized in that: include: A first decomposition module is configured to perform Gaussian filtering on the current frame image and perform a first downsampling to obtain a first downsampled image, and perform Gaussian pyramid decomposition on the first downsampled image to obtain a current frame image group including a plurality of scale images; an upsampling module, configured to upsample the first downsampled image to obtain an upsampled image, and to perform a difference between the current frame image and the upsampled image to obtain high frequency information of the current frame image; an estimation module, configured to perform motion estimation and texture estimation at different scales on the current frame image group and the pre-cached reference frame image group, and to correct the motion estimation weights and texture estimation weights at each scale by guided filtering; a filtering module configured to perform spatiotemporal filtering on the first downsampled image and the reference frame image of the corresponding scale in the reference frame image group based on the motion estimation weight at the corresponding scale to obtain a downsampled image after noise reduction; a fusion module configured to upsample the denoised downsampled image to obtain a denoised upsampled image, and weightedly fuse the denoised upsampled image with the high-frequency information according to a texture estimation weight at a corresponding scale to obtain a final image; The second decomposition module downsamples the final image and performs Gaussian pyramid decomposition to obtain a reference frame image group for denoising the next frame image.

7. The image noise reduction system according to claim 6, characterized in that: The maximum size of the reference frame image group is the image size after the reference frame image is down-sampled for the first time.

8. The image noise reduction system according to claim 6, characterized in that: The motion estimation is performed using an inter-frame difference method, an optical flow method, or a density-based clustering algorithm.

9. The image noise reduction system according to claim 6, characterized in that: The texture estimation is performed using an edge extraction operator.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the image noise reduction method according to any one of claims 1 to 5 can be implemented.

Images