Image noise reduction method and noise reduction system

By combining noise level and global feature index in image denoising, calculating the noise compression multiple and expected target weights, combined with similar weights and nonlinear shrinkage technology, the problem of difficulty in balancing noise distribution changes and detail retention of traditional methods is solved, and the more Fine noise suppression and detail retention improve image quality.

CN118735814BActive Publication Date: 2024-11-12GUOCHUANG RUISHI ARTIFICIAL INTELLIGENCE TECHNOLOGY (SICHUAN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411214905.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-11-12
Estimated Expiration
2044-09-02

AI Technical Summary

Technical Problem

The traditional image noise reduction method based on wavelet transform is poor when the noise distribution changes, and it is difficult to balance noise suppression and detail retention, especially fixed threshold selection and simple wavelet thresholding ignore the local characteristics of the image.

Method used

By acquiring the high-frequency and low-frequency components of the image after multi-scale decomposition, the noise level and global feature index are calculated, and the noise reduction processing of the high-frequency components of the image is combined with the noise compression multiple and the expected target weight, and the similar weights and nonlinear shrinkage are used to reduce the noise. Technology, combining multi-scale images for finer noise suppression and detail retention.

Benefits of technology

Achieve adaptive noise reduction effect under different noise levels and image characteristics, improve the flexibility and stability of the noise reduction process, ensure the balance between noise suppression and detail retention, and improve the overall quality of the image after noise reduction .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118735814B_ABST
    Figure CN118735814B_ABST
Patent Text Reader

Abstract

The present invention provides an image noise reduction method and a noise reduction system, which relate to the technical field of image processing. Symlets The wavelet function performs multi-scale decomposition and more finely separates the low-frequency and high-frequency components of the image, which is conducive to targeted processing of noise of different scales. At the same time, threshold processing is introduced to pre-process the coefficients of the high-frequency components, which can remove the obvious noise of the image before denoising, and prepare for the subsequent more refined denoising. The global feature index of the image is calculated using local binary values, which can accurately reflect the complexity and uniformity of the image texture, and combined with the noise level, it provides a more accurate and reasonable basis for the subsequent calculation of the noise compression multiple. It fully considers the local characteristics of the image, realizes adaptive control of noise, improves the flexibility and pertinence of the denoising strategy, and the calculation of the expected target weight further optimizes the denoising process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image noise reduction method and a noise reduction system. Background Art

[0002] In the field of digital image processing, image noise reduction is a crucial technology used to improve the quality of images. With the popularity of digital images in daily life and industrial applications, the demand for image noise reduction is growing. High-quality images are essential for accurate analysis and decision-making.

[0003] Traditional image denoising methods mainly include spatial filtering, frequency filtering and model-based methods. However, image details are often blurred while removing noise. Model-based methods, such as machine learning models, can better adapt to different types of noise, but have high computational complexity and require a large amount of training data. In recent years, multiscale analysis methods, especially image denoising techniques based on wavelet transform, have attracted widespread attention due to their ability to effectively remove noise while maintaining image details. However, traditional image denoising methods based on wavelet transform have limitations, such as fixed threshold selection may not adapt to changes in noise distribution in the image, resulting in poor denoising or over-smoothing. In addition, simple wavelet thresholding often ignores the local features of the image, making it difficult to strike a balance between denoising and detail retention.

[0004] Therefore, it is necessary to provide an image noise reduction method and a noise reduction system to solve the above technical problems. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides an image denoising method and a denoising system, which can achieve a more refined noise suppression while retaining the beneficial effects of image details and structures to the maximum extent.

[0006] The present invention provides an image noise reduction method and a noise reduction system, comprising the following steps:

[0007] S1: Obtain the high-frequency components and low-frequency components of the image to be denoised after multi-scale decomposition, and calculate the noise level of the high-frequency components of the image;

[0008] S2: Estimating image features of the high-frequency components of the image at all scales to obtain a global feature index corresponding to the high-frequency components of the image, wherein the global feature index is calculated based on the LBP value of each pixel point of the high-frequency components of the image;

[0009] S3: traverse the high-frequency components of the image, and calculate the noise compression multiple of the high-frequency components of the image in combination with the corresponding noise level and the global feature index. At the same time, according to the noise compression multiple, calculate the expected target weight of the high-frequency components of the image;

[0010] S4: Based on the LBP value of each pixel in the high-frequency component of the image, the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood is calculated;

[0011] S5: Based on the similarity weight and the expected target weight, all pixels in the high-frequency component of the image are denoised to obtain the high-frequency component of the denoised image, the low-frequency component of the image at the same scale and the high-frequency component of the denoised image are merged to obtain a multi-scale denoised image, and the multi-scale denoised images are fused to obtain the denoised image.

[0012] Preferably, in step S1, the method of multi-scale decomposition of the image to be denoised is wavelet transform.

[0013] Preferably, step S1 further includes:

[0014] After the image to be denoised is decomposed at multiple scales, the coefficients of the high-frequency components of each image are thresholded.

[0015] Preferably, step S2 includes the following specific steps:

[0016] S201: Performing local binary calculation on each pixel point in the high-frequency component of the image at all scales to obtain the LBP value of each pixel point in the high-frequency component of the image;

[0017] S202: Establishing an LBP histogram corresponding to the high-frequency components of the image based on the LBP value of each pixel of the high-frequency components of the image;

[0018] S203: Calculate the global feature index corresponding to the high-frequency component of the image based on the statistical characteristics of the LBP histogram corresponding to the high-frequency component of the image.

[0019] Preferably, in step S3, the relationship between the noise compression factor, the noise level and the global characteristic index is:

[0020]

[0021] in, is the noise compression factor, is the noise level, The global is the characteristic index, , , , is a constant.

[0022] Preferably, in step S3, the relationship between the expected target weight and the noise compression factor is:

[0023]

[0024] in, is the expected target weight, is the noise compression factor.

[0025] Preferably, step S4 includes the following specific steps:

[0026] S401: Calculate the LBP local difference measure between each pixel and each pixel in its neighborhood based on the LBP value of each pixel of the high-frequency component of the image;

[0027] S402: Convert the LBP local difference measure into a similarity weight. The conversion formula is:

[0028]

[0029] in is the similarity weight, is the LBP local difference metric, is a positive real number parameter.

[0030] Preferably, in step S5, the similarity weights are nonlinearly shrunk to obtain adjusted similarity weights, the total weights of the pixels are calculated based on the adjusted similarity weights, and the total weights of the pixels are normalized to the desired target weights, wherein the nonlinear shrinkage is performed using Sigmoid Function, the total weight of a pixel is the cumulative sum of the adjusted similarity weights between each pixel and each pixel in its neighborhood.

[0031] Preferably, in step S5, the multi-scale denoised images are fused using a pyramid fusion method.

[0032] The present invention also provides an image noise reduction system, which is applied to an image noise reduction method. The noise reduction system comprises:

[0033] The noise level calculation module obtains the high-frequency components and low-frequency components of the image to be denoised after multi-scale decomposition, and calculates the noise level of the high-frequency components of the image;

[0034] A feature evaluation module estimates the image features of the high-frequency components of the image at all scales to obtain a global feature index corresponding to the high-frequency components of the image, wherein the global feature index is calculated based on the LBP value of each pixel point of the high-frequency components of the image;

[0035] The noise compression multiple and expected target weight calculation module traverses the high-frequency components of the image, and calculates the noise compression multiple of the high-frequency components of the image in combination with the corresponding noise level and the global feature index. At the same time, according to the noise compression multiple, the expected target weight of the high-frequency components of the image is calculated;

[0036] The similarity weight calculation module calculates the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood based on the LBP value of each pixel in the high-frequency component of the image;

[0037] The denoising processing module performs denoising on all pixels in the high-frequency components of the image to obtain the high-frequency components of the denoised image, combines the low-frequency components of the image at the same scale and the high-frequency components of the denoised image to obtain a multi-scale denoised image, and fuses the multi-scale denoised images to obtain a denoised image.

[0038] Compared with the related art, the image noise reduction method and noise reduction system provided by the present invention have the following beneficial effects:

[0039] The present invention adopts Symlets The wavelet function performs multi-scale decomposition and more finely separates the low-frequency and high-frequency components of the image, which is conducive to targeted processing of noise of different scales. At the same time, threshold processing is introduced to pre-process the coefficients of high-frequency components, which can remove the obvious noise of the image before denoising, and prepare for the subsequent more refined denoising. The global feature index of the image is calculated using local binary values, which can accurately reflect the complexity and uniformity of the image texture, and combined with the noise level, it provides a more accurate and reasonable basis for the subsequent calculation of the noise compression multiple. It fully considers the local characteristics of the image, realizes adaptive control of noise, and improves the flexibility and pertinence of the denoising strategy. The calculation of the expected target weight further optimizes the denoising process, ensuring the denoising effect and detail retention under different noise levels and image characteristics. Sigmoid The function performs nonlinear shrinkage on similar weights, which can adjust the weight distribution more smoothly, avoid the extreme weight values, and improve the stability and reliability of the denoising process. Finally, the pyramid fusion technology is used to merge multi-scale denoised images, taking into account the detail richness and noise residual level of images at each scale, achieving a balance between detail retention and noise suppression, and improving the overall quality of the denoised image. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of an image noise reduction method of the present invention;

[0041] Figure 2 This is a module structure of an image noise reduction system of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention, rather than all structures, are shown in the accompanying drawings. In addition, the embodiments of the present invention and the features in the embodiments may be combined with each other without conflict.

[0043] It should also be noted that, for ease of description, only the parts related to the present invention, but not all of the contents, are shown in the accompanying drawings. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the various operations (or steps) as being processed sequentially, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the various operations can be rearranged. The processing can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The processing can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0044] Embodiment 1

[0045] The present invention provides an image noise reduction method. In a specific implementation process, Figure 1 As shown, it shows a flow chart of an image noise reduction method provided by the present invention, including:

[0046] Step S1, obtaining high-frequency components and low-frequency components of the image to be denoised after multi-scale decomposition, and calculating the noise level of the high-frequency components of the image.

[0047] Specifically, in step S1, the method of multi-scale decomposition of the image to be denoised is wavelet transform.

[0048] Specifically, step S1 also includes:

[0049] After the image to be denoised is decomposed at multiple scales, the coefficients of the high-frequency components of each image are thresholded.

[0050] Based on the description of the above steps, in the specific implementation process, first, using wavelet transform technology, select Symlets The wavelet function performs multi-scale decomposition on the image to be denoised to obtain low-frequency and high-frequency components of the image at different scales, and threshold processing is performed on the high-frequency components of the image at each scale. For example, a threshold T is set, and the high-frequency coefficients below the threshold T are set to zero to preliminarily remove the obvious noise components in the image. Then, on this basis, the noise level of the high-frequency components of the image at each scale is estimated respectively. The estimation of the noise level is achieved by calculating a certain statistical value of the coefficients of the high-frequency components of the image at each scale, which includes but is not limited to the variance and standard deviation, so as to provide a quantitative indicator for the subsequent denoising processing and help determine the strength of the denoising strategy.

[0051] Step S2, performing image feature estimation on the high-frequency components of the image at all scales to obtain a global feature index corresponding to the high-frequency components of the image, wherein the global feature index is calculated based on the LBP value of each pixel point of the high-frequency components of the image.

[0052] Specifically, step S2 includes the following specific steps:

[0053] S201: Performing local binary calculation on each pixel point in the high-frequency component of the image at all scales to obtain the LBP value of each pixel point in the high-frequency component of the image;

[0054] S202: Establishing an LBP histogram corresponding to the high-frequency components of the image based on the LBP value of each pixel of the high-frequency components of the image;

[0055] S203: Calculate the global feature index corresponding to the high-frequency component of the image based on the statistical characteristics of the LBP histogram corresponding to the high-frequency component of the image.

[0056] Based on the description of the above steps, in the specific implementation process, first, for each pixel point of the high-frequency component of each image, a local binary calculation is performed. For example, for each pixel point, a central pixel and eight neighboring pixels around it are defined; then, the grayscale value of the central pixel is compared with the grayscale value of each neighboring pixel. If the grayscale value of the neighboring pixel is greater than or equal to the grayscale value of the central pixel, it is recorded as a logical value of 1; otherwise, it is recorded as a logical value of 0. In this way, each central pixel will obtain a binary sequence of length eight, which is the LBP value of the pixel point; wherein, for pixel points at corners and edges, their neighboring pixels cannot be eight. After calculating their LBP values, zeros are added to the prefix of their binary representation until the number of bits of their LBP values ​​is eight, thereby achieving the unification of the number of bits of the LBP values.

[0057] After completing the local binary calculation of all pixels of the high-frequency components of each image, the LBP histogram corresponding to the high-frequency components of each image is constructed based on the LBP values ​​of all pixels of the high-frequency components of each image; the LBP histogram counts the frequency of occurrence of different LBP values; finally, based on the LBP histogram corresponding to the high-frequency components of each image, the entropy of the LBP histogram and the uniformity of the LBP histogram are obtained, and the global feature index of the high-frequency components of each image is calculated by combining the entropy of the LBP histogram and the uniformity of the LBP histogram, normalizing and weighting them.

[0058] Step S3, traverse the high-frequency components of the image, and calculate the noise compression multiples of the high-frequency components of the image in combination with the corresponding noise level and the global feature index, and at the same time calculate the expected target weight of the high-frequency components of the image according to the noise compression multiples.

[0059] Specifically, in step S3, the relationship between the noise compression factor, the noise level and the global characteristic index is:

[0060]

[0061] in, is the noise compression factor, is the noise level, is the global feature index, , , , is a constant.

[0062] Specifically, in step S3, the relationship between the expected target weight and the noise compression factor is:

[0063]

[0064] in, is the expected target weight, is the noise compression factor.

[0065] Based on the description of the above steps, in the specific implementation process, the noise compression multiples of the high-frequency components of the images at all scales are calculated based on the above-mentioned relationship between the noise level of the high-frequency components of each image and the global feature index and the noise compression multiple, where: , , , is a constant, which is a parameter determined based on experiments and is used to adjust the shape and response characteristics of the function. The expected target weight is calculated based on the noise compression factor calculated above and the relationship between the expected target weight and the noise compression factor. Here, the noise compression factor itself is calculated based on the noise level and the global feature index, which reflects the image noise characteristics and texture complexity. Therefore, the calculation of the expected target weight not only considers the distribution of noise, but also the texture characteristics of the image, ensuring the adaptability and intelligence of the denoising strategy.

[0066] Step S4, based on the LBP value of each pixel in the high-frequency component of the image, calculate the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood.

[0067] Specifically, step S4 includes the following specific steps:

[0068] S401: Calculate the LBP local difference measure between each pixel and each pixel in its neighborhood based on the LBP value of each pixel of the high-frequency component of the image;

[0069] S402: Convert the LBP local difference measure into a similarity weight. The conversion formula is:

[0070]

[0071] in is the similarity weight, is the LBP local difference metric, is a positive real number parameter.

[0072] Based on the description of the above steps, in the specific implementation process, first, based on the LBP value of each pixel of the high-frequency component of the image, the LBP local difference metric between each pixel and each pixel in its neighborhood is calculated. Exemplarily, the LBP local difference metric is obtained by calculating the Hamming distance between the LBP values ​​of two pixels. The Hamming distance directly reflects the bit difference of the binary sequence. This bit difference accurately reflects the difference of the LBP value in the local texture features. Based on the obtained Hamming distance, according to the Gaussian function, the Hamming distance is converted into a similarity weight, as shown in the conversion formula shown in step S402, wherein, is the LBP local difference metric, namely the above-mentioned Hamming distance, is a positive real number parameter obtained from actual experiments, which is used to control the rate at which the weight changes with the difference.

[0073] Step S5, based on the similarity weight and the expected target weight, all pixels in the high-frequency component of the image are denoised to obtain the high-frequency component of the image after denoising, the low-frequency component of the image at the same scale and the high-frequency component of the image after denoising are merged to obtain a multi-scale denoised image, and the multi-scale denoised images are fused to obtain a denoised image.

[0074] Specifically, in step S5, the similarity weights are nonlinearly shrunk to obtain adjusted similarity weights, the total weights of the pixels are calculated based on the adjusted similarity weights, and the total weights of the pixels are normalized to the desired target weights, wherein the nonlinear shrinkage is performed using Sigmoid Function, the total weight of a pixel is the cumulative sum of the adjusted similarity weights between each pixel and each pixel in its neighborhood.

[0075] Specifically, in step S5, the multi-scale denoised images are fused using a pyramid fusion method.

[0076] Based on the above steps, in the specific implementation process, first, the similarity weights between a pixel and all pixels in its neighborhood are nonlinearly shrunk to obtain the adjusted similarity weights. The nonlinear shrinkage is performed using Sigmoid Function, after nonlinear shrinkage, the similarity weights of all pixels in each pixel area are adjusted to the interval (0, 1). Based on the adjusted similarity weights, the cumulative sum of the adjusted similarity weights is calculated to obtain the total weight. For example, the total weight is recorded as , according to the above, the expected target weight is recorded as , according to the total weight and the expected target weight Calculate the normalization factor. For example, the normalization factor is set to , The value of is given by the formula Calculated, based on the normalization factor obtained , calculate the weighted average of the pixels, that is, the pixel value after noise reduction. Exemplarily, the pixel value after noise reduction is calculated according to the following formula:

[0077]

[0078] in, is the pixel value after noise reduction, is the first pixel in the neighborhood i The similarity weight after pixel adjustment is is the first pixel in the neighborhood i The pixel value of a pixel, is the normalization factor mentioned above. Through this formula, all pixels in the high-frequency component of the image are denoised to obtain the high-frequency component of the image after denoising. Symlets The wavelet function inversely transforms the low-frequency components of the image at the same scale and the high-frequency components of the denoised image to obtain the denoised image at that scale. The denoised images at multiple scales are then fused using the pyramid fusion technology to obtain the final denoised image. After the above steps, the total weight of all pixels in the neighborhood of the pixel is ensured. and expected target weight Therefore, the contribution of each pixel in the neighborhood of the pixel can be effectively distributed according to the noise reduction strategy during weighted averaging, thus achieving high-quality image noise reduction effect.

[0079] The working principle of an image noise reduction method provided by the present invention is as follows:

[0080] First, use Symlets The wavelet function performs multi-scale decomposition on the image to be denoised to obtain low-frequency and high-frequency components of the image at multiple scales. The high-frequency components of the image contain the details and texture information of the image, as well as the noise information. Therefore, the high-frequency components of the image are directly denoised, and the threshold processing is used to set the coefficients of the high-frequency components of the image that are lower than the set threshold to zero, thereby removing the obvious noise in the high-frequency components of the image and preparing for subsequent more detailed denoising processing.

[0081] Then, the global feature index of the image is obtained by estimating the global features of the high-frequency components of the image. The global feature index is calculated based on the LBP value of each pixel of the high-frequency components of the image. The LBP value of the pixel is a quantification of the local texture features of the pixel, reflecting the texture features of the high-frequency components of the image.

[0082] Secondly, based on the noise level of the high-frequency components of the image and the global special index, the noise compression multiples corresponding to the high-frequency components of the image at all scales are calculated, and then the expected target weight is calculated according to the noise compression multiples.

[0083] Thirdly, based on the LBP value of each pixel in the high-frequency component of the image, the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood is calculated. The similarity weight fully considers the local texture characteristics of the image.

[0084] Finally, based on the similarity weight, the pixel value after denoising is obtained by weighted calculation. The same weighted calculation is performed on all the pixels of the high-frequency components of the image to obtain the high-frequency components of the image after denoising at this scale. Then, the low-frequency components of the image at this scale are compared with the high-frequency components of the image after denoising. Symlets The inverse transform of the wavelet function is used to obtain the denoised image at that scale, and the denoised images at multiple scales are fused through the pyramid fusion technology to obtain the final denoised image.

[0085] Embodiment 2

[0086] The present invention also provides an image denoising system applied to an image denoising method. In a specific implementation process, Figure 2 As shown, it shows a schematic diagram of the module structure of an image noise reduction system provided by the present invention, including:

[0087] The noise level calculation module 100 obtains the image high-frequency components and image low-frequency components of the image to be denoised after multi-scale decomposition, and calculates the noise level of the image high-frequency components.

[0088] The feature evaluation module 200 estimates the image features of the high-frequency components of the image at all scales to obtain a global feature index corresponding to the high-frequency components of the image, wherein the global feature index is calculated based on the LBP value of each pixel point of the high-frequency components of the image.

[0089] The noise compression factor and expected target weight calculation module 300 traverses the high-frequency components of the image, and calculates the noise compression factor of the high-frequency components of the image in combination with the corresponding noise level and the global feature index, and calculates the expected target weight of the high-frequency components of the image based on the noise compression factor.

[0090] The similarity weight calculation module 400 calculates the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood based on the LBP value of each pixel in the high-frequency component of the image.

[0091] The noise reduction processing module 500 performs noise reduction processing on all pixel points in the high-frequency component of the image to obtain the high-frequency component of the image after noise reduction, combines the low-frequency component of the image at the same scale and the high-frequency component of the image after noise reduction to obtain a multi-scale noise reduction image, and fuses the multi-scale noise reduction images to obtain a noise reduction image.

[0092] The working principle of an image noise reduction system provided by the present invention is:

[0093] The noise level calculation module 100 uses Symlets The wavelet function treats the denoised image at multiple scales, decomposes the image into low-frequency components and high-frequency components, and calculates the noise level of the high-frequency components; the feature evaluation module 200 calculates the LBP value of each pixel of the high-frequency component of the image, establishes an LBP histogram based on the LBP value, and calculates the global feature index according to the statistical characteristics of the LBP histogram to reflect the texture characteristics of the high-frequency component of the image; the noise compression factor and expected target weight calculation module 300 calculates the noise compression factor in combination with the noise level and the global feature index, and calculates the expected target weight according to the noise compression factor; the similarity weight calculation module 400 calculates the LBP value difference between each pixel of the high-frequency component of the image and each pixel in its neighborhood, uses the Hamming distance as the LBP local difference metric, and converts the difference metric into a similarity weight through a Gaussian function; the denoising processing module 500 uses Sigmoid The function performs nonlinear shrinkage on the similarity weights to adjust the weight distribution, calculates the cumulative sum of the adjusted similarity weights between each pixel and each pixel in its neighborhood, and obtains the total weight. The total weight is normalized to the expected target weight to ensure that the sum of the similarity weights of all pixels in the domain of each pixel matches the expected target weight. The high-frequency components of the image are denoised using the adjusted similarity weights and the expected target weights, and are merged with the low-frequency components to obtain a multi-scale denoised image. The multi-scale denoised images are fused using pyramid fusion technology, and denoised images of different scales are gradually fused to reconstruct the final denoised image.

[0094] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, the storage medium including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically-erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0096] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or, also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. An image denoising method, characterized in that: The following steps are involved: S1: Obtain the high-frequency components and low-frequency components of the image to be denoised after multi-scale decomposition, and calculate the noise level of the high-frequency components of the image; S2: Estimating image features of the high-frequency components of the image at all scales to obtain a global feature index corresponding to the high-frequency components of the image, wherein the global feature index is calculated based on the LBP value of each pixel point of the high-frequency components of the image; S3: traverse the high-frequency components of the image, and calculate the noise compression multiples of the high-frequency components of the image in combination with the corresponding noise level and the global feature index, and calculate the expected target weight of the high-frequency components of the image according to the noise compression multiples; S4: Based on the LBP value of each pixel in the high-frequency component of the image, the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood is calculated; S5: Based on the similarity weight and the expected target weight, all pixels in the high-frequency component of the image are subjected to denoising to obtain the high-frequency component of the denoised image, the low-frequency component of the image at the same scale and the high-frequency component of the denoised image are combined to obtain a multi-scale denoised image, and the multi-scale denoised images are fused to obtain a denoised image; Step S2 includes the following specific steps: S201: Performing local binary calculation on each pixel point in the high-frequency component of the image at all scales to obtain the LBP value of each pixel point in the high-frequency component of the image; S202: Establishing an LBP histogram corresponding to the high-frequency components of the image based on the LBP value of each pixel of the high-frequency components of the image; S203: Calculating a global feature index corresponding to the high-frequency component of the image based on the statistical characteristics of the LBP histogram corresponding to the high-frequency component of the image; In step S3, the relationship between the noise compression factor, the noise level and the global characteristic index is: in, is the noise compression factor, is the noise level, is the global feature index, , , , is a constant; In step S3, the relationship between the expected target weight and the noise compression factor is: in, is the expected target weight, is the noise compression factor; Step S4 includes the following specific steps: S401: Calculate the LBP local difference measure between each pixel and each pixel in its neighborhood based on the LBP value of each pixel of the high-frequency component of the image; S402: Convert the LBP local difference measure into a similarity weight. The conversion formula is: in is the similarity weight, is the LBP local difference metric, is a positive real number parameter; In step S5, the similarity weights are nonlinearly shrunk to obtain adjusted similarity weights, the total weights of the pixels are calculated based on the adjusted similarity weights, and the total weights of the pixels are normalized to the desired target weights, wherein the nonlinear shrinkage is performed using Sigmoid Function, the total weight of a pixel is the cumulative sum of the adjusted similarity weights between each pixel and each pixel in its neighborhood.

2. The image denoising method according to claim 1, characterized in that: In step S1, the method of multi-scale decomposition of the image to be denoised is wavelet transform.

3. The image denoising method according to claim 2, characterized in that: Step S1 also includes: After the image to be denoised is decomposed at multiple scales, the coefficients of the high-frequency components of each image are thresholded.

4. The image denoising method according to claim 1, characterized in that: In step S5, the multi-scale denoised images are fused using a pyramid fusion method.

5. An image noise reduction system, characterized in that: Applied to an image denoising method according to any one of claims 1 to 4, the denoising system comprises: The noise level calculation module obtains the high-frequency components and low-frequency components of the image to be denoised after multi-scale decomposition, and calculates the noise level of the high-frequency components of the image; A feature evaluation module estimates the image features of the high-frequency components of the image at all scales to obtain a global feature index corresponding to the high-frequency components of the image, wherein the global feature index is calculated based on the LBP value of each pixel point of the high-frequency components of the image; The noise compression multiple and expected target weight calculation module traverses the high-frequency components of the image, and calculates the noise compression multiple of the high-frequency components of the image in combination with the corresponding noise level and the global feature index. At the same time, according to the noise compression multiple, the expected target weight of the high-frequency components of the image is calculated; The similarity weight calculation module calculates the similarity weight between each pixel in the high-frequency component of the image and each pixel in its neighborhood based on the LBP value of each pixel in the high-frequency component of the image; The denoising processing module performs denoising on all pixels in the high-frequency components of the image to obtain the high-frequency components of the denoised image, combines the low-frequency components of the image at the same scale and the high-frequency components of the denoised image to obtain a multi-scale denoised image, and fuses the multi-scale denoised images to obtain a denoised image.