Guided filter denoising method and device, electronic equipment and readable storage medium

Abstract

The application provides a guided filter denoising method and device, electronic equipment and readable storage medium, and belongs to the technical field of image denoising. The method comprises the following steps: processing an original image through an average pooling layer of a neural network to obtain low-frequency information; obtaining high-frequency information according to the original image and the low-frequency information; training the high-frequency information through a first convolutional network to obtain a denoising coefficient; performing denoising processing on the high-frequency information according to the denoising coefficient to obtain a first image; performing artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image; and determining a target image according to the first image and the second image, which can adapt to image denoising in various scenes, improve the filtering effect and improve the high-frequency detail recovery effect.

Description

Technical Field

[0001] This application relates to the field of image denoising technology, and in particular to a guided filtering denoising method, apparatus, electronic device, and readable storage medium. Background Technology

[0002] In image processing tasks, image filtering techniques play a crucial role in suppressing noise while preserving necessary edge information. Image filtering techniques are commonly used in tasks such as image restoration, boundary detection, texture segmentation, and image detail enhancement. Traditional filtering algorithms, such as Gaussian filtering and mean filtering, while capable of quickly processing input images, also face challenges such as poor suppression of non-stationary burst noise, difficulty adapting to complex and changing scenes, and excessively smooth images after noise reduction.

[0003] To address the aforementioned problems, guided filtering has attracted significant attention from the academic community. The key idea behind guided filtering is to utilize an additional guiding image as a structural prior and transfer the structural information from the guiding image to the target image. In this way, edge information can be preserved while noise is suppressed. However, existing guided filtering algorithms suffer from poor filtering performance in flat regions and relatively poor high-frequency detail recovery. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a guided filtering noise reduction method, apparatus, electronic device, and readable storage medium.

[0005] In a first aspect, this application provides a guided filtering noise reduction method, the method comprising:

[0006] The original image is processed by the average pooling layer of a neural network to obtain low-frequency information;

[0007] High-frequency information is obtained based on the original image and the low-frequency information;

[0008] The high-frequency information is trained using a first convolutional network to obtain the noise reduction coefficients;

[0009] The high-frequency information is denoised according to the denoising coefficient to obtain a first image;

[0010] The low-frequency information is processed by a second convolutional network to remove artifacts, resulting in a second image.

[0011] The target image is determined based on the first image and the second image.

[0012] Preferably, obtaining high-frequency information based on the original image and the low-frequency information includes:

[0013] The high-frequency information is obtained by subtracting the low-frequency information from the original image.

[0014] Preferably, the noise reduction coefficient is a weight matrix, and each weight in the weight matrix represents the noise reduction strength.

[0015] Preferably, the step of training the high-frequency information through a first convolutional network to obtain noise reduction coefficients includes:

[0016] The high-frequency information is convolved using the first convolutional network to obtain a feature map;

[0017] The feature map is determined as the noise reduction coefficient.

[0018] Preferably, the step of performing noise reduction processing on the high-frequency information according to the noise reduction coefficient to obtain the first image includes:

[0019] The first image is obtained by multiplying the noise reduction coefficient and the high-frequency information.

[0020] Preferably, determining the target image based on the first image and the second image includes:

[0021] The first image and the second image are superimposed to obtain the target image.

[0022] Preferably, the first image is an image after noise removal and high-frequency texture enhancement, and the second image is an image after artifact removal.

[0023] Secondly, this application provides a guided filtering noise reduction device, the device comprising:

[0024] The first processing module is used to process the original image through the average pooling layer of the neural network to obtain low-frequency information;

[0025] The first acquisition module is used to acquire high-frequency information based on the original image and the low-frequency information;

[0026] The training module is used to train the high-frequency information through the first convolutional network to obtain the noise reduction coefficients;

[0027] The second acquisition module is used to perform noise reduction processing on the high-frequency information according to the noise reduction coefficient to obtain the first image;

[0028] The second processing module is used to perform artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image;

[0029] A determining module is used to determine a target image based on the first image and the second image.

[0030] Thirdly, this application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the computer program executes the guided filtering noise reduction method provided in the first aspect when the processor is running.

[0031] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, executes the guided filtering noise reduction method provided in the first aspect.

[0032] The guided filtering denoising method, apparatus, electronic device, and readable storage medium provided in this application process the original image through the average pooling layer of a neural network to obtain low-frequency information; obtain high-frequency information based on the original image and the low-frequency information; train the high-frequency information through a first convolutional network to obtain denoising coefficients; perform denoising processing on the high-frequency information based on the denoising coefficients to obtain a first image; perform artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image; and determine a target image based on the first image and the second image. This method is adaptable to image denoising in various scenarios, improves filtering effects, and enhances the recovery of high-frequency details. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation on the scope of protection of this application. In the various drawings, similar components are numbered similarly.

[0034] Figure 1 A flowchart of the guided filtering noise reduction method provided in this application is shown;

[0035] Figure 2 Another flowchart of the guided filtering noise reduction method provided in this application is shown;

[0036] Figure 3 A schematic diagram of the structure of the convolutional network provided in this application is shown;

[0037] Figure 4 A schematic diagram of the guided filtering noise reduction device provided in this application is shown.

[0038] Main icons: 400 - Guided filtering noise reduction device, 401 - First processing module, 402 - First acquisition module, 403 - Training module, 404 - Second acquisition module, 405 - Second processing module, 406 - Determination module. Detailed Implementation

[0039] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0040] The components of this application, typically described and illustrated in the accompanying drawings, can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0041] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations thereof.

[0042] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0043] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0044] In existing technologies, the key idea of ​​guided filtering is to use an additional guiding image as a structural prior and transfer the structural information of the guiding image to the target image. In this way, edge information of the image can be preserved while suppressing noise. However, existing guided filtering algorithms also suffer from poor filtering performance in flat regions and poor high-frequency detail recovery.

[0045] Example 1

[0046] This application provides a guided filtering noise reduction method.

[0047] For details, see Figure 1 The guided filtering noise reduction method includes steps S101-S106, and each step is explained below.

[0048] Step S101: The original image is processed through the average pooling layer of the neural network to obtain low-frequency information.

[0049] Please see Figure 2 The original image is input into a neural network and processed through an average pooling layer to obtain low-frequency information. The neural network can be a Unet network or other types of neural networks; no specific limitation is made here.

[0050] It should be noted that low-frequency information, or low-frequency components, represents areas in an image where grayscale values ​​change slowly. These areas correspond to large, flat regions in the image. Low-frequency information primarily provides a comprehensive measure of the overall intensity of the image. For an image, this means the content within the edges constitutes low-frequency information, which represents most of the image's information—its general outline and contours—and is an approximate representation of the image.

[0051] Step S102: Obtain high-frequency information based on the original image and the low-frequency information.

[0052] It should be further explained that high-frequency information, or high-frequency components, represents areas in an image where grayscale values ​​change drastically, corresponding to the image's edges and contours, noise, and details. For an image, after removing low-frequency information, the remaining information is the high-frequency information.

[0053] In one embodiment, step S102 includes: subtracting the low-frequency information from the original image to obtain the high-frequency information.

[0054] Please see Figure 2 The original image is subtracted from the low-frequency information output by the average pooling layer to obtain high-frequency information, which is then input into the first convolutional network.

[0055] See Figure 3 The first convolutional network consists of multiple sequentially connected convolutional modules, that is, the whole of multiple sequentially connected convolutional modules is called the first convolutional network.

[0056] Step S103: Train the high-frequency information using the first convolutional network to obtain the noise reduction coefficient.

[0057] In this application, the noise reduction coefficients are obtained by training the first convolutional network on the high-frequency information.

[0058] In one embodiment, the denoising coefficient is used to control the denoising intensity and enhance texture strength. The denoising coefficient is a weight matrix, and each weight in the weight matrix represents the denoising intensity. When the first convolutional network trains the high-frequency information, it needs to determine the high-frequency pixels and noise pixels of the high-frequency information. High-frequency pixels refer to edge and contour points and detail points. Higher weights are assigned to high-frequency pixels, and lower weights are assigned to noise pixels. For example, the weight of noise pixels is set to 0. The denoising coefficient is determined based on the weight values ​​of all high-frequency pixels and noise pixels.

[0059] In one embodiment, step S103 includes:

[0060] The high-frequency information is convolved using the first convolutional network to obtain a feature map;

[0061] The feature map is determined as the noise reduction coefficient.

[0062] For example, see Figure 2 The first convolutional network performs convolution operations on high-frequency information, and the output feature map is used as the noise reduction coefficient.

[0063] Step S104: Perform noise reduction processing on the high-frequency information according to the noise reduction coefficient to obtain the first image.

[0064] In this application, since the noise reduction coefficient is used to control the noise reduction intensity and enhance the texture intensity, the high-frequency information can be denoised according to the noise reduction coefficient to obtain a first image, which is an image after removing noise and enhancing the high-frequency texture.

[0065] In one embodiment, step S104 includes:

[0066] The first image is obtained by multiplying the noise reduction coefficient and the high-frequency information.

[0067] For example, see Figure 2 The noise reduction coefficient is multiplied by the high-frequency information to obtain the first image. The noise reduction coefficient is a weight matrix, and the high-frequency information is a pixel value matrix. Multiplying the two is equivalent to multiplying the weight value of the noise reduction coefficient by the pixel value of the high-frequency information. The noise reduction coefficient and the high-frequency information have the same size. For high-frequency pixels (i,j) in the high-frequency information, the weight value of (i,j) corresponding to the noise reduction coefficient is set to a larger value. The result of the multiplication enhances the information of the high-frequency pixels, thereby enhancing the high-frequency texture. Conversely, for noisy pixels (i,j) in the high-frequency information, if the weight value of (i,j) corresponding to the noise reduction coefficient is set to 0, the noise can be suppressed, thus achieving the effect of noise suppression.

[0068] Step S105: The low-frequency information is processed by a second convolutional network to remove artifacts, resulting in a second image.

[0069] In this application, the second convolutional network consists of multiple cascaded convolutional modules. The number of cascaded convolutional modules can be set according to user needs and is not limited here. This second convolutional network is trained and has appropriate weights. After performing artifact removal processing on low-frequency information using the second convolutional network, an image with artifacts removed can be obtained. Specifically, the second image is an image with low-frequency noise and artifacts such as false color removed.

[0070] See Figure 2 The low-frequency information is input into the second convolutional network, and the low-frequency information is convolved by the second convolutional network to obtain the second image.

[0071] Step S106: Determine the target image based on the first image and the second image.

[0072] In one embodiment, step S106 includes:

[0073] The first image and the second image are superimposed to obtain the target image.

[0074] For example, see Figure 2 The first and second images are superimposed to obtain the target image. This approach integrates artificial intelligence (AI) into the traditional guided filtering algorithm, enabling adaptive image denoising and reconstruction through a learnable mechanism. By adaptively removing noise and enhancing high-frequency texture details, the resulting target image is clearer and of better quality.

[0075] The AI-driven guided filtering method provided in this application is implemented through a learnable approach based on deep learning. Because it is data-driven, it can adapt to image denoising in various scenarios and has a stronger ability to suppress non-stationary burst noise. High-frequency information can be adaptively removed through cascaded convolutional networks to enhance high-frequency texture details, thereby improving the detail representation and image resolution. Similarly, low-frequency information can also be adaptively removed from low-frequency noise and artifacts such as false color through cascaded convolutional networks, ensuring a cleaner final reconstructed image.

[0076] The guided filtering denoising method provided in this application processes the original image through the average pooling layer of a neural network to obtain low-frequency information; obtains high-frequency information based on the original image and the low-frequency information; trains the high-frequency information through a first convolutional network to obtain denoising coefficients; performs denoising processing on the high-frequency information based on the denoising coefficients to obtain a first image; performs artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image; and determines a target image based on the first image and the second image. This method is adaptable to image denoising in various scenarios, improves filtering effect, and enhances high-frequency detail recovery.

[0077] Example 2

[0078] In addition, this application provides a guided filtering noise reduction device.

[0079] like Figure 4 As shown, the guided filtering noise reduction device 400 includes:

[0080] The first processing module 401 is used to process the original image through the average pooling layer of the neural network to obtain low-frequency information;

[0081] The first acquisition module 402 is used to acquire high-frequency information based on the original image and the low-frequency information;

[0082] Training module 403 is used to train the high-frequency information through the first convolutional network to obtain noise reduction coefficients;

[0083] The second acquisition module 404 is used to perform noise reduction processing on the high-frequency information according to the noise reduction coefficient to obtain a first image;

[0084] The second processing module 405 is used to perform artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image.

[0085] The determining module 406 is used to determine the target image based on the first image and the second image.

[0086] In one embodiment, the first acquisition module 402 is used to subtract the low-frequency information from the original image to obtain the high-frequency information.

[0087] In one embodiment, the noise reduction coefficient is a weight matrix, and each weight of the weight matrix represents the noise reduction strength.

[0088] In one embodiment, the training module 403 is further configured to perform convolution operations on the high-frequency information through the first convolutional network to obtain a feature map;

[0089] The feature map is determined as the noise reduction coefficient.

[0090] In one embodiment, the second acquisition module 404 is further configured to multiply the noise reduction coefficient and the high-frequency information to obtain the first image.

[0091] In one embodiment, the determining module 406 is further configured to superimpose the first image and the second image to obtain the target image.

[0092] In one embodiment, the first image is an image after noise removal and high-frequency texture enhancement, and the second image is an image after artifact removal.

[0093] The guided filtering noise reduction device 400 provided in this application can implement the guided filtering noise reduction method provided in Embodiment 1. To avoid repetition, it will not be described again here.

[0094] The guided filtering noise reduction device provided in this application processes the original image through the average pooling layer of a neural network to obtain low-frequency information; obtains high-frequency information based on the original image and the low-frequency information; trains the high-frequency information through a first convolutional network to obtain noise reduction coefficients; performs noise reduction processing on the high-frequency information based on the noise reduction coefficients to obtain a first image; performs artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image; and determines a target image based on the first image and the second image. This device is adaptable to image noise reduction in various scenarios, improves filtering effects, and enhances the recovery of high-frequency details.

[0095] Example 3

[0096] In addition, this application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the computer program executes the guided filtering noise reduction method provided in Embodiment 1 when it is run on the processor.

[0097] The electronic device provided in this application can implement the guided filtering noise reduction method provided in Embodiment 1. To avoid repetition, it will not be described again here.

[0098] Example 4

[0099] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the guided filtering noise reduction method provided in Embodiment 1.

[0100] In this application, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0101] The computer-readable storage medium provided in this application can implement the guided filtering noise reduction method provided in Embodiment 1. To avoid repetition, it will not be described again here.

[0102] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0103] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0104] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A guided filtering noise reduction method, characterized in that, The method includes: The original image is processed by the average pooling layer of a neural network to obtain low-frequency information; High-frequency information is obtained based on the original image and the low-frequency information; The high-frequency information is trained using a first convolutional network to obtain the noise reduction coefficients; The high-frequency information is denoised according to the denoising coefficient to obtain a first image; The low-frequency information is processed by a second convolutional network to remove artifacts, resulting in a second image. Determine the target image based on the first image and the second image; The step of training the high-frequency information through a first convolutional network to obtain noise reduction coefficients includes: The high-frequency information is convolved using the first convolutional network to obtain a feature map; The feature map is determined as the denoising coefficient, which is a weight matrix used to control the denoising intensity and enhance the texture strength; The step of performing noise reduction processing on the high-frequency information according to the noise reduction coefficient to obtain a first image includes: The first image is obtained by multiplying the noise reduction coefficient and the high-frequency information.

2. The method according to claim 1, characterized in that, The step of obtaining high-frequency information based on the original image and the low-frequency information includes: The high-frequency information is obtained by subtracting the low-frequency information from the original image.

3. The method according to claim 1, characterized in that, Each weight in the weight matrix represents the noise reduction intensity.

4. The method according to claim 1, characterized in that, Determining the target image based on the first image and the second image includes: The first image and the second image are superimposed to obtain the target image.

5. The method according to claim 1, characterized in that, The first image is an image after noise removal and high-frequency texture enhancement, and the second image is an image after artifact removal.

6. A guided filtering noise reduction device, characterized in that, The device includes: The first processing module is used to process the original image through the average pooling layer of the neural network to obtain low-frequency information; The first acquisition module is used to acquire high-frequency information based on the original image and the low-frequency information; The training module is used to train the high-frequency information through the first convolutional network to obtain the noise reduction coefficients; The second acquisition module is used to perform noise reduction processing on the high-frequency information according to the noise reduction coefficient to obtain the first image; The second processing module is used to perform artifact removal processing on the low-frequency information through a second convolutional network to obtain a second image; The determining module is used to determine the target image based on the first image and the second image; The training module is further configured to perform convolution operations on the high-frequency information through the first convolutional network to obtain a feature map; The feature map is determined as the denoising coefficient, which is a weight matrix used to control the denoising intensity and enhance the texture strength; The second acquisition module is further configured to multiply the noise reduction coefficient and the high-frequency information to obtain the first image.

7. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that executes the guided filtering noise reduction method according to any one of claims 1 to 5 when the processor is running.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when run on a processor, executes the guided filtering noise reduction method according to any one of claims 1 to 5.

Images