Image denoising method and device, electronic equipment, storage medium and program product

Abstract

Embodiments of the present application provide a denoising method and device for images, electronic equipment, storage medium and program product. The method comprises: generating a first denoised image by denoising an original input image through a convolutional neural network; determining a noise residual image based on the original input image and the first denoised image; obtaining a second denoised image by adjusting the contrast of the first denoised image; and generating a target denoised image by combining the second denoised image with the noise residual image and the attribute features of the original input image. The method of the present application breaks through the limitation of fixed mapping relationship in related technologies, and no longer relies on fixed mapping of the model for denoising. The method not only retains the advantages of efficient denoising of the convolutional neural network, but also designs noise addition by combining the noise residual image with the attribute features of the image, so that the denoising process has flexible adaptation ability. The method can adjust the image denoising by regulating the noise addition according to actual needs, so that the quality of the denoised image matches the actual needs.

Description

Technical Field

[0001] This application relates to the field of image processing, and more particularly to an image noise reduction method, apparatus, electronic device, storage medium, and program product. Background Technology

[0002] In the field of image processing, image denoising is a key technology. Its goal is to remove noise interference from images while preserving the effective information of the image to the greatest extent possible, so as to improve the visual quality of the image and the reliability of subsequent processing.

[0003] In related technologies, noise reduction methods are usually based on deep learning. Convolutional neural network methods based on deep learning directly learn the mapping relationship between noise and clean images through end-to-end training, and perform noise reduction processing according to the mapping relationship.

[0004] However, the above noise reduction methods are not very flexible, resulting in image quality that does not match actual needs. Summary of the Invention

[0005] This application provides an image noise reduction method, apparatus, electronic device, storage medium, and program product to improve noise reduction flexibility and achieve the technical effect of matching image quality with actual needs.

[0006] In a first aspect, embodiments of this application provide an image noise reduction method, comprising:

[0007] The original input image is denoised using a convolutional neural network to generate the first denoised image.

[0008] Based on the first denoised image and the original input image, determine the noise residual map;

[0009] The image contrast of the first denoised image is adjusted to generate a second denoised image with adjusted contrast.

[0010] Based on the attribute features of the original input image and the noise residual map, the second denoised image is subjected to noise back-addition processing to generate the target denoised image.

[0011] In one possible implementation, based on the attribute features of the original input image and the noise residual map, the second denoised image is subjected to noise back-addition processing to generate the target denoised image, including:

[0012] Based on the attribute features of the original input image, the original input image is partitioned to obtain partitioned images;

[0013] Based on the partitioned image, the second denoised image is subjected to noise re-addition processing to generate the third denoised image;

[0014] Determine target image information based on noise residual map;

[0015] Based on the target image information, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

[0016] In one possible implementation, the original input image is partitioned based on its attribute features to obtain partitioned images, including:

[0017] Based on a preset image filtering algorithm, the original input image is frequency-divided to obtain a first frequency region and a second frequency region, thus obtaining a partitioned image; or,

[0018] Based on the pixels of the original input image, the regions in the original input image with pixels greater than a preset pixel threshold are divided into first brightness regions, and the regions with pixels less than or equal to the preset pixel threshold are divided into second brightness regions, so as to obtain a partitioned image.

[0019] In one possible implementation, the second denoised image is subjected to noise re-addition processing based on the partitioned image to generate a third denoised image, including:

[0020] In response to user input on debugging operations for each region in the partitioned image, determine the weight debugging data for each region;

[0021] Based on the weighted adjustment data of each region, noise is added back to the second denoised image to generate the third denoised image.

[0022] In one possible implementation, determining target image information based on the noise residual map includes:

[0023] The original input image is denoised based on the noise residual map to generate a fourth denoised image.

[0024] The fourth denoised image is filtered to determine the target image information, which includes image contour information.

[0025] In one possible implementation, the third denoised image is subjected to noise re-addition processing again based on the target image information to generate the target denoised image, including:

[0026] Based on the original input image, determine the radial map of the original input image;

[0027] In response to the user's debugging input operation on the target image information, determine the weight debugging data of the target image information;

[0028] In response to user input on the radial plot of the image, determine the weighted debugging data for the radial plot of the image;

[0029] Based on the weighted adjustment data of the target image information and the weighted adjustment data of the image radial plot, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

[0030] Secondly, embodiments of this application provide an image noise reduction apparatus, comprising:

[0031] The generation module is used to denoise the original input image using a convolutional neural network to generate the first denoised image.

[0032] The determination module is used to determine the noise residual map based on the first denoised image and the original input image;

[0033] The generation module is also used to adjust the image contrast of the first denoised image to generate a second denoised image after contrast adjustment.

[0034] The generation module is also used to perform noise back-addition processing on the second denoised image based on the attribute features of the original input image and the noise residual map, so as to generate the target denoised image.

[0035] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0036] The memory stores instructions that the computer executes;

[0037] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0038] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0039] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0040] The image denoising method, apparatus, electronic device, storage medium, and program product provided in this application embodiment denoise the original input image to generate a first denoised image. Then, a noise residual map is determined based on the original input image and the first denoised image. The contrast of the first denoised image is adjusted to obtain a second denoised image. Finally, noise back-addition processing is performed on the second denoised image, combining the attribute features of the original input image and the noise residual map, to generate a target denoised image. The method of this application overcomes the limitations of fixed mapping relationships in related technologies, no longer relying on fixed mappings of the model for denoising. It retains the advantages of efficient denoising of convolutional neural networks, and through the noise back-addition design combining the noise residual map with image attribute features, the denoising process has flexible adaptability. It can adjust the image denoising by regulating the noise back-addition processing according to actual needs, so that the quality of the denoised image matches the actual requirements. Attached Figure Description

[0041] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0042] Figure 1 A schematic flowchart of an image noise reduction method provided in an embodiment of this application;

[0043] Figure 2 A schematic diagram of a first denoised image provided in an embodiment of this application;

[0044] Figure 3 A schematic diagram of a second denoised image provided in an embodiment of this application;

[0045] Figure 4 A flowchart illustrating a method for generating a target denoised image provided in an embodiment of this application;

[0046] Figure 5 A schematic diagram illustrating the generation of a target denoised image provided in an embodiment of this application;

[0047] Figure 6 A schematic diagram of a frequency division diagram provided in an embodiment of this application;

[0048] Figure 7 A schematic diagram of a light-dark partitioning map provided in an embodiment of this application;

[0049] Figure 8 A schematic diagram of a filtered image provided in an embodiment of this application;

[0050] Figure 9 A schematic diagram of an image radial pattern provided in an embodiment of this application;

[0051] Figure 10A schematic diagram of the structure of an image noise reduction device provided in an embodiment of this application;

[0052] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0053] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0054] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0055] Noise in an image is essentially non-target interference signals introduced during image acquisition or transmission. These signals are not valid content of the image itself, but they directly disrupt the visual coherence of the image. For example, they can create messy spots in flat areas or cause blurring or breaks at texture edges, thereby significantly reducing the user's ability to recognize image details and their subjective visual experience.

[0056] Image denoising, a key aspect of image processing, can be used to separate noise signals from valid information. This allows for precise suppression of noise components, eliminating their interference with the visual experience, while preserving, to a greater extent, the image's texture, edges, and color information. Ultimately, this results in a clearer and more accurate representation of the image's content.

[0057] In related technologies, deep learning is widely used for image denoising. Deep learning-based convolutional neural network methods directly learn the mapping relationship between noise and clean images through end-to-end training, and then perform denoising processing based on this mapping relationship.

[0058] However, the aforementioned image denoising methods based on convolutional neural networks generally adopt an end-to-end training approach, i.e., fixed input and fixed output. The weight parameters in the network cannot be manually adjusted, resulting in the output results not being able to be flexibly adjusted according to different scenarios and user preferences, leading to poor flexibility and inability to adapt to diverse needs.

[0059] Therefore, to address the aforementioned problems in related technologies, this application proposes an image denoising method. Specifically, a first denoised image is generated by performing denoising processing on the original input image using a convolutional neural network, and a noise residual map is determined based on the first denoised image and the original input image. After adjusting the image contrast of the first denoised image to obtain a second denoised image, noise back-addition processing is performed on the second denoised image based on the attribute features of the original input image and the noise residual map, thereby generating the target denoised image. The technical concept of this application not only relies on convolutional neural networks to achieve effective denoising, but also breaks the limitation of fixed end-to-end input and output through the noise back-addition design combining the noise residual map with image attribute features. This allows the denoised output to flexibly adapt to different scenarios and user needs, solving the problem of insufficient flexibility in related denoising methods.

[0060] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0061] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating an image noise reduction method provided in an embodiment of this application. The execution entity of this method can be an image noise reduction device. This image noise reduction device can be implemented through a computer program; it can also be implemented through a medium storing the relevant computer program, such as a USB flash drive and / or optical disc; or it can be implemented through a physical device integrating or installing the relevant computer program, such as a chip or electronic device. The electronic device can be a server, server cluster, smart terminal, etc. Figure 1 As shown, the method may include the following steps:

[0062] S101. The original input image is denoised using a convolutional neural network to generate the first denoised image.

[0063] The original input image is obtained and fed into a pre-trained convolutional neural network. The convolutional neural network performs feature extraction and noise filtering on the original input image, outputting the first denoised image, such as... Figure 2 As shown, Figure 2 This is a schematic diagram of a first denoised image provided in an embodiment of this application.

[0064] The training process includes acquiring training samples, building the network structure, and setting up the training framework. The network structure typically uses a residual network, where the model output is a negative noise map. The denoised output image is obtained by adding the input image to the negative noise map.

[0065] Since convolutional neural networks have stronger noise reduction capabilities and produce output images with less residual noise, this embodiment uses a convolutional neural network to perform noise reduction processing on the original input image.

[0066] S102. Determine the noise residual map based on the first denoised image and the original input image.

[0067] Based on the pixel data of the original input image and the first denoised image, a pixel-by-pixel difference operation is performed on the two images. The difference between the pixel value of the original input image and the pixel value of the first denoised image at the same coordinate position is calculated point by point. The difference results of all pixels are integrated to generate a noise residual map with the same size as the original input image.

[0068] S103. Adjust the image contrast of the first denoised image to generate a second denoised image with adjusted contrast.

[0069] The first denoised image is then subjected to subsequent post-processing steps, including image contrast adjustment and noise re-addition.

[0070] To make the output image more transparent and the information in the bright and dark areas more prominent, the contrast of the first denoised image needs to be adjusted.

[0071] Optionally, the first denoised image after convolutional neural network denoising is divided into multiple pixel sub-blocks of preset size to reduce local overexposure caused by single histogram equalization. The grayscale statistical histogram of each pixel sub-block is calculated to determine the number of pixels corresponding to each grayscale level. A contrast limit threshold is set, and pixels with grayscale levels exceeding this threshold in the histogram are cropped. The cropped excess pixels are evenly distributed across all grayscale levels in the histogram to obtain a corrected histogram, reducing excessive local contrast enhancement. Then, the corrected histogram of each pixel sub-block is equalized, stretching the grayscale range of the sub-block to a preset range through grayscale mapping to enhance the difference between bright and dark areas and display information within the sub-block. Finally, a bilinear interpolation algorithm is used to stitch all equalized pixel sub-blocks into a complete image. By eliminating the stitching edges between sub-blocks, a second denoised image with contrast adjustment is generated. Figure 3 As shown, Figure 3 This is a schematic diagram of a second denoised image provided in an embodiment of this application. Figure 3 and Figure 2 compared to, Figure 3 The displayed image has a more transparent overall effect, with more obvious bright and dark areas, and the image is clearer.

[0072] The reason why contrast enhancement was not performed on the original input image in this embodiment is that if contrast enhancement is performed on the original input image, noise will also be enhanced simultaneously. Since the first denoised image after denoising by the convolutional neural network has less residual noise, enhancing the contrast on the first denoised image can avoid this problem.

[0073] S104. Based on the attribute features of the original input image and the noise residual map, perform noise back-addition processing on the second denoised image to generate the target denoised image.

[0074] Convolutional neural network (CNN) denoising methods excel in noise reduction, but while efficiently removing image noise, they inevitably lose some image detail, leading to blurred image texture edges. This also means that the noise residual map obtained from the noise residual map determined by the first denoised image and the original input image contains not only the noise components to be removed but also valid image information that was mistakenly identified as noise and removed.

[0075] Therefore, in this embodiment, noise re-addition processing can also be performed on the second denoised image based on the attribute features of the original input image and the noise residual map. By performing noise re-addition processing, more detail information can be recovered as much as possible, thereby reducing the problem of blurred image texture edges.

[0076] In related technologies, if the user is not satisfied with the output denoised image, the convolutional neural network model needs to be retrained, resulting in low efficiency in the image denoising process. However, in this embodiment, by performing noise re-addition processing on the second denoised image and flexibly adjusting the re-addition intensity based on the attribute features of the original input image, the details and edge performance of the denoised image can be optimized in a targeted manner, quickly adapting to different user needs for denoised images without relying on model retraining, thereby significantly improving the efficiency of the image denoising process.

[0077] In the above embodiments of this application, a first denoised image is generated by denoising the original input image using a convolutional neural network. Then, a noise residual map is determined based on the original input image and the first denoised image. The contrast of the first denoised image is adjusted to obtain a second denoised image. Finally, noise back-addition processing is performed on the second denoised image, combining the attribute features of the original input image and the noise residual map, to generate the target denoised image. The method of this application overcomes the limitations of fixed mapping relationships in related technologies, no longer relying on fixed mappings of the model for denoising. It retains the advantages of efficient denoising using convolutional neural networks, and through the noise back-addition design combining the noise residual map with image attribute features, it gives the denoising process a flexible adaptability. This allows the image denoising to be adjusted according to actual needs by regulating the noise back-addition processing, ensuring that the quality of the denoised image matches the actual requirements.

[0078] Furthermore, based on the above embodiments, the following embodiments illustrate the process of performing noise back-addition processing on the second denoised image to generate the target denoised image based on the attribute features of the original input image and the noise residual map.

[0079] Please see 4. Figure 4 This application provides a flowchart illustrating a method for generating a target denoised image, which may include the following steps:

[0080] S401. Based on the attribute features of the original input image, the original input image is partitioned to obtain partitioned images. Based on the partitioned images, the second denoised image is subjected to noise back-addition processing to generate the third denoised image.

[0081] To facilitate a better understanding of the method in this embodiment, the following will be combined with... Figure 5 To explain, Figure 5 This is a schematic diagram illustrating the generation of a target denoised image, as provided in an embodiment of this application.

[0082] Optionally, the attribute feature can be frequency. Based on a preset image filtering algorithm, the frequency of the original input image is divided to obtain a first frequency region and a second frequency region, so as to obtain a partitioned image.

[0083] Since a smearing effect is prone to occur in the flat areas of the second denoised image, adding a larger degree of noise residual in the flat areas can solve this problem. Conversely, if a large degree of noise residual is added back in the textured areas of the second denoised image, the image edges are prone to being uneven and not straight. Therefore, it is necessary to reduce the degree of noise residual addition in the textured areas.

[0084] In this embodiment, in order to distinguish between the flat area and the textured area in the second denoised image, the image is frequency-divided using an image filtering method.

[0085] Specifically, such as Figure 5 As shown, the original input image noisy_img is frequency-divided using a Gaussian low-pass filtering algorithm to obtain a first frequency region, also known as the low-frequency region, and a second frequency region, also known as the high-frequency region, resulting in a frequency partition map freq_mask (or simply frequency division map). Figure 6 As shown, Figure 6 This is a schematic diagram of a frequency division diagram provided in an embodiment of this application, wherein the low-frequency region (dashed frame) corresponds to the flat region and the high-frequency region (solid frame) corresponds to the texture region.

[0086] Understandable, Figure 6 The annotations in the text do not fully indicate the low-frequency and high-frequency regions. Figure 6 The annotations in the document are for illustrative purposes only and do not limit this application.

[0087] After determining the partitions, in response to the user's debugging input operations on each region of the partitioned image, the weight debugging data of each region is determined, and based on the weight debugging data of each region, noise re-addition processing is performed on the second denoised image to generate the third denoised image.

[0088] Specifically, using the frequency division map obtained above as the weight map, the weight adjustment data input by the user for the low-frequency and high-frequency regions is obtained. For example, a higher weight value is set for the low-frequency region and a lower weight value is set for the high-frequency region. Then, the regional weight adjustment parameter w1 is obtained based on the weight values ​​corresponding to different regions. freq_mask.

[0089] The noise residual map (noise_map) is then multiplied pixel-by-pixel with the weight map of the region to obtain a weighted residual map, which is then added back to the second denoised image (denoised_img) to finally obtain the third denoised image (denoised_img1). This results in less residual noise in high-frequency regions and more residual noise in low-frequency regions, which solves the smearing problem and makes the edges smoother.

[0090] Optionally, the attribute feature can be pixels. Based on the pixels of the original input image, the regions in the original input image with pixels greater than a preset pixel threshold are divided into first brightness regions, and the regions with pixels less than or equal to the preset pixel threshold are divided into second brightness regions, so as to obtain a partitioned image.

[0091] Because the effective signal is weak in darker areas of an image, the strong denoising capability of convolutional neural networks can remove almost all the effective information. Therefore, a greater degree of noise residual re-addition is needed in these areas to recover the removed effective information. However, if too much noise is re-added to bright areas of the image, it will mask the effective information and reduce image quality. Therefore, different levels of noise residual re-addition processing are needed for bright and dark areas of the denoised image.

[0092] In this embodiment, in order to distinguish between the bright and dark areas in the second denoised image, the image is partitioned according to its pixels.

[0093] Specifically, such as Figure 5 As shown, a preset pixel threshold, such as the pixel mean, is obtained for the original input image noisy_img. Regions with pixel values ​​less than or equal to the pixel mean are classified as second brightness regions (also called dark regions), and regions with pixel values ​​greater than the pixel mean are classified as first brightness regions (also called bright regions), resulting in a brightness-dark partition map light_mask. Figure 7 As shown, Figure 7 This is a schematic diagram of a light-dark partitioning map provided in an embodiment of this application, wherein the dashed frame corresponds to the dark area and the solid frame corresponds to the bright area.

[0094] Understandable, Figure 7 The annotations in the document are for illustrative purposes only and do not limit this application.

[0095] After determining the partitions, in response to the user's debugging input operations on each region of the partitioned image, the weight debugging data of each region is determined, and based on the weight debugging data of each region, noise re-addition processing is performed on the second denoised image to generate the third denoised image.

[0096] Specifically, using the aforementioned light-dark partition map as a weight map, the user-input weight adjustment data for dark and bright areas is obtained. For example, a higher weight value is set for dark areas and a lower weight value is set for bright areas. Then, based on the weight values ​​corresponding to different areas, the area weight adjustment parameter w2 is obtained. light_mask.

[0097] The noise residual map (noise_map) is then multiplied pixel-by-pixel with the weight map of the corresponding region to obtain a weighted residual map. This weighted residual map is then added back to the second denoised image (denoised_img) to finally obtain the third denoised image (denoised_img1). This results in more residual noise in dark areas and less residual noise in bright areas, thereby improving the image display quality.

[0098] S402. Determine the target image information based on the noise residual map.

[0099] After generating the third denoised image through the above steps, the image details still lack sharpness and clarity. Therefore, this embodiment can also obtain the high-frequency information of the image, i.e., the target image information. This is because high-frequency information usually contains contour details and noise. By adding back the high-frequency information, this application can increase the contour details of the image.

[0100] Low-frequency image filtering removes high-frequency information such as edges and noise from an image while preserving low-frequency information, making it a fundamental image denoising method. For example, the commonly used Gaussian filter determines the filtering strength, i.e., the denoising power, by controlling the size and variance of the filter kernel. However, some noise and image details belong to the same high-frequency information, and low-frequency image filtering methods struggle to distinguish between them, leading to situations where noise and details are simultaneously removed or preserved. Therefore, this application utilizes the characteristics of low-frequency image filtering to obtain the high-frequency information of an image through image filtering methods.

[0101] One possible implementation is to perform noise reduction processing on the original input image based on the noise residual map to generate a fourth noise-reduced image, and then perform filtering processing on the fourth noise-reduced image to determine the target image information, which includes image contour information.

[0102] like Figure 5 As shown, the original input image noisy_img is moderately denoised using the noise residual map noise_map and the user-input debugging denoising weights W3, resulting in the fourth denoised image W3. The fourth denoised image, `noise_map+noisy_img`, is processed using a filtering algorithm to obtain the high-frequency information `hf_img`. For example... Figure 8 As shown, Figure 8 This is a schematic diagram of a filtered image provided in an embodiment of this application. The image contains detailed information such as the image's outline and texture, and has minimal residual noise.

[0103] S403. Based on the target image information, perform noise re-addition processing on the third denoised image again to generate the target denoised image.

[0104] Based on the original input image, determine the radial graph of the original input image.

[0105] Due to the limitations of the camera lens, the original input image often exhibits a higher signal-to-noise ratio in the central region compared to the corner regions. Therefore, combining the image's radial plot, such as... Figure 9 As shown, Figure 9 This is a schematic diagram of an image radial map provided in an embodiment of this application. Taking the center of the original input image noisy_img as the origin, the distance from each pixel to the center is calculated to generate an image radial map radial_img, which reflects the distribution characteristics of "high signal-to-noise ratio at the center and low signal-to-noise ratio at the four corners".

[0106] In response to the user's debugging input operation on the target image information, the weight debugging data of the target image information is determined, and in response to the user's debugging input operation on the radial plot of the image, the weight debugging data of the radial plot of the image is determined.

[0107] like Figure 5 As shown, the weight adjustment data w4 input by the user for high-frequency information hf_img and the weight adjustment data w5 input for the radial image radial_img are obtained and fused into a detail weight map.

[0108] Based on the weighted adjustment data of the target image information and the weighted adjustment data of the image radial plot, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

[0109] Based on the weight adjustment data w4 of the target image information, the weight adjustment parameters w4 of the target image information, i.e., the high-frequency information, are obtained. hf_img, based on the weight adjustment data w5 of the radial plot of the image, obtains the weight adjustment parameters w5 of the radial plot of the image. radial_img.

[0110] Then, based on the above-mentioned adjustment parameters, the target image information, i.e. high-frequency information, is added back to the third noise-reduced image to perform noise addition processing again, and finally the target noise-reduced image is generated.

[0111] In the above embodiments of this application, the step-by-step processing method of performing a first round of noise back-addition based on feature partitioning of the original input image attribute features, and then performing a second round of noise back-addition based on the target image information determined by the noise residual map, can improve the targeting of noise back-addition. This method relies on feature partitioning to achieve differentiated noise residual back-addition, specifically compensating for the loss of details in different regions during convolutional neural network denoising. Furthermore, by extracting target image information from the noise residual map for a second back-addition, it further supplements key image details. This dual back-addition design can further improve the display quality of the denoised image. Simultaneously, it eliminates the need to retrain the model, flexibly adapting to different needs, and significantly improves the flexibility of denoising and the quality of image output compared to related technologies.

[0112] Please see Figure 10 , Figure 10 This is a schematic diagram of the structure of an image noise reduction device provided in an embodiment of this application, as shown below. Figure 10 As shown, it includes:

[0113] The generation module 1001 is used to perform noise reduction processing on the original input image through a convolutional neural network to generate a first noise-reduced image.

[0114] The determination module 1002 is used to determine the noise residual map based on the first denoised image and the original input image.

[0115] The generation module 1001 is also used to adjust the image contrast of the first denoised image to generate a second denoised image after contrast adjustment.

[0116] The generation module 1001 is also used to perform noise back-addition processing on the second denoised image based on the attribute features of the original input image and the noise residual map to generate the target denoised image.

[0117] In one possible implementation, the generation module 1001 is specifically used for:

[0118] Based on the attribute features of the original input image, the original input image is partitioned to obtain partitioned images.

[0119] Based on the partitioned image, the second denoised image is subjected to noise re-addition processing to generate the third denoised image.

[0120] The target image information is determined based on the noise residual map.

[0121] Based on the target image information, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

[0122] In one possible implementation, the generation module 1001 is specifically used for:

[0123] Based on a preset image filtering algorithm, the frequencies of the original input image are divided to obtain a first frequency region and a second frequency region, thus creating a partitioned image. Alternatively,

[0124] Based on the pixels of the original input image, the regions in the original input image with pixels greater than a preset pixel threshold are divided into first brightness regions, and the regions with pixels less than or equal to the preset pixel threshold are divided into second brightness regions, so as to obtain a partitioned image.

[0125] In one possible implementation, the generation module 1001 is specifically used for:

[0126] In response to user input on debugging operations for each region in the partitioned image, determine the weight debugging data for each region.

[0127] Based on the weighted adjustment data of each region, noise is added back to the second denoised image to generate the third denoised image.

[0128] In one possible implementation, the generation module 1001 is specifically used for:

[0129] The original input image is denoised based on the noise residual map to generate a fourth denoised image.

[0130] The fourth denoised image is filtered to determine the target image information, which includes image contour information.

[0131] In one possible implementation, the generation module 1001 is specifically used for:

[0132] Based on the original input image, determine the radial graph of the original input image.

[0133] In response to the user's debugging input operation on the target image information, the weight debugging data of the target image information is determined.

[0134] In response to user input on the radial plot of the image, determine the weighted debugging data for the radial plot of the image.

[0135] Based on the weighted adjustment data of the target image information and the weighted adjustment data of the image radial plot, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

[0136] The image noise reduction device provided in this embodiment can perform the image noise reduction method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0137] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 11 As shown, the electronic device provided in this embodiment includes at least one processor 1101 and a memory 1102. Optionally, the device further includes a communication component 1103. The processor 1101, memory 1102, and communication component 1103 are connected via a bus 1104.

[0138] In a specific implementation, at least one processor 1101 executes computer execution instructions stored in memory 1102, causing at least one processor 1101 to perform the above-described method.

[0139] The specific implementation process of processor 1101 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0140] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0141] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0142] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0143] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0144] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0145] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory, electrically erasable programmable read-only memory, erasable programmable read-only memory, programmable read-only memory, read-only memory, magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0146] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0147] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0148] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0150] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0151] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0152] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for denoising images, characterized in that, include: The original input image is denoised using a convolutional neural network to generate the first denoised image. Based on the first denoised image and the original input image, a noise residual map is determined; The image contrast of the first denoised image is adjusted to generate a second denoised image with adjusted contrast. Based on the attribute features of the original input image and the noise residual map, the second denoised image is subjected to noise back-addition processing to generate the target denoised image.

2. The method according to claim 1, characterized in that, The step of performing noise back-addition processing on the second denoised image based on the attribute features of the original input image and the noise residual map to generate the target denoised image includes: Based on the attribute features of the original input image, the original input image is partitioned to obtain partitioned images; Based on the partitioned image, the second denoised image is subjected to noise re-addition processing to generate the third denoised image; Based on the noise residual map, the target image information is determined; Based on the target image information, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

3. The method according to claim 2, characterized in that, The step of partitioning the original input image based on its attribute features to obtain partitioned images includes: Based on a preset image filtering algorithm, the original input image is frequency-divided to obtain a first frequency region and a second frequency region, thus obtaining the partitioned image; or, Based on the pixels of the original input image, the regions in the original input image with pixels greater than a preset pixel threshold are divided into a first brightness region, and the regions with pixels less than or equal to the preset pixel threshold are divided into a second brightness region, so as to obtain the partitioned image.

4. The method according to claim 2, characterized in that, The step of performing noise re-addition processing on the second denoised image based on the partitioned image to generate a third denoised image includes: In response to user input operations for debugging each region in the partitioned image, determine the weight debugging data for each region; Based on the weighted adjustment data of each region, the second denoised image is subjected to noise re-addition processing to generate the third denoised image.

5. The method according to claim 2, characterized in that, The step of determining the target image information based on the noise residual map includes: The original input image is denoised based on the noise residual map to generate a fourth denoised image. The fourth denoised image is filtered to determine the target image information, which includes image contour information.

6. The method according to claim 5, characterized in that, The step of performing noise re-addition processing on the third denoised image again based on the target image information to generate the target denoised image includes: Based on the original input image, determine the image radial map of the original input image; In response to the user's debugging input operation on the target image information, the weight debugging data of the target image information is determined; In response to the user's debugging input operation on the radial plot of the image, the weight debugging data of the radial plot of the image is determined; Based on the weighted adjustment data of the target image information and the weighted adjustment data of the image radial plot, the third denoised image is subjected to noise re-addition processing again to generate the target denoised image.

7. An image noise reduction device, characterized in that, include: The generation module is used to denoise the original input image using a convolutional neural network to generate the first denoised image. The determining module is used to determine a noise residual map based on the first denoised image and the original input image; The generation module is further configured to adjust the image contrast of the first denoised image to generate a second denoised image with adjusted contrast. The generation module is further configured to perform noise back-addition processing on the second denoised image based on the attribute features of the original input image and the noise residual map, to generate the target denoised image.

8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images