Low-frequency noise processing method and device

Abstract

The invention provides a low-frequency noise processing method and device. According to the method, down-sampling is performed on data to be subjected to noise reduction so as to reduce the data processing amount of subsequent tasks and low-frequency noise in the data to be subjected to noise reduction. And performing guide filtering on the chrominance information according to the brightness information in the down-sampling data so as to establish a statistical relationship between the brightness and the chrominance. In combination with an up-sampling operation, a statistical relationship between brightness and chromaticity is mapped to a resolution space corresponding to data to be denoised, so that low-frequency noise in the data to be denoised can be suppressed based on brightness information representing high-frequency features and edge detail features and the statistical relationship between the brightness and the chromaticity, and image details are reserved at the same time. Efficient noise reduction can be realized in a scene with limited hardware resources, and the image quality is improved.

Description

Technical Field

[0001] The present application relates to the technical field of image noise reduction, and in particular to a method and device for processing low-frequency noise. Background Art

[0002] Image denoising is a crucial step in image processing. It can remove graininess, speckling, blur, and other artifacts from images, enhancing the effectiveness of details. Furthermore, image denoising facilitates subsequent image processing tasks, such as object detection, image segmentation, and feature extraction, improving the accuracy of these tasks.

[0003] In the process of image denoising, the denoised pixel can be obtained by taking all the pixels in the neighborhood of the current pixel and performing a weighted average. The larger the neighborhood selected by this denoising method, the more pixels will be involved in the denoising, and the better the denoising effect can be achieved.

[0004] However, a noise reduction method that only relies on increasing the neighborhood to achieve better noise reduction effects will occupy more hardware resources and increase hardware power consumption, leading to increased hardware design and manufacturing costs. Summary of the Invention

[0005] The present application provides a low-frequency noise processing method and device to solve the problem of difficult balance between image noise reduction effect and hardware resource requirements.

[0006] In a first aspect, an embodiment of the present application provides a low-frequency noise processing method, comprising:

[0007] Upon receiving the data to be denoised that conforms to the target data format, downsampling the data to be denoised to obtain downsampled data;

[0008] performing guided filtering on the first color information in the downsampled data based on the first luminance information in the downsampled data to obtain guided filtering conversion coefficients; the guided filtering conversion coefficients are used to describe the statistical relationship between luminance and chrominance in the image corresponding to the downsampled data;

[0009] Upsampling the guided filter conversion coefficients to map the statistical relationship between luminance and chrominance represented by the guided filter conversion coefficients to a target resolution space; the target resolution space is a resolution space corresponding to the data to be denoised;

[0010] fusing the guided filter conversion coefficient and the second brightness information in the data to be denoised to obtain second color information;

[0011] A target image is output based on the second brightness information and the second color information; low-frequency noise in the target image is lower than low-frequency noise in the image corresponding to the data to be denoised.

[0012] In some feasible embodiments, the target data format includes a YUV format; the first color information includes a first color component and a second color component, and the first color component and the second color component describe different chromaticities; and performing downsampling on the data to be denoised further includes:

[0013] First brightness information, a first color component, and a second color component in the downsampled data are obtained.

[0014] In some feasible embodiments, the step of performing guided filtering on the first color information in the downsampled data based on the first brightness information in the downsampled data to obtain the guided filtering conversion coefficient includes:

[0015] performing guided filtering on the first color component based on the first luminance information to obtain a first guided filtering conversion coefficient and a second guided filtering conversion coefficient corresponding to the first color component;

[0016] Guided filtering is performed on the second color component based on the first luminance information to obtain a third guided filtering conversion coefficient and a fourth guided filtering conversion coefficient corresponding to the second color component.

[0017] In some feasible embodiments, the second color information includes a third color component and a fourth color component; the chroma described by the third color component and the fourth color component are different, the chroma described by the third color component and the first color component are the same, and the chroma described by the fourth color component and the second color component are the same; the low-frequency noise of the third color component is lower than that of the first color component; and the step of fusing the guided filter conversion coefficient with the second brightness information in the data to be denoised includes:

[0018] The second brightness information, the first guided filter conversion coefficient, and the second guided filter conversion coefficient are fused based on a first fusion formula to obtain a third color component; the first fusion formula is as follows:

[0019] U_N_NR=U_A_N-1*Y_N+U_B_N-1;

[0020] Among them, U_N_NR is the third color component; U_A_N-1 is the first guided filter conversion coefficient; U_B_N-1 is the second guided filter conversion coefficient; Y_N is the second brightness information;

[0021] The second brightness information, the third guided filter conversion coefficient, and the fourth guided filter conversion coefficient are fused based on a second fusion formula to obtain a fourth color component; the second fusion formula is as follows:

[0022] V_N_NR=V_A_N-1*Y_N+V_B_N-1;

[0023] Among them, V_N_NR is the fourth color component; V_A_N-1 is the third guided filter conversion coefficient; V_B_N-1 is the fourth guided filter conversion coefficient.

[0024] In some feasible embodiments, the step of performing downsampling processing on the data to be denoised includes:

[0025] Setting a downsampling ratio, wherein the downsampling ratio is used to reduce the image size corresponding to the downsampled data to at least one fourth of the image size corresponding to the data to be denoised;

[0026] Downsampling processing is performed on the data to be denoised according to the downsampling ratio.

[0027] In some feasible embodiments, the following further comprises:

[0028] When receiving the data to be denoised that does not conform to the target data format, the data to be denoised that does not conform to the target data format is converted into data in a YUV format.

[0029] In some feasible embodiments, the step of outputting a target image based on the second brightness information and the second color information includes:

[0030] When the received initial format of the data to be denoised does not conform to the target format, performing inverse conversion on the second luminance information and the second color information based on a conversion relationship between the YUV format and the initial format of the data to be denoised, so that an output format of the target image conforms to the initial format of the data to be denoised;

[0031] The output data format is the target image in the initial format.

[0032] In a second aspect, an embodiment of the present application provides a low-frequency noise processing device, comprising: a downsampling module, a guided filter coefficient calculation module, an upsampling module, and a guided filter application module;

[0033] The downsampling module is configured to, upon receiving the data to be denoised that conforms to the target data format, perform downsampling on the data to be denoised to obtain downsampled data;

[0034] The guided filter coefficient calculation module is used to perform guided filtering on the first color information in the downsampled data based on the first luminance information in the downsampled data to obtain guided filter conversion coefficients; the guided filter conversion coefficients are used to describe the statistical relationship between luminance and chrominance in the image corresponding to the downsampled data;

[0035] The upsampling module is used to perform upsampling on the guided filter conversion coefficients to map the statistical relationship between luminance and chrominance represented by the guided filter conversion coefficients to a target resolution space; the target resolution space is a resolution space corresponding to the data to be denoised;

[0036] The guided filter application module is used to fuse the guided filter conversion coefficient and the second brightness information in the data to be denoised to obtain second color information;

[0037] The guided filtering application module is further configured to output a target image based on the second brightness information and the second color information; the low-frequency noise in the target image is lower than the low-frequency noise in the image corresponding to the data to be denoised.

[0038] In some feasible embodiments, a color space conversion module is further included;

[0039] The color space conversion module is used to convert the data format of the data to be denoised into the target data format when the data format of the data to be denoised does not conform to the target data format; the target data format includes a YUV format.

[0040] In some feasible embodiments, a color space inverse conversion module is further included;

[0041] The color space inverse conversion module is configured to, when the data format of the data to be denoised does not conform to the target data format, perform an inverse conversion on the second brightness information and the second color information based on a conversion relationship between the YUV format and the initial format of the data to be denoised, so that the output format of the target image conforms to the initial format of the data to be denoised;

[0042] The color space inverse conversion module is further configured to output a target image in the initial format. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0044] Figure 1 A flow chart of a low-frequency noise processing method provided in an embodiment of the present application;

[0045] Figure 2 A schematic diagram of guided filtering execution provided in an embodiment of the present application;

[0046] Figure 3 This is a module composition diagram of the first low-frequency noise processing device provided in an embodiment of the present application;

[0047] Figure 4 A diagram showing the module composition of a second low-frequency noise processing device provided in an embodiment of the present application;

[0048] Figure 5 This is a module composition diagram of the third low-frequency noise processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The following embodiments are described in detail, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numbers in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following embodiments are not intended to represent all possible implementations consistent with the present application. They are merely examples of systems and methods consistent with certain aspects of the present application, as detailed in the claims.

[0050] Image noise reduction is a crucial step in image processing in an image signal processor (ISP). A common noise reduction method involves obtaining all pixels in the neighborhood of the current pixel and performing a weighted average of these pixels to obtain the denoised pixel.

[0051] The noise reduction capabilities of the above-mentioned denoising techniques are almost entirely dependent on the number of pixels in the neighborhood. In other words, the larger the neighborhood, the more pixels that can participate in the noise reduction, and the better the noise reduction effect.

[0052] Image noise, on the other hand, can be categorized into medium- and high-frequency noise and low-frequency noise based on its spatial frequency characteristics. The lower the frequency of the noise, the more demanding the denoising capability is, meaning higher requirements are placed on the number of pixels involved in noise reduction and the size of the pixel neighborhood. Removing low-frequency color noise in images is particularly urgent, as the human eye is particularly sensitive to this type of noise.

[0053] Obviously, the most direct way to address low-frequency color noise is to increase the number of neighborhood pixels, that is, to expand the neighborhood size. However, since image signal processors are generally semiconductor integrated circuits, expanding the neighborhood size requires more hardware resources. This will significantly increase the difficulty of hardware design and manufacturing costs, and thus make key performance indicators such as chip production cost and power consumption inconsistent with design standards.

[0054] In order to solve the above problems, the embodiment of the present application provides a low-frequency noise processing method, which can be applied to an image signal processor to provide a low-frequency noise reduction function during image processing. Figure 1 As shown, the low-frequency noise processing method includes:

[0055] S100: When receiving data to be denoised that conforms to a target data format, downsampling is performed on the data to be denoised to obtain downsampled data.

[0056] In some embodiments, the circuitry for executing the low-frequency noise processing method is referred to as a low-frequency noise processing module. Upon receiving the data to be denoised from the upper-level module, and if the data format of the data to be denoised conforms to the target data format, the low-frequency noise processing module may downsample the data to obtain downsampled data.

[0057] In some embodiments, the target data format may be a YUV format, where the Y component represents luminance information, and the U and V components represent chrominance information. The image corresponding to the data to be denoised may be the image to be denoised. Thus, downsampling can reduce the size of the image to be denoised, i.e., reduce the size of the image represented by each data channel of the image to be denoised.

[0058] For example, the initial Y, U, and V channel data of the data to be denoised are recorded as Y_N, U_N, and V_N, respectively. After downsampling, the denoised data can be obtained as Y_DS_N, U_DS_N, and V_DS_N, where Y_DS_N is the reduced luminance information, and U_DS_N and V_DS_N are the reduced chrominance information. In this way, downsampling is equivalent to performing noise reduction within a larger neighborhood, effectively reducing the impact of noise on image quality. Furthermore, downsampling can reduce the amount of data processing in subsequent steps and reduce computational complexity. This optimizes hardware resource utilization while ensuring effective noise reduction, improving the overall performance of the image signal processor.

[0059] In some embodiments, the Y channel, U channel, and V channel data can be first extracted from the data to be denoised, and then the Y channel, U channel, and V channel data can be downsampled to obtain Y_DS_N, U_DS_N, and V_DS_N, respectively. That is, the step of downsampling the data to be denoised also includes:

[0060] First brightness information, a first color component, and a second color component in the downsampled data are obtained.

[0061] Here, the Y channel data represents the first luminance information, while the U and V channel data represent the first and second color components of the first color information, respectively. By downsampling the Y, U, and V channels separately, we can more precisely control the noise reduction effect of each channel, ensuring that image detail and color balance are maintained while reducing noise.

[0062] In some embodiments, when downsampling the denoised data, it is necessary to set a downsampling ratio as needed. That is, the steps of downsampling the denoised data include:

[0063] A downsampling ratio is set, where the downsampling ratio is used to at least reduce the image size corresponding to the downsampled data to a quarter of the image size corresponding to the data to be denoised.

[0064] Downsampling processing is performed on the data to be denoised according to the downsampling ratio.

[0065] It can be understood that the effect of downsampling is reflected in reducing the size of the image to be denoised on the one hand, and averaging the noise over a large scale on the other hand, thereby effectively reducing the interference of low-frequency noise.

[0066] In some embodiments, the downsampling ratio can be set to 4:1, which means that the image size after downsampling is reduced to one-fourth of the original image size. This setting helps significantly improve noise reduction when processing larger-scale images, thereby improving image quality.

[0067] It is understandable that, depending on the specific size of the image to be denoised, a larger downsampling ratio can be selected for processing the image to be denoised. For example, when the downsampling ratio is set to 8:1, the image size will be reduced to one-eighth of the original size, further reducing noise interference and improving image clarity. At the same time, this large-scale downsampling can effectively reduce the amount of data processing and optimize the allocation of computing resources.

[0068] It should be noted that the embodiments of the present application do not limit the specific method of downsampling. The specific method of downsampling may adopt, but is not limited to, bilinear downsampling, bicubic downsampling, and other downsampling methods. The embodiments of the present application aim to reduce the subsequent data processing volume and achieve a certain degree of low-frequency noise reduction effect based on the downsampling method, thereby improving the processing efficiency of the system while ensuring image quality.

[0069] S200: Performing guided filtering on first color information in the downsampled data based on first brightness information in the downsampled data to obtain guided filtering conversion coefficients.

[0070] like Figure 2 As shown, in some embodiments, based on the correlation between Y_DS_N and U_DS_N and V_DS_N, a conversion relationship between Y_DS_N and U_DS_N and V_DS_N can be established, that is, guided filtering is performed on the first color component and the second color component respectively based on the first luminance information to obtain guided filtering conversion coefficients corresponding to the first color component and the second color component. That is, the step of performing guided filtering on the first color information in the downsampled data based on the first luminance information in the downsampled data to obtain the guided filtering conversion coefficients includes:

[0071] Guided filtering is performed on the first color component based on the first luminance information to obtain a first guided filtering conversion coefficient and a second guided filtering conversion coefficient corresponding to the first color component.

[0072] Guided filtering is performed on the second color component based on the first luminance information to obtain a third guided filtering conversion coefficient and a fourth guided filtering conversion coefficient corresponding to the second color component.

[0073] In some embodiments, the guided filtering conversion coefficients corresponding to the first color component include a first guided filtering conversion coefficient U_A_DS_N-1 and a second guided filtering conversion coefficient U_B_DS_N-1, and the guided filtering conversion coefficients corresponding to the second color component include a third guided filtering conversion coefficient V_A_DS_N-1 and a fourth guided filtering conversion coefficient V_B_DS_N-1.

[0074] Taking the guided filter conversion coefficient corresponding to the first color component as an example, U_A_DS_N-1 can be used to control the local correlation between brightness and chrominance in the image. For example, at the edge of the image, U_A_DS_N-1 can use a larger value to maintain the consistency of chrominance changes with brightness, thereby helping to protect edge color transitions and alleviate image blur. In flat areas of the image, U_A_DS_N-1 uses a smaller value to ensure smooth chrominance changes, thereby avoiding over-emphasizing chrominance details and causing color distortion.

[0075] U_B_DS_N-1 can be used to control the baseline shift of chromaticity in an image. For example, in flat areas, U_B_DS_N-1 uses a smaller value to maintain chromaticity stability and achieve smooth noise reduction. In edge areas, U_B_DS_N-1 uses a larger value to adapt to chromaticity changes, enhance edge contrast, and ensure clear details. This adaptive adjustment achieves uniform noise reduction in flat areas and helps alleviate color deviation in the image.

[0076] It should be noted that the specific implementation method of guided filtering is not specifically limited in the embodiments of the present application. The specific implementation method of guided filtering can be selected according to actual needs to achieve different effects, thereby adapting to the needs of different fields such as security, industrial shooting, etc.

[0077] For example, when performing guided filtering on the first color component and the second color component based on the first brightness information, a sliding window can be established by defining a local window, and then the parameters used to calculate the guided filtering within each window can be statistically analyzed. The parameters used to calculate the guided filtering can include the luminance mean, chrominance mean, luminance variance, and luminance-chrominance covariance within each window. In this way, based on these parameters, the guided filtering conversion coefficient corresponding to the first color component or the second color component can be calculated in combination with the guided filtering coefficient calculation formula. In this way, the filtering intensity can be adaptively adjusted in different areas to ensure that the image maintains the authenticity of details and colors while reducing noise.

[0078] In this way, the relationship between luminance and chrominance can be established by calculating the guided filter conversion coefficients. Dynamically generated guided filter conversion coefficients can be used to suppress noise while preserving image edge features. This is particularly beneficial given that the parameters used to calculate the guided filter have large-scale statistical properties, further facilitating the suppression of low-frequency noise in the image.

[0079] S300: Upsampling the guided filter conversion coefficients to map the statistical relationship between luminance and chrominance represented by the guided filter conversion coefficients to a target resolution space.

[0080] It can be understood that the target resolution space is the resolution space corresponding to the data to be denoised. Furthermore, in some embodiments, the upsampling operation can, on the one hand, enlarge the size corresponding to the guided filter conversion coefficient to the same size as the brightness information Y_N corresponding to the data to be denoised, and on the other hand, map the statistical relationship between brightness and chromaticity represented by the guided filter conversion coefficient to the target resolution space, thereby restoring the statistical relationship between brightness and chromaticity of the image described by the upsampled guided filter conversion coefficient in the target resolution space, ensuring that the image still maintains delicate details and accurate color restoration at a higher resolution, effectively improving the overall quality of the image. Through this fine upsampling process, it is not only helpful to alleviate the problem of detail loss in the noise reduction process, but also to further optimize the texture performance of the image during the amplification process, so that the image presents a natural visual effect at different scales, meeting the needs of high-resolution application scenarios.

[0081] In some embodiments, the specific implementation of upsampling may include, but is not limited to, bilinear upsampling and bicubic upsampling. The specific implementation of upsampling may correspond to the specific implementation of downsampling. For example, if bilinear downsampling is used in the downsampling stage, bilinear upsampling may be used in the upsampling stage accordingly.

[0082] The upsampled guided filter conversion coefficients can be recorded as U_A_N-1, U_B_N-1, V_A_N-1, and V_B_N-1, respectively. The upsampled guided filter conversion coefficients can be fused with the brightness information of the data to be denoised to determine the U channel data and V channel data in the target image.

[0083] S400: Fusing the guided filter conversion coefficient and the second brightness information in the data to be denoised to obtain second color information.

[0084] In some embodiments, the Y channel data in the data to be denoised is used to represent the second luminance information, which corresponds to the initial resolution of the data to be denoised. Therefore, the second luminance information contains the high-frequency edges and detail features of the image to be denoised. The up-sampled guided filter conversion coefficient can map the statistical relationship between luminance and chrominance to the target resolution space after up-sampling. Therefore, during the fusion process, low-frequency denoising can be achieved for the denoised data based on the statistical relationship between luminance and chrominance obtained by the guided filter, while retaining the high-frequency edges and detail features of the image to be denoised. Through this fusion process, it is possible to reduce image noise while ensuring image clarity and color accuracy, further improving the overall image perception.

[0085] In some embodiments, the second color information includes the color components of the target image obtained after noise reduction. For example, the second color information may include a third color component U_N_NR and a fourth color component V_N_NR. The third color component U_N_NR and the first color component U_DS_N correspond to different image scales, but describe the same chromaticity, both used to describe the hue changes of blue and yellow in the image. The fourth color component V_N_NR and the second color component V_DS_N correspond to different image scales, but describe the same chromaticity, both used to describe the hue changes of red and cyan in the image.

[0086] It is understandable that the second brightness information and the guided filter conversion coefficient can be fused through a fusion formula to obtain the second color information. That is, the step of fusing the guided filter conversion coefficient and the second brightness information in the data to be denoised includes:

[0087] The third color component is obtained by fusing the second brightness information, the first guided filter conversion coefficient, and the second guided filter conversion coefficient based on a first fusion formula. The first fusion formula is as follows:

[0088] U_N_NR=U_A_N-1*Y_N+U_B_N-1.

[0089] The second brightness information, the third guided filter conversion coefficient, and the fourth guided filter conversion coefficient are fused based on a second fusion formula to obtain a fourth color component; the second fusion formula is as follows:

[0090] V_N_NR=V_A_N-1*Y_N+V_B_N-1.

[0091] It should be noted that the above fusion formula combines the high-frequency edge and detail features contained in the second brightness information with the low-frequency noise reduction properties of the guided filter conversion coefficients. This helps reduce low-frequency noise in the image while preserving high-frequency details and edges, ensuring that the image maintains clarity and color fidelity after noise reduction. This processing not only improves overall image quality but also achieves low-cost low-frequency noise reduction, suitable for a variety of image processing scenarios. In particular, in scenarios with limited hardware resources, such as security equipment, it can still provide significant noise reduction effects, meeting real-time and efficiency requirements.

[0092] In addition, even though the guided filter conversion coefficient corresponding to the previous frame of the current frame is involved in the process of fusing the third color component and the fourth color component, the guided filter conversion coefficient is calculated based on the downsampled image and is therefore inclusive of inter-frame motion, that is, the guided filter conversion coefficient matches the signal characteristics of the current frame, ensuring that the noise reduction effect remains stable between different frames.

[0093] S500: Outputting a target image based on the second brightness information and the second color information; low-frequency noise in the target image is lower than low-frequency noise in the image corresponding to the data to be denoised.

[0094] In some embodiments, after determining the second color information including the third color component and the fourth color component, a target image may be output based on the second brightness information Y_N, U_N_NR, and V_N_NR.

[0095] In other embodiments, if the initial format of the data to be denoised is not in YUV format, then when outputting the target image, it is necessary to inversely convert the second luminance information and the second color information so that the data format of the target image is the same as the initial format of the data to be denoised. That is, the step of outputting the target image based on the second luminance information and the second color information further includes:

[0096] When the received initial format of the data to be denoised does not conform to the target format, performing inverse conversion on the second luminance information and the second color information based on a conversion relationship between the YUV format and the initial format of the data to be denoised, so that an output format of the target image conforms to the initial format of the data to be denoised;

[0097] The output data format is the target image in the initial format.

[0098] For example, when the initial format of the data to be denoised is RGB format, when outputting the target image, the second brightness information and the second color information need to be inversely converted according to the conversion relationship between the RGB format and the YUV format to ensure that the target image is consistent with the original image in color restoration and detail presentation.

[0099] It can be understood that the low-frequency noise reduction method provided in the embodiment of the present application needs to be executed based on the data to be denoised in YUV format. Therefore, when the low-frequency noise processing module receives the data to be denoised, it is necessary to obtain the data format of the data to be denoised and determine whether format conversion needs to be performed.

[0100] In some embodiments, when the low-frequency noise reduction processing module receives the data to be denoised in a non-YUV format, it can first convert the data format of the data to be denoised into a YUV format, and then perform a series of processing on the YUV format data, such as downsampling, guided filtering, upsampling, and guided filtering application.

[0101] like Figure 3 As shown, some embodiments of the present application further provide a low-frequency noise processing device, including: a downsampling module 10, a guided filter coefficient calculation module 20, an upsampling module 30, and a guided filter application module 40.

[0102] The downsampling module 10 is configured to, upon receiving the data to be denoised that conforms to a target data format, perform downsampling on the data to be denoised to obtain downsampled data.

[0103] The guided filter coefficient calculation module 20 is used to perform guided filtering on the first color information in the downsampled data based on the first brightness information in the downsampled data to obtain guided filter conversion coefficients; the guided filter conversion coefficients are used to describe the statistical relationship between brightness and chrominance in the image corresponding to the downsampled data.

[0104] The upsampling module 30 is used to perform upsampling on the guided filter conversion coefficients to map the statistical relationship between luminance and chrominance represented by the guided filter conversion coefficients to a target resolution space; the target resolution space is a resolution space corresponding to the data to be denoised.

[0105] The guided filter application module 40 is used to fuse the guided filter conversion coefficient and the second brightness information in the data to be denoised to obtain the second color information.

[0106] The guided filtering application module 40 is further configured to output a target image based on the second brightness information and the second color information; the low-frequency noise in the target image is lower than the low-frequency noise in the image corresponding to the data to be denoised.

[0107] In some embodiments, the downsampling module 10, the guided filter coefficient calculation module 20, the upsampling module 30, and the guided filter application module 40 can all be integrated circuits, that is, low-frequency noise reduction of the noise reduction data can be achieved through hardware processing. Downsampling can achieve a low-frequency noise reduction method equivalent to expanding the domain, and can reduce the amount of data processing in subsequent tasks, thereby ensuring a certain noise reduction effect and improving the real-time performance of image processing tasks.

[0108] Furthermore, by calculating the guided filter conversion coefficients, the relationship between luminance and chrominance in the data to be denoised can be statistically analyzed, effectively reducing low-frequency noise based on this relationship. Combined with upsampling the guided filter conversion coefficients, this relationship between luminance and chrominance can be mapped to the resolution space corresponding to the data to be denoised. This allows for image denoising and restoration in high-resolution scenarios, combining the luminance information of the data to be denoised.

[0109] In addition, if Figure 4 As shown, the low-frequency noise processing device may further include a storage module 50, which may be a static random-access memory (SRAM) or a double data rate synchronous dynamic random-access memory (DDR SDRAM), for caching intermediate data to improve processing efficiency. Thus, each upper-level module in the low-frequency noise processing device can store the data in the storage module 50 after processing the data, and the lower-level modules can obtain the data from the storage module 50. For example, the downsampling module 10 can store the obtained downsampled data in the storage module 50, and the guided filter coefficient calculation module can obtain the downsampled data from the storage module 50 to perform guided filter coefficient calculation.

[0110] The collaborative work of these modules achieves efficient and cost-effective low-frequency noise reduction, ensuring improved image quality. This method is applicable to a variety of image processing scenarios, especially in scenarios with limited hardware resources, where it can significantly reduce low-frequency noise and improve visual effects.

[0111] like Figure 5 When the low-frequency noise processing device further includes a color space conversion module 60 and a color space inverse conversion module 70. Thus, when the data format of the received data to be denoised does not conform to the target data format, the data format of the data to be denoised can be converted into a YUV format by the color space conversion module 60.

[0112] In some embodiments, the color space conversion module 60 is activated only when the data format of the data to be denoised does not conform to the target data format. If the data format of the data to be denoised conforms to the target data format, the color space conversion module 60 does not need to perform the format conversion operation.

[0113] In other embodiments, if the initial format of the data to be denoised does not conform to the target data format, then in the output stage of the target image, the color space inverse conversion module 70 is required to convert the YUV format data back to the initial format to ensure that the output image is consistent with the original image format.

[0114] As can be seen from the above technical content, the present application provides a low-frequency noise processing method and device. The method performs downsampling on the data to be denoised to reduce the data processing amount of subsequent tasks and the low-frequency noise in the data to be denoised. Then, guided filtering is performed on the chromaticity information based on the luminance information in the downsampled data to establish a statistical relationship between luminance and chromaticity. Combined with the upsampling operation, the statistical relationship between luminance and chromaticity is mapped to the resolution space corresponding to the data to be denoised. In this way, the low-frequency noise in the data to be denoised can be suppressed based on the luminance information that characterizes high-frequency features and edge detail features and the statistical relationship between luminance and chromaticity, while retaining image details. It is conducive to achieving efficient noise reduction and improving image quality in scenarios with limited hardware resources.

[0115] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without expending creative work shall fall within the scope of protection of this application.

Claims

1. A low-frequency noise processing method, characterized in that: include: Upon receiving the data to be denoised that conforms to the target data format, downsampling the data to be denoised to obtain downsampled data; performing guided filtering on first color information in the downsampled data based on first luminance information in the downsampled data to obtain a guided filtering conversion coefficient; The guided filter conversion coefficient is used to describe the statistical relationship between brightness and chrominance in the image corresponding to the downsampled data; performing upsampling on the guided filter conversion coefficients to map the statistical relationship between luminance and chrominance represented by the guided filter conversion coefficients to a target resolution space; The target resolution space is the resolution space corresponding to the data to be denoised; fusing the guided filter conversion coefficient and the second brightness information in the data to be denoised to obtain second color information; outputting a target image based on the second brightness information and the second color information; The low-frequency noise in the target image is lower than the low-frequency noise in the image corresponding to the data to be denoised.

2. The method according to claim 1, characterized in that The target data format includes a YUV format; the first color information includes a first color component and a second color component, and the chromaticity described by the first color component and the second color component is different; The performing downsampling on the data to be denoised further includes: First brightness information, a first color component, and a second color component in the downsampled data are obtained.

3. The method according to claim 2, characterized in that The step of performing guided filtering on the first color information in the downsampled data based on the first brightness information in the downsampled data to obtain a guided filtering conversion coefficient comprises: performing guided filtering on the first color component based on the first luminance information to obtain a first guided filtering conversion coefficient and a second guided filtering conversion coefficient corresponding to the first color component; Guided filtering is performed on the second color component based on the first luminance information to obtain a third guided filtering conversion coefficient and a fourth guided filtering conversion coefficient corresponding to the second color component.

4. The method according to claim 3, characterized in that The second color information includes a third color component and a fourth color component; the chromaticity described by the third color component and the fourth color component is different, the chromaticity described by the third color component and the first color component is the same, and the chromaticity described by the fourth color component and the second color component is the same; The third color component has lower low-frequency noise than the first color component; The step of fusing the guided filter conversion coefficient and the second brightness information in the data to be denoised comprises: The second brightness information, the first guided filter conversion coefficient, and the second guided filter conversion coefficient are fused based on a first fusion formula to obtain a third color component; the first fusion formula is as follows: U_N_NR=U_A_N-1*Y_N+U_B_N-1; Among them, U_N_NR is the third color component; U_A_N-1 is the first guided filter conversion coefficient; U_B_N-1 is the second guided filter conversion coefficient; Y_N is the second brightness information; The second brightness information, the third guided filter conversion coefficient, and the fourth guided filter conversion coefficient are fused based on a second fusion formula to obtain a fourth color component; the second fusion formula is as follows: V_N_NR=V_A_N-1*Y_N+V_B_N-1; Among them, V_N_NR is the fourth color component; V_A_N-1 is the third guided filter conversion coefficient; V_B_N-1 is the fourth guided filter conversion coefficient.

5. The method according to claim 1, wherein The step of performing downsampling processing on the data to be denoised comprises: Setting a downsampling ratio, wherein the downsampling ratio is used to reduce the image size corresponding to the downsampled data to at least one fourth of the image size corresponding to the data to be denoised; Downsampling processing is performed on the data to be denoised according to the downsampling ratio.

6. The method according to claim 2, characterized in that Also includes: When receiving the data to be denoised that does not conform to the target data format, the data to be denoised that does not conform to the target data format is converted into data in a YUV format.

7. The method according to claim 2, characterized in that The step of outputting a target image based on the second brightness information and the second color information includes: When the received initial format of the data to be denoised does not conform to the target format, performing inverse conversion on the second luminance information and the second color information based on a conversion relationship between the YUV format and the initial format of the data to be denoised, so that an output format of the target image conforms to the initial format of the data to be denoised; The output data format is the target image in the initial format.

8. A low-frequency noise processing device, characterized in that: include: Downsampling module, guided filter coefficient calculation module, upsampling module, guided filter application module; The downsampling module is configured to, upon receiving the data to be denoised that conforms to the target data format, perform downsampling on the data to be denoised to obtain downsampled data; The guided filter coefficient calculation module is used to perform guided filtering on the first color information in the downsampled data based on the first brightness information in the downsampled data to obtain a guided filter conversion coefficient; The guided filter conversion coefficient is used to describe the statistical relationship between brightness and chrominance in the image corresponding to the downsampled data; The upsampling module is used to perform upsampling on the guided filter conversion coefficients to map the statistical relationship between luminance and chrominance represented by the guided filter conversion coefficients to a target resolution space; the target resolution space is a resolution space corresponding to the data to be denoised; The guided filter application module is used to fuse the guided filter conversion coefficient and the second brightness information in the data to be denoised to obtain second color information; The guided filtering application module is further configured to output a target image based on the second brightness information and the second color information; The low-frequency noise in the target image is lower than the low-frequency noise in the image corresponding to the data to be denoised.

9. The device according to claim 8, characterized in that Also includes a color space conversion module; The color space conversion module is used to convert the data format of the data to be denoised into the target data format when the data format of the data to be denoised does not conform to the target data format; the target data format includes a YUV format.

10. The device according to claim 9, characterized in that It also includes a color space inverse conversion module; The color space inverse conversion module is configured to, when the data format of the data to be denoised does not conform to the target data format, perform an inverse conversion on the second brightness information and the second color information based on a conversion relationship between the YUV format and the initial format of the data to be denoised, so that the output format of the target image conforms to the initial format of the data to be denoised; The color space inverse conversion module is further configured to output a target image in the initial format.

Images