De-noising parameter determination and image de-noising method, device, equipment and medium

Abstract

The present disclosure provides a de-noising parameter determination and image de-noising method, device, equipment and medium, which can acquire a plurality of de-noised images corresponding to an un-de-noised image; for any de-noised image: determining a difference value between any de-noised image and the un-de-noised image to obtain a difference value image corresponding to any de-noised image; determining a difference value discrete degree of the difference value image corresponding to any de-noised image; performing singular value decomposition on the difference value image corresponding to any de-noised image to obtain a singular value set corresponding to any de-noised image; determining a singular value discrete degree of the singular value set corresponding to any de-noised image; and determining a target parameter value group for image de-noising processing based on the difference value discrete degrees, the singular value discrete degrees corresponding to the plurality of de-noised images respectively, and parameter value groups corresponding to the plurality of de-noised images respectively. Thus, the accuracy of determining the de-noising parameter can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of image signal processing technology, and in particular to a method, apparatus, device, and medium for determining denoising parameters and for image denoising. Background Technology

[0002] Denoising is an important module in an ISP (Image Signal Processor). Generally, a denoising module has a large number of parameters. When debugging this module, after adjusting a set of parameters, the technician often analyzes the changes in image effect by visually comparing the results. This approach is inefficient, often consumes a lot of time and effort, and it is difficult to obtain the optimal parameters. Summary of the Invention

[0003] To address the aforementioned technical problems, this disclosure provides a method, apparatus, device, and medium for determining denoising parameters and for image denoising.

[0004] A first aspect of this disclosure provides a method for determining denoising parameters, including: Obtain multiple denoised images corresponding to the undated image; each denoised image is used to denoise the undated image using multiple denoising parameters under the corresponding parameter value group. For any denoised image: Determine the difference between any denoised image and an undenoised image to obtain the difference image corresponding to any denoised image; Determine the degree of dispersion of the difference image corresponding to any denoised image; Perform singular value decomposition on the difference image corresponding to any denoised image to obtain the set of singular values ​​corresponding to any denoised image; determine the degree of singular value dispersion of the set of singular values ​​corresponding to any denoised image. Based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value set corresponding to each of the denoised images, the target parameter value set for image denoising processing is determined.

[0005] A second aspect of this disclosure provides an image denoising method, comprising: Obtain the image to be denoised and the target parameter value set of the denoising parameters, wherein the target parameter value set is determined by any of the denoising parameter determination methods in the first aspect above; Denoising is performed on the image to be denoised based on the target parameter value set.

[0006] A third aspect of this disclosure provides a noise reduction parameter determination apparatus, comprising: The first acquisition unit is configured to: acquire multiple denoised images corresponding to the denoised image, and obtain the denoised image by performing denoising processing on the denoised image using multiple denoising parameters under the corresponding parameter value group for any denoised image; The first determining unit is configured to: for any denoised image: Determine the difference between any denoised image and an undenoised image to obtain the difference image corresponding to any denoised image; Determine the degree of dispersion of the difference image corresponding to any denoised image; Perform singular value decomposition on the difference image corresponding to any denoised image to obtain the set of singular values ​​corresponding to any denoised image; determine the degree of singular value dispersion of the set of singular values ​​corresponding to any denoised image. The second determining unit is configured to: determine a target parameter value set for image denoising based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value set corresponding to the multiple denoised images respectively.

[0007] A fourth aspect of this disclosure provides an image denoising apparatus, comprising: The second acquisition unit is configured to acquire the image to be denoised and a target parameter value group of denoising parameters, wherein the target parameter value group is determined by any of the denoising parameter determination devices in the first aspect described above. The denoising unit is configured to perform denoising processing on the image to be denoised based on the target parameter value set.

[0008] A fifth aspect of this disclosure provides an electronic device, comprising: Memory, used to store computer programs; A processor is configured to execute a computer program stored in a memory, wherein, when the computer program is executed, it implements the method of either the denoising parameter determination of the first aspect or the image denoising method of the second aspect of the present disclosure.

[0009] A sixth aspect of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any embodiment of the denoising parameter determination of the first aspect or the image denoising method of the second aspect described above.

[0010] A seventh aspect of this disclosure provides a computer program including computer-readable code, wherein the computer program instructions, when executed by a processor, implement the method of any embodiment of the denoising parameter determination of the first aspect or the image denoising method of the second aspect described above.

[0011] Based on the embodiments of this disclosure, by acquiring multiple denoised images corresponding to an undenoised image, a correspondence between different parameter configurations and denoising effects can be established; by determining the difference between any denoised image and an undenoised image to obtain a difference image, the impact of denoising processing on the undenoised image can be quantified more accurately; by determining the degree of dispersion of the difference image corresponding to any denoised image, the intensity of denoising can be measured, with a larger degree of dispersion indicating more effective denoising; by performing singular value decomposition on the difference image corresponding to any denoised image to obtain a set of singular values ​​and determining its singular values. The degree of dispersion can determine whether the denoising process excessively removes real signals (such as texture details and edge contours) from the image. The smaller the dispersion of singular values, the more noise is removed rather than useful information. By determining the target parameter set for image denoising based on the dispersion of differences, the dispersion of singular values, and the parameter value sets corresponding to multiple denoised images, we can screen out parameter value sets that effectively remove noise while preserving image details. This avoids the inefficiency and inconsistency of traditional human visual and subjective evaluations, and can improve the accuracy of determining denoising parameters. Furthermore, performing denoising processing on the image to be denoised based on the target parameter value set can improve the denoising effect. Attached Figure Description

[0012] Figure 1 This is an exemplary application scenario diagram to which this disclosure applies.

[0013] Figure 2 This is a flowchart illustrating a method for determining denoising parameters provided in an exemplary embodiment of this disclosure.

[0014] Figure 3 It is an illustrative grayscale Gaussian noise image.

[0015] Figure 4 It is a histogram of Gaussian noise.

[0016] Figure 5 This is a schematic diagram of the distribution of singular values.

[0017] Figure 6 This is a schematic diagram of the difference image.

[0018] Figure 7 It is aimed at Figure 6 A schematic diagram of pixel values ​​in the flat region of the difference image.

[0019] Figure 8 It is aimed at Figure 6 A schematic diagram of singular values ​​in the texture region of the difference image.

[0020] Figure 9 This is a schematic flowchart illustrating the process of determining a set of target parameter values ​​for image denoising processing, provided by an exemplary embodiment of this disclosure.

[0021] Figure 10 This is a schematic diagram of a process for obtaining multiple denoised images corresponding to an undenoised image, provided by an exemplary embodiment of this disclosure.

[0022] Figure 11 This is a schematic diagram of a process for determining N first parameter values ​​provided by an exemplary embodiment of this disclosure.

[0023] Figure 12 This is a schematic flowchart illustrating the process of determining a set of target parameter values ​​for image denoising processing, provided by an exemplary embodiment of this disclosure.

[0024] Figure 13 This is a schematic diagram of a process for determining the difference between any denoised image and an undenoised image, provided by an exemplary embodiment of this disclosure.

[0025] Figure 14 This is a schematic diagram of a process for performing singular value decomposition on the difference image corresponding to any denoised image, provided by an exemplary embodiment of this disclosure.

[0026] Figure 15 This is a schematic flowchart illustrating the process of determining the degree of difference dispersion of a difference image corresponding to any denoised image, provided by an exemplary embodiment of this disclosure.

[0027] Figure 16 This is a schematic flowchart illustrating the process of determining the degree of singular value dispersion of a set of singular values ​​corresponding to any denoised image, provided by an exemplary embodiment of this disclosure.

[0028] Figure 17 This is a schematic flowchart illustrating the process of determining a set of target parameter values ​​for image denoising processing, provided by an exemplary embodiment of this disclosure.

[0029] Figure 18 This is a schematic diagram of another process for determining a set of target parameter values ​​for image denoising processing, provided by an exemplary embodiment of this disclosure.

[0030] Figure 19 This is a schematic flowchart of an image denoising method provided in an exemplary embodiment of this disclosure.

[0031] Figure 20 This is a schematic diagram of the structure of a noise reduction parameter determination device provided in an exemplary embodiment of the present disclosure.

[0032] Figure 21 This is a schematic diagram of the structure of a noise reduction parameter determination device provided in another exemplary embodiment of this disclosure.

[0033] Figure 22 This is a schematic diagram of the structure of an image denoising apparatus provided in an exemplary embodiment of the present disclosure.

[0034] Figure 23 This is a structural diagram of an electronic device provided in an exemplary embodiment of this disclosure. Detailed Implementation

[0035] To explain this disclosure, exemplary embodiments of the disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the disclosure, and not all of them. It should be understood that the disclosure is not limited to exemplary embodiments.

[0036] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0037] Application Overview Noise reduction is a crucial module in ISPs. In developing this disclosure, the inventors discovered that ISP debugging engineers spend considerable time and effort debugging the noise reduction module, yet it remains difficult to find a truly optimal set of parameters. Furthermore, this process is highly dependent on human perception; different people perceive noise reduction differently, leading to significant subjectivity and a lack of objectivity. Some methods for automatically adjusting noise reduction module parameters employ neural networks to collect a large number of images and invite professionals to subjectively score them, ultimately finding a set of optimal parameters. However, this method is resource-intensive, difficult to implement, and its effectiveness in selecting the optimal parameters is poor.

[0038] Exemplary Overview Figure 1 This is an exemplary application scenario diagram to which this disclosure applies.

[0039] In any scenario requiring image denoising, such as autonomous driving, industrial inspection, security monitoring, and medical imaging, the denoising parameter determination method disclosed herein can be used to determine the target parameter value set for image denoising processing, and the image denoising method disclosed herein can be used to denoise the image to be denoised, thereby improving image quality.

[0040] Specifically, taking autonomous driving scenarios as an example, multiple denoising parameters can be used to denoise the undated image under corresponding parameter value groups, thereby obtaining multiple denoised images corresponding to the undated image. For example, using parameter value group 1 to denoise the undated image yields denoised image 1, using parameter value group 2 yields denoised image 2, and using parameter value group n yields denoised image n. The parameter value groups are different from each other; for example, the denoising parameters corresponding to the parameter values ​​in different parameter value groups are different, or the corresponding parameter values ​​in different parameter value groups have different values. Thus, multiple denoised images corresponding to the undated image can be obtained.

[0041] Furthermore, for any denoised image, the difference between any denoised image and an undenoised image can be determined to obtain a difference image corresponding to any denoised image. For example, determining the difference between denoised image 1 and an undenoised image yields difference image 1 corresponding to denoised image 1; determining the difference between denoised image 2 and an undenoised image yields difference image 2 corresponding to denoised image 2; and so on, determining the difference between denoised image n and an undenoised image yields difference image n corresponding to denoised image n. Then, the degree of difference dispersion of the difference image corresponding to any denoised image is determined. For example, the degree of difference dispersion 1 of difference image 1 corresponding to denoised image 1 is determined; the degree of difference dispersion 2 of difference image 2 corresponding to denoised image 2 is determined; and so on, the degree of difference dispersion n of difference image n corresponding to denoised image n is determined. Subsequently, singular value decomposition (SVD) is performed on the difference image corresponding to any denoised image to obtain the singular value set corresponding to any denoised image. For example, SVD is performed on the difference image 1 corresponding to denoised image 1 to obtain singular value set 1 corresponding to denoised image 1; SVD is performed on the difference image 2 corresponding to denoised image 2 to obtain singular value set 2 corresponding to denoised image 2, and so on. SVD is performed on the difference image n corresponding to denoised image n to obtain singular value set n corresponding to denoised image n. Then, the singular value dispersion of the singular value set corresponding to any denoised image is determined. For example, the singular value dispersion 1 of the singular value set 1 corresponding to denoised image 1 is determined; the singular value dispersion 2 of the singular value set 2 corresponding to denoised image 2 is determined; and so on, the singular value dispersion n of the singular value set n corresponding to denoised image n is determined. Finally, based on the difference dispersion, singular value dispersion, and parameter value groups corresponding to multiple denoised images, a target parameter value group for image denoising is determined. For example, based on difference dispersion 1, difference dispersion 2, ..., difference dispersion n, singular value dispersion 1, singular value dispersion 2, ..., singular value dispersion n, parameter value group 1, parameter value group 2, ..., parameter value group n, a target parameter value group for image denoising is determined.

[0042] When performing image denoising, the image to be denoised can be processed based on the target parameter value group to obtain the denoised image.

[0043] In some alternative implementations, the denoising parameters or image denoising methods can be specifically implemented using ASIC (Application-Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or CPU (Central Processing Unit).

[0044] Optionally, the vehicle can further perform visual environment perception, vehicle localization, multi-sensor fusion, path planning, and vehicle control based on the denoised images, thereby achieving autonomous driving.

[0045] Based on this, by acquiring multiple denoised images corresponding to the undated image, a correspondence between different parameter configurations and denoising effects can be established; by determining the difference between any denoised image and the undated image, a difference image can be obtained, which can more accurately quantify the impact of denoising on the undated image; by determining the degree of dispersion of the difference image corresponding to any denoised image, the intensity of denoising can be measured, with a larger degree of dispersion indicating more effective denoising; and by performing singular value decomposition on the difference image corresponding to any denoised image to obtain a set of singular values ​​and determine its singular value dispersion... The degree of singular value dispersion can determine whether the denoising process excessively removes real signals (such as texture details and edge contours) from the image. The smaller the singular value dispersion, the more noise is removed rather than useful information. By determining the target parameter value set for image denoising based on the difference dispersion, singular value dispersion, and parameter value sets corresponding to multiple denoised images, a parameter value set that effectively removes noise while preserving image details can be selected. This avoids the inefficiency and inconsistency of traditional human eye and subjective evaluation, and can improve the accuracy of determining denoising parameters. Furthermore, performing denoising processing on the image to be denoised based on the target parameter value set can improve the denoising effect.

[0046] Exemplary methods Figure 2 This is a schematic flowchart illustrating the determination of denoising parameters according to an exemplary embodiment of this disclosure. This embodiment can be applied to electronic devices such as intelligent driving vehicles, etc. Figure 2 As shown, it includes the following steps: Step 201: Obtain multiple denoised images corresponding to the denoised image. Each denoised image is used to denoise the denoised image by applying multiple denoising parameters to the denoised image under the corresponding parameter value group.

[0047] An undenoised image can be an image that needs to be denoised. In some cases, an undenoised image can be the original image saved after the denoising module of the image acquisition device (e.g., a camera) is turned off. In other cases, the camera can be pointed at a scene with rich texture details, the denoising module can be turned off, and a textured image can be saved (e.g., as a lossless compressed BMP image), named ori_noise, and used as the undenoised image. This undenoised image contains noise.

[0048] A denoised image can be obtained by applying a corresponding set of parameters to an undenoised image. Each denoised image can correspond to one set of parameters. In some cases, denoised images can be saved as BMP images. Multiple denoised images corresponding to an undenoised image can be named denoise_img0, denoise_img1, denoise_img2, ..., denoise_imgN, respectively.

[0049] Here, denoised and undenoised images are relative terms. In some cases, a denoised image can be used as an undenoised image, and the image denoising method described in this disclosure can be used to denoise it, thereby obtaining a denoised image.

[0050] Denoising parameters can be configurable parameters used for denoising in the denoising module of an image acquisition device. These parameters can be used to control the strength, threshold, and other parameters of the denoising algorithm. In some cases, the number of denoising parameters can be denoted as M, and each parameter can be divided into multiple segments, either evenly or randomly, such as 16 or 20 segments.

[0051] A parameter value group can include the values ​​of multiple denoising parameters. Each parameter value is the value of the corresponding denoising parameter. In some cases, each parameter value group can include a combination of M denoising parameter values, i.e., M parameter values. In some cases, the above parameter value groups can be used as the parameter values ​​selected in the first round (also known as coarse adjustment).

[0052] In some optional embodiments, each parameter file may contain a set of parameter value groups. Thus, multiple parameter files can be input to the camera's denoising module. The denoising module performs denoising processing on the undated image based on the multiple sets of parameter value groups contained in each parameter file, generating multiple corresponding denoised images. Furthermore, the aforementioned execution entity can obtain the multiple denoised images corresponding to the undated image.

[0053] In some alternative implementations, multiple sets of parameter values ​​can be used to denoise the entire undated image separately, thereby obtaining multiple denoised images; or, multiple sets of parameter values ​​can be used to denoise multiple different image regions of the undated image separately, thereby obtaining multiple denoised images.

[0054] Step 202: For any denoised image, perform the following steps (including steps 2021-2023): Step 2021: Determine the difference between any denoised image and any non-denoised image to obtain the difference image corresponding to any denoised image.

[0055] A difference image can be an image formed by the difference between the corresponding pixel values ​​of a denoised image and an undenoised image.

[0056] In some optional embodiments, the RGB (Red, Green, Blue) values ​​of denoise_img0 to N denoised images can be subtracted from the RGB values ​​of the undenoised orifice_noise image to obtain a difference image of the three RGB channels, denoted as a diff image.

[0057] Here, the difference image quantifies the changes introduced by the denoising process, that is, the changes in the undenoised image relative to the denoised image.

[0058] Step 2022: Determine the degree of difference dispersion of the difference image corresponding to any denoised image.

[0059] The degree of dispersion of the difference can represent the degree of dispersion of pixel values ​​in the difference image.

[0060] In some alternative implementations, the variance (stdNoise), standard deviation, or mean deviation of the difference image corresponding to each denoised image can be calculated to obtain the degree of difference dispersion of the difference image.

[0061] Step 2023: Perform singular value decomposition on the difference image corresponding to any denoised image to obtain the singular value set corresponding to any denoised image; determine the singular value dispersion of the singular value set corresponding to any denoised image.

[0062] Singular Value Decomposition (SVD) is a matrix factorization method. SVD decomposes the matrix corresponding to a difference image into three matrices: U, Σ, and V. The diagonal elements of Σ are the singular values. Here, U represents the left singular vector; Σ represents a diagonal matrix with non-zero elements only on the diagonal; and V represents the right singular vector.

[0063] A set of singular values ​​can be multiple singular values ​​obtained after singular value decomposition of a difference image. The set of singular values ​​can characterize the structural properties of the difference image. Here, the singular values ​​in the set of singular values ​​can also be called eigenvalues.

[0064] The degree of dispersion of singular values ​​can represent the degree of dispersion of values ​​in a set of singular values.

[0065] In some alternative implementations, the degree of singularity dispersion of the singular value set can be obtained by calculating the standard deviation, variance, or mean deviation of the singular value set corresponding to each denoised image.

[0066] Step 203: Based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value group corresponding to the multiple denoised images, determine the target parameter value group for image denoising processing.

[0067] The target parameter value set can be a parameter value set determined from the parameter value sets corresponding to multiple denoised images, based on the degree of difference dispersion and the degree of singular value dispersion corresponding to multiple denoised images respectively.

[0068] In some alternative implementations, a weighted sum of the difference dispersion and the singular value dispersion corresponding to each parameter value group can be calculated, and the target parameter value group can be determined from the parameter value groups corresponding to multiple denoised images based on this result.

[0069] Based on the embodiments of this disclosure, by acquiring multiple denoised images corresponding to an undenoised image, a correspondence between different parameter configurations and denoising effects can be established; by determining the difference between any denoised image and an undenoised image to obtain a difference image, the impact of denoising processing on the undenoised image can be quantified more accurately; by determining the degree of dispersion of the difference image corresponding to any denoised image, the intensity of denoising can be measured, with a larger degree of dispersion indicating more effective denoising; by performing singular value decomposition on the difference image corresponding to any denoised image to obtain a set of singular values ​​and determining its singular values. The degree of dispersion can determine whether the denoising process excessively removes real signals (such as texture details and edge contours) from the image. The smaller the dispersion of singular values, the more noise is removed rather than useful information. By determining the target parameter set for image denoising based on the dispersion of differences, the dispersion of singular values, and the parameter value sets corresponding to multiple denoised images, we can screen out parameter value sets that effectively remove noise while preserving image details. This avoids the inefficiency and inconsistency of traditional human visual and subjective evaluations, and can improve the accuracy of determining denoising parameters. Furthermore, performing denoising processing on the image to be denoised based on the target parameter value set can improve the denoising effect.

[0070] Here, refer to Figures 3-8 The noise in the image is analyzed. for Figure 3 The grayscale Gaussian noise image shown can be analyzed using histograms, see [link to relevant documentation]. Figure 4 As shown. Figure 4The gray-filled area and the smooth curve above conform to a Gaussian distribution, indicating that the noise histogram almost perfectly fits the Gaussian distribution.

[0071] exist Figure 3 Based on the grayscale Gaussian noise image shown, Gaussian white noise with a standard deviation of 8 (this is just an example; other values ​​are also possible) is added. The calculated SVD eigenvalues ​​are basically on a horizontal line, as shown below. Figure 5 As shown.

[0072] If we calculate the standard deviation of the eigenvalues ​​(also known as singular values, SVD eigenvalues), we will find that this value is very small. From this, we can conclude that the smaller the variance of the eigenvalues, the higher the probability that Gaussian white noise exists in the image.

[0073] The difference between a denoised and an undenoised image is generally considered to be noise. However, many denoising algorithms are not perfect, resulting in the removal of not only noise but also texture details and edge contours during denoising. For example... Figure 6 As shown. Figure 6 It is the difference image between the denoised image and the undated image, where the green box is the approximately flat area and the red box is the texture area.

[0074] Figure 6 The green box represents a nearly flat area, where pixel values ​​(e.g., grayscale, RGB values) are represented using a 10-bit width. Figure 7 As shown. Figure 7 The standard deviation of the pixel values ​​is 193.5. Its eigenvalues ​​(singular values) are [948.49, 689.54, 588.68, 580.07, 425.7, 275.77, 213.83, 4.33], and the standard deviation of the eigenvalues ​​is 279.6.

[0075] Figure 6 The red box in the middle represents the texture area, where specific pixel values ​​(such as grayscale values ​​and RGB values) are represented using a 10-bit width. Figure 8 As shown. Figure 8 The standard deviation of the pixel values ​​is 615.4. Its eigenvalues ​​(singular values) are [3179.21, 2633.79, 1912.73, 1198.23, 1161.98, 691.52, 463.97, 164.04], and the standard deviation of the eigenvalues ​​is 996.8.

[0076] It is evident that the standard deviation of eigenvalues ​​in the textured region is much greater than that in the flat region.

[0077] Therefore, the larger the pixel value of the difference image itself, the better, indicating that the denoising is more effective. If the difference image is 0, it means that there is no denoising at all. However, at the same time, the smaller the standard deviation of the feature value (singular value) of the difference image, the better, indicating that only noise is removed, rather than removing the texture details and edge contours.

[0078] Based on the above analysis, in some optional embodiments, such as Figure 9 As shown above, in the above Figure 2 Based on the illustrated embodiment, step 203 may include the following steps: Step 2031: For any one of the denoised images from the plurality of denoised images, perform the following steps (including steps 311-312): Step 311: Based on the degree of difference dispersion and the degree of singular value dispersion of any denoised image, obtain the evaluation value of the denoising parameters corresponding to any denoised image.

[0079] The denoising parameter evaluation value can be used to quantitatively evaluate the denoising effect.

[0080] In some alternative implementations, the denoising parameter evaluation value DeNoiseK = the dispersion of the difference images / (the dispersion of the corresponding singular values ​​+ alpha). A larger DeNoiseK value indicates that the denoising is more satisfactory. alpha is a constant greater than 0 to prevent the denominator from being zero.

[0081] In some alternative implementations, the denoising parameter evaluation value DeNoiseK = the degree of dispersion of the difference images / the degree of dispersion of the corresponding singular values.

[0082] In some alternative implementations, a weighted summation of the difference dispersion and the singular value dispersion of the denoised image can be performed, and the result can be used as the evaluation value of the denoising parameters corresponding to the denoised image.

[0083] Step 312: Based on the denoising parameter evaluation values ​​corresponding to the multiple denoised images and the parameter value groups corresponding to the multiple denoised images, determine the target parameter value group for image denoising processing.

[0084] In some alternative implementations, the parameter value group corresponding to the largest denoising parameter evaluation value among the parameter value groups corresponding to multiple denoised images can be determined as the target parameter value group for image denoising processing.

[0085] Based on the embodiments of this disclosure, the denoising parameter evaluation value determined by obtaining the denoising parameter evaluation value corresponding to any denoised image according to the difference dispersion and singular value dispersion of any denoised image can comprehensively reflect the intensity of denoising (corresponding to the difference dispersion) and the degree of preservation of the real signal (corresponding to the singular value dispersion), avoiding the one-sidedness of single index evaluation; by determining the target parameter value group for image denoising processing based on the denoising parameter evaluation values ​​corresponding to multiple denoised images and the parameter value groups corresponding to multiple denoised images, the parameter value group with better denoising effect can be selected, thereby improving the efficiency and accuracy of parameter value group determination.

[0086] In some alternative embodiments, in the above... Figure 2 Based on the illustrated embodiment, step 201 may include the following steps: Step 2011: For any of the M denoising parameters: Determine N first parameter values ​​from the range of values ​​of any denoising parameter, where M and N are both positive integers.

[0087] M represents the total number of noise reduction parameters that need to be adjusted.

[0088] The value range indicates the optional range of values ​​for the denoising parameter. For example, the value range for the denoising parameter can be 0-255.

[0089] N represents the number of parameter values ​​selected from the range of values.

[0090] The first parameter value can be a parameter value determined from the range of values.

[0091] In some optional implementations, when N is 16 and its value range is 0-255, sampling can start from 8 and proceed at 16 intervals, i.e., the N first parameter values ​​are 8, 24, 40, 56, 72, 88, 104, 120, 136, 152, 168, 184, 200, 216, 232, and 248.

[0092] In some alternative implementations, the range of values ​​for the denoising parameters can be divided into N sub-ranges, and then the median value can be selected from each sub-range, or Latin Hypercube Sampling (LHS) can be used to obtain N first parameter values.

[0093] In some optional implementation manners, when M is greater than or equal to 16, the Latin hypercube sampling value can be configured as a first value (such as 5000, 6000), and when 2 < M < 16, the Latin hypercube sampling value can be configured as a second value (such as 1000). Here, the first value is greater than the second value. Thus, when the parameter dimension is relatively high, the Latin hypercube sampling method can effectively reduce the sampling quantity, solve the problem of combinatorial explosion caused by a high parameter dimension, improve the search efficiency, and ensure the coverage of the parameter space at the same time.

[0094] Step 2012: Based on N first parameter values corresponding to any one of the M denoising parameters, determine a set of first parameter value groups. In each first parameter value group in the set of first parameter value groups, the first parameter values correspond one by one to the denoising parameters among the M denoising parameters, and in any two first parameter value groups in the set of first parameter value groups, at least one pair of different first parameter values of the same denoising parameter is included.

[0095] The set of first parameter value groups may include multiple first parameter value groups. Each set of first parameter value groups may correspond to one denoising parameter. Each first parameter value group may be determined based on M first parameter values (corresponding to one denoising parameter respectively). For example, each first parameter value group may include M first parameter values.

[0096] In some optional implementation manners, each of the M denoising parameters corresponds to N first parameter values. Thus, the number of first parameter values corresponding to the M denoising parameters is M × N. Based on this, one first parameter value can be respectively selected from the N first parameter values corresponding to each of the M denoising parameters, so as to obtain the set of first parameter value groups.

[0097] Step 2013: For any one first parameter value group in the set of first parameter value groups, perform denoising processing on the non-denoised image by using the any one first parameter value group to obtain a first denoised image.

[0098] The first denoised image may be an image obtained by performing denoising processing on the non-denoised image by using the first parameter value group.

[0099] In some optional implementation manners, if the set of first parameter value groups includes first parameter value group 1 and first parameter value group 2, then, the non-denoised image may be denoised by using first parameter value group 1 to obtain a first denoised image 1; the non-denoised image may be denoised by using first parameter value group 2 to obtain a first denoised image 2.

[0100] Based on the embodiments of this disclosure, by determining N first parameter values ​​from the value range of any of the M denoising parameters, preliminary sampling can be performed in the parameter space to obtain M×N first parameter values. By determining a set of first parameter value groups based on the N first parameter values ​​corresponding to any of the M denoising parameters, discrete sample points covering the parameter space can be constructed. By applying any first parameter value group from the set of first parameter value groups to denoise the undenoised image to obtain a first denoised image, the denoising effect corresponding to different first parameter value groups can be obtained. Furthermore, by comparing the denoising effects, the target parameter value group with the best denoising effect can be determined more accurately.

[0101] In some alternative embodiments, such as Figure 11 As shown above, in the above Figure 10 Based on the illustrated embodiment, step 2011 may include the following steps: Step 20111: Based on the value range of any denoising parameter and the preset selection quantity, determine the sampling interval corresponding to any denoising parameter.

[0102] The preset selection quantity can be a predetermined number of first parameter values ​​selected for each noise reduction parameter. For example, the preset selection quantity could be 16, 18, etc.

[0103] The sampling interval represents the step size between two adjacent sampling points (i.e., the first parameter value). The sampling interval determines the density of sampling points.

[0104] In some alternative implementations, the difference between the upper and lower limits of the value range can be calculated first, and then the sampling interval can be determined based on this difference and a preset selection quantity. For example, if the value range of a denoising parameter is 0-255 and the preset selection quantity is 16, then 255 ÷ 16 ≈ 16 can be used to determine the sampling interval.

[0105] Step 20112: Based on the sampling interval, determine N first parameter values ​​from the range of any denoising parameter.

[0106] In some optional implementations, N first parameter values ​​can be determined from the range of any denoising parameter based on a preset strategy (such as random sampling or Latin hypercube sampling) and the sampling interval. For example, if the range of a denoising parameter is 0-255 and the preset selection quantity is 16, sampling can start from 8 and proceed at 16 intervals, resulting in N first parameter values ​​of 8, 24, 40, 56, 72, 88, 104, 120, 136, 152, 168, 184, 200, 216, 232, and 248.

[0107] Based on the embodiments of this disclosure, by determining the sampling interval corresponding to any denoising parameter based on its value range and a preset selection quantity, it is possible to ensure a uniform distribution of sampling points (first parameter values). By determining N first parameter values ​​from the value range of any denoising parameter based on the sampling interval, the entire value range of the denoising parameter can be covered, avoiding the subjectivity and incompleteness of manual selection. By setting the sampling interval, the computational load can be controlled while ensuring coverage, providing balanced first parameter values, and avoiding the omission of potential optimal parameter value sets.

[0108] In some alternative embodiments, such as Figure 12 As shown above, in the above Figure 10 Based on the illustrated embodiment, step 203 may include the following steps: Step 2032: Based on the degree of difference dispersion and the degree of singular value dispersion corresponding to the multiple denoised images, determine the initial parameter value group from the parameter value group corresponding to the multiple denoised images.

[0109] The initial parameter value set can be a preliminary set of parameter values ​​determined from the parameter value sets corresponding to multiple denoised images, based on the degree of dispersion of the difference and the degree of dispersion of the singular values. In some cases, the initial parameter value set can also be referred to as the optimal parameter combination selected in the coarse adjustment stage.

[0110] In some alternative implementations, the initial parameter value set can be determined by selecting the parameter value set with the largest denoising parameter evaluation value from the parameter value sets corresponding to multiple denoised images. Alternatively, the initial parameter value set can be determined from the parameter value sets corresponding to multiple denoised images by performing a weighted summation of the difference dispersion and singular value dispersion of the parameters.

[0111] Step 2033: For any of the preset M denoising parameters, perform the following steps (including steps 331-332): Step 331: Determine the initial parameter value corresponding to any denoising parameter from the initial parameter value group.

[0112] The initial parameter value can be any value from the initial parameter value group. Each initial parameter value corresponds to a denoising parameter.

[0113] Here, since each parameter value in the initial parameter value group corresponds to a denoising parameter, the parameter value corresponding to any denoising parameter, i.e., the initial parameter value, can be determined from the initial parameter value group.

[0114] Step 332: From the range of any denoising parameter, determine X second parameter values ​​within the neighborhood of the initial parameter value, where X is a positive integer.

[0115] The neighborhood range can be a local region near the initial parameter value within the range of the corresponding denoising parameter. For example, if the denoising parameter ranges from 0 to 255 and the initial parameter value is 40, then the neighborhood range of the initial parameter value can be defined as 30-50. The neighborhood range limits the scope of the fine-grained search.

[0116] The second parameter value can be a parameter value selected from the above-mentioned range.

[0117] In some alternative implementations, X second parameter values ​​can be selected from the neighborhood range according to a preset strategy (e.g., random or Latin hypercube sampling). For example, if the neighborhood range is 25-55 and X is 31, then the X second parameter values ​​could be 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, and 55.

[0118] Step 2034: Based on the X second parameter values ​​corresponding to any one of the M denoising parameters, determine the set of second parameter value groups. The second parameter value in each second parameter value group in the set of second parameter value groups corresponds one-to-one with the denoising parameter in the M denoising parameters. Any two second parameter value groups in the set of second parameter value groups contain at least one pair of different second parameter values ​​of the same denoising parameter.

[0119] The set of second parameter value groups can include multiple second parameter value groups. Each set of second parameter value groups can correspond to one denoising parameter. Each second parameter value group can be determined based on X second parameter values ​​(each corresponding to a denoising parameter), for example, each second parameter value group can contain X second parameter values.

[0120] In some optional implementations, each of the M denoising parameters corresponds to X second parameter values. Thus, the number of second parameter values ​​corresponding to the M denoising parameters is M×X. Based on this, one second parameter value can be selected from each of the X second parameter values ​​corresponding to the M denoising parameters to obtain a set of second parameter value groups.

[0121] Step 2035: For any second parameter value group in the set of second parameter value groups, perform denoising processing on the undenoised image using any second parameter value group to obtain the second denoised image.

[0122] The second denoised image can be an image obtained by denoising the undenoised image using the second set of parameter values.

[0123] In some alternative implementations, if the set of second parameter value groups includes second parameter value group 1 and second parameter value group 2, then the second parameter value group 1 can be used to denoise the undenoised image to obtain the second denoised image 1; the second parameter value group 2 can be used to denoise the undenoised image to obtain the second denoised image 2.

[0124] Step 2036: Based on the difference dispersion, singular value dispersion and parameter value group corresponding to the multiple second denoised images respectively, determine the target parameter value group for image denoising processing.

[0125] In some alternative implementations, a weighted sum of the difference dispersion and singular value dispersion corresponding to multiple second denoised images can be calculated, and the target parameter value set can be determined from the parameter value sets corresponding to the multiple second denoised images based on the result.

[0126] In some optional implementations, the parameter value group corresponding to the largest denoising parameter evaluation value among the parameter value groups corresponding to multiple second denoised images can be determined as the target parameter value group for image denoising processing. The denoising parameter evaluation value is determined based on the degree of difference dispersion and the degree of singular value dispersion.

[0127] Based on the embodiments of this disclosure, by determining an initial parameter value set from the parameter value sets corresponding to multiple denoised images based on the degree of difference dispersion and singular value dispersion of the images, a parameter value set with better performance can be selected in the coarse adjustment stage as the starting point for fine adjustment. By determining an initial parameter value corresponding to any denoising parameter from the initial parameter value set for any of the preset M denoising parameters, and determining X second parameter values ​​within the neighborhood of the initial parameter value from the value range of any denoising parameter, a local fine search can be performed near the initial parameter value, improving the accuracy of parameter value optimization. By using M denoising parameters... The first step involves determining a set of X second parameter values ​​corresponding to any denoising parameter in the noise parameters. This set of second parameter values ​​helps identify parameter values ​​near the initial parameter values. Then, by applying any set of second parameter values ​​from this set to the undenoised image, a second denoised image is obtained, providing data on the denoising effect during the fine-tuning stage. Finally, by determining the target parameter value set for image denoising based on the difference dispersion, singular value dispersion, and the parameter value sets corresponding to multiple second denoised images, the optimal parameter value set (the target parameter value set) can be found locally. This two-stage search method ensures both search efficiency and parameter optimization accuracy, avoiding the computational burden of exhaustive searching in a high-dimensional parameter space.

[0128] In some alternative embodiments, such as Figure 13 As shown above, in the above Figure 2 Based on the illustrated embodiment, step 202 may include the following steps: Step 2021: Determine the difference between the pixel value of any channel of any denoised image and the pixel value of the corresponding channel of any undisturbed image, and obtain the difference image corresponding to any channel of any denoised image.

[0129] Channels can represent the color components of a denoised image. As an example, the channels mentioned above can be one of the following: R channel, G channel, or B channel.

[0130] A pixel value can be the numerical value of each pixel in a denoised image. Pixel values ​​represent the intensity or color component value of an image pixel.

[0131] In some optional implementations, if the denoised image includes denoised image 1 and denoised image 2, then for denoised image 1: the difference between the pixel values ​​of the R channel of denoised image 1 and the pixel values ​​of the R channel of the undated image can be determined to obtain the difference image corresponding to the R channel of denoised image 1; the difference between the pixel values ​​of the G channel of denoised image 1 and the pixel values ​​of the G channel of the undated image can be determined to obtain the difference image corresponding to the G channel of denoised image 1; and the difference between the pixel values ​​of the B channel of denoised image 1 and the pixel values ​​of the B channel of the undated image can be determined to obtain the difference image corresponding to the B channel of denoised image 1. For the denoised image 2: the difference between the pixel values ​​of the R channel of the denoised image 2 and the pixel values ​​of the R channel of the undated image 2 can be determined to obtain the difference image corresponding to the R channel of the denoised image 2; the difference between the pixel values ​​of the G channel of the denoised image 2 and the pixel values ​​of the G channel of the undated image 2 can be determined to obtain the difference image corresponding to the G channel of the denoised image 2; the difference between the pixel values ​​of the B channel of the denoised image 2 and the pixel values ​​of the B channel of the undated image 2 can be determined to obtain the difference image corresponding to the B channel of the denoised image 2.

[0132] Based on the embodiments of this disclosure, by determining the difference between the pixel value of any channel of any denoised image and the pixel value of the corresponding channel of any undisturbed image, a difference image corresponding to any channel of any denoised image is obtained. Noise analysis and evaluation can be performed on each channel separately, avoiding mutual interference between different color channels. This channel separation processing method can more accurately capture the noise characteristics of different color components. Through channel separation processing, more accurate noise analysis and parameter adjustment can be performed on the characteristics of different color components, thereby improving the denoising effect.

[0133] In some alternative embodiments, such as Figure 14 As shown above, in the above Figure 13 Based on the illustrated embodiment, step 2021 may include the following steps: Step 211: For the difference image of any channel of any denoised image, perform the following steps (including steps 2111-2112): Step 2111: Divide the difference image of any channel to obtain multiple image blocks of any channel.

[0134] An image patch can be an image region obtained by dividing a difference image. Dividing an image into patches facilitates the analysis of local features.

[0135] In some alternative implementations, the difference image of the R channel can be divided to obtain multiple image blocks of the R channel; the difference image of the G channel can be divided to obtain multiple image blocks of the G channel; and the difference image of the B channel can be divided to obtain multiple image blocks of the B channel.

[0136] In some alternative implementations, the difference image of any channel can be divided into multiple image blocks for any channel by setting blocks of a fixed size (such as 8×8, 16×8, 16×16 or 32×32 pixels).

[0137] Step 2112: Perform singular value decomposition on the pixel values ​​of any image block in any channel to obtain the singular value set of any image block in any channel of any denoised image.

[0138] In some optional implementations, singular value decomposition (SVD) can be performed on the pixel values ​​of image block 1 in the R channel of denoised image 1 to obtain the singular value set of image block 1 in the R channel of denoised image 1; singular value decomposition can be performed on the pixel values ​​of image block 2 in the R channel of denoised image 1 to obtain the singular value set of image block 2 in the R channel of denoised image 1; ...; singular value decomposition can be performed on the pixel values ​​of image block n in the R channel of denoised image 1 to obtain the singular value set of image block n in the R channel of denoised image 1; singular value decomposition can be performed on the pixel values ​​of image block 1 in the G channel of denoised image 1 to obtain the singular value set of image block 1 in the G channel of denoised image 1; ...; singular value decomposition can be performed on the pixel values ​​of image block 2 in the G channel of denoised image 1 to obtain the singular value set of image block n in the R channel of denoised image 1; singular value decomposition can be performed on the pixel values ​​of image block 2 in the G channel of denoised image 1 to obtain the singular value set of image block 1 in the G channel of denoised image 1; ... Perform singular value decomposition (SVD) to obtain the singular value set of image block 2 in the G channel of the denoised image 1; ...; Perform SVD on the pixel values ​​of image block n in the G channel of the denoised image 1 to obtain the singular value set of image block n in the G channel of the denoised image 1; Perform SVD on the pixel values ​​of image block 1 in the B channel of the denoised image 1 to obtain the singular value set of image block 1 in the B channel of the denoised image 1; Perform SVD on the pixel values ​​of image block 2 in the B channel of the denoised image 1 to obtain the singular value set of image block 2 in the B channel of the denoised image 1; ...; Perform SVD on the pixel values ​​of image block n in the B channel of the denoised image 1 to obtain the singular value set of image block n in the B channel of the denoised image 1. The processing method for denoised image 2-n can be referred to the above description.

[0139] Based on the embodiments of this disclosure, by dividing the difference image of any channel of any denoised image into multiple image blocks of any channel, global difference image analysis can be transformed into local image block analysis, and the noise characteristic differences of different image blocks can be obtained. By performing singular value decomposition on the pixel values ​​of any image block of any channel, the singular value set of any image block of any channel of any denoised image can be obtained. The structural features of local image blocks can be analyzed to determine whether the image block contains random noise or meaningful image structure, thereby determining more accurate denoising parameters for different regional characteristics.

[0140] In some alternative embodiments, such as Figure 15 As shown above, in the above Figure 14 Based on the illustrated embodiment, step 202 may include the following steps: Step 2022: For any image block of any channel in any denoised image: determine the degree of dispersion of pixel values ​​of any image block of any channel, and obtain the degree of difference dispersion of any image block of any channel in any denoised image.

[0141] Discreteness refers to how far apart the pixel values ​​of a pixel are from the mean. For example, discreteness can be determined based on the standard deviation and / or variance.

[0142] In some optional implementations, the dispersion of pixel values ​​in image block 1 of the R channel of the denoised image 1 can be determined to obtain the difference dispersion of image block 1 in the R channel of the denoised image 1; the dispersion of pixel values ​​in image block 2 of the R channel of the denoised image 1 can be determined to obtain the difference dispersion of image block 2 in the R channel of the denoised image 1, ..., the dispersion of pixel values ​​in image block n of the R channel of the denoised image 1 can be determined to obtain the difference dispersion of image block n in the R channel of the denoised image 1; the dispersion of pixel values ​​in image block 1 of the G channel of the denoised image 1 can be determined to obtain the difference dispersion of image block 1 in the G channel of the denoised image 1; the dispersion of pixel values ​​in image block 2 of the G channel of the denoised image 1 can be determined to obtain the difference dispersion of image block 1 in the G channel of the denoised image 1; the dispersion of pixel values ​​in image block 2 of the G channel of the denoised image 1 can be determined to obtain the difference dispersion of image block 1 in the G channel of the denoised image 1. The dispersion of pixel values ​​in image block n of the G channel of denoised image 1 is determined, and the dispersion of the difference values ​​in image block n of the G channel of denoised image 1 is obtained; the dispersion of pixel values ​​in image block 1 of the B channel of denoised image 1 is determined, and the dispersion of the difference values ​​in image block n of the B channel of denoised image 1 is obtained; the dispersion of pixel values ​​in image block 2 of the B channel of denoised image 1 is determined, and the dispersion of the difference values ​​in image block 2 of the B channel of denoised image 1 is obtained; the dispersion of pixel values ​​in image block n of the B channel of denoised image 1 is determined, and the dispersion of the difference values ​​in image block n of the B channel of denoised image 1 is obtained. The processing method for denoised image 2-n can be referred to the above description.

[0143] Based on the embodiments of this disclosure, by determining the degree of dispersion of pixel values ​​of any image block in any channel of any denoised image, the degree of difference dispersion of any image block in any channel of any denoised image is obtained, which can quantitatively evaluate the denoising intensity of a local region (image block). By calculating the local dispersion, the differences in noise characteristics of different regions of the difference image can be captured, thereby distinguishing the denoising effect of different regions. For example, flat regions and textured regions have different sensitivities to denoising parameters, thus providing a more fine-grained evaluation index for global optimization.

[0144] In some alternative embodiments, such as Figure 16 As shown above, in the above Figure 14 Based on the illustrated embodiment, step 202 may include the following steps: Step 2023, for any image block of any channel of any denoised image: based on the set of singular values ​​of any image block of any channel of any denoised image, determine the degree of singular value dispersion of the set of singular values ​​of any image block of any channel of any denoised image.

[0145] In some optional implementations, the degree of singular value dispersion of the singular value set of image block 1 in the R channel of denoised image 1 can be determined based on the singular value set of image block 1 in the R channel of denoised image 1; the degree of singular value dispersion of the singular value set of image block 2 in the R channel of denoised image 1 can be determined based on the singular value set of image block 2 in the R channel of denoised image 1, ..., the degree of singular value dispersion of the singular value set of image block n in the R channel of denoised image 1 can be determined based on the singular value set of image block 1 in the G channel of denoised image 1; the degree of singular value dispersion of the singular value set of image block 1 in the G channel of denoised image 1 can be determined based on the singular value set of image block 2 in the G channel of denoised image 1. The process involves determining the degree of singular value dispersion of the singular value set of image patch 2 in the G channel of denoised image 1, and so on. This process is repeated for each image patch, and the processing of denoised image 2-n follows the same steps.

[0146] Based on the embodiments of this disclosure, by taking any image patch of any channel in any denoised image, and determining the singular value dispersion of the singular value set of any image patch of any channel in any denoised image, it is possible to analyze whether a local region (image patch) contains structured information. The smaller the singular value dispersion, the more likely the region is mainly random noise; the larger the singular value dispersion, the more likely the region contains more meaningful image structures, such as textures or edges. By calculating the singular value dispersion, noise and real signals can be objectively distinguished, thereby determining the target parameter value set with better denoising effect.

[0147] In some alternative embodiments, such as Figure 17 As shown above, in the above Figure 16Based on the illustrated embodiment, step 2023 may include the following steps: Step 20231: Based on the difference dispersion, singular value dispersion of each channel of each image block of multiple denoised images, and the parameter value group corresponding to each of the multiple denoised images, determine the target parameter value group for image denoising processing.

[0148] In some optional embodiments, the denoising parameter evaluation value corresponding to the denoised image 1 can be determined based on the difference dispersion and singular value dispersion corresponding to the R channel of image block 1 of the denoised image 1, the difference dispersion and singular value dispersion corresponding to the G channel of image block 1 of the denoised image 1, the difference dispersion and singular value dispersion corresponding to the B channel of image block 1 of the denoised image 1, the difference dispersion and singular value dispersion corresponding to the R channel of image block 2 of the denoised image 1, the difference dispersion and singular value dispersion corresponding to the G channel of image block 2 of the denoised image 1, the difference dispersion and singular value dispersion corresponding to the B channel of image block 2 of the denoised image 1, ..., the difference dispersion and singular value dispersion corresponding to the R channel of image block n of the denoised image 1, the difference dispersion and singular value dispersion corresponding to the G channel of image block n of the denoised image 1, and the difference dispersion and singular value dispersion corresponding to the B channel of image block n of the denoised image 1. The processing method for denoised images 2-n can be referred to the above description. This yields the denoising parameter evaluation values ​​corresponding to denoised images 1-n respectively. Then, from the denoising parameter evaluation values ​​corresponding to denoised images 1-n respectively, the denoising parameter evaluation value with the largest value is determined, and the parameter value group used in the denoised image corresponding to the denoising parameter evaluation value with the largest value is taken as the target parameter value group.

[0149] In some optional embodiments, the denoising parameter evaluation value corresponding to each image block can be determined based on the denoising parameter evaluation values ​​calculated for the three RGB channels of that image block. For example, a weighted average or weighted summation can be performed on the denoising parameter evaluation values ​​calculated for the three RGB channels of that image block (e.g., the weights for the RGB channels are 0.3, 0.6, and 0.1 respectively, to conform to the characteristics of human visual perception) to obtain the denoising parameter evaluation value corresponding to that image block. Further, the denoising parameter evaluation values ​​corresponding to each image block of the denoised image can be accumulated to obtain the denoising parameter evaluation value of the denoised image.

[0150] Based on the embodiments of this disclosure, by considering the degree of difference dispersion, the degree of singular value dispersion corresponding to each channel of each image block of multiple denoised images, and the parameter value groups corresponding to each of the multiple denoised images, a target parameter value group for image denoising processing is determined. This can comprehensively consider the denoising effect of local regions (image blocks) and color channels of the denoised image, avoid the global performance degradation caused by local optimization, thereby balancing the denoising requirements of different regions and channels, finding the overall optimal target parameter value group, and ensuring that the target parameter value group can achieve good denoising effect on various image contents and color components.

[0151] In some optional embodiments, the number of undisturbed images is a preset number, wherein different undisturbed images correspond to different scenes.

[0152] The preset quantity can be a pre-defined positive integer, for example, the preset quantity can be 5.

[0153] A scene can represent the type or environment of an image. In some cases, the scene corresponding to an undenoised image can be a scene with rich texture details, such as colorful withered leaves, blue sky, white clouds, and green trees.

[0154] Based on this, such as Figure 18 As shown above, in the above Figure 2 Based on the illustrated embodiment, step 203 may include the following steps: Step 2037: For any denoised image among multiple denoised images, based on the degree of difference dispersion, the degree of singular value dispersion of the multiple denoised images corresponding to any denoised image, and the parameter value group corresponding to the multiple denoised images corresponding to any denoised image, determine the parameter value group for image denoising processing corresponding to any denoised image.

[0155] The parameter set used for image denoising can be a set of parameter values ​​determined for a single scene.

[0156] In some alternative implementations, parameter value sets can be determined independently for each scenario.

[0157] Step 2038: Determine the target parameter value set based on the parameter value sets for image denoising processing corresponding to the multiple undated images.

[0158] The target parameter value set can be a set of parameter values ​​obtained based on the optimization results of multiple scenarios.

[0159] In some alternative implementations, the median of the corresponding parameter values ​​from the parameter value sets obtained from each scenario (i.e., the parameter value sets for image denoising corresponding to multiple undenoised images) can be used as the target parameter value set.

[0160] Based on the embodiments of this disclosure, by setting the number of undenoised images to a preset number, where different undenoised images correspond to different scenarios, diverse image content and noise characteristics can be covered, improving the robustness of determining the target parameter value set; by determining the parameter value set for image denoising for any undenoised image among multiple undenoised images based on the difference dispersion, singular value dispersion, and parameter value set corresponding to the multiple denoised images corresponding to any undenoised image, the optimal parameter value set for each specific scenario can be found; by determining the target parameter value set based on the parameter value set for image denoising corresponding to multiple undenoised images, the optimization results of multiple scenarios can be combined to obtain a more adaptable target parameter value set.

[0161] Any of the denoising parameter determination methods provided in this disclosure can be executed by any suitable device with data processing capabilities, including but not limited to: terminal devices and servers. Alternatively, any of the denoising parameter determination methods provided in this disclosure can be executed by a processor, such as by a processor executing any of the denoising parameter determination methods mentioned in this disclosure by calling corresponding instructions stored in memory. Further details will not be elaborated below.

[0162] Figure 19 This is a schematic flowchart illustrating an exemplary embodiment of the present disclosure of image denoising. This embodiment can be applied to electronic devices such as intelligent driving vehicles, etc. Figure 19 As shown, it includes the following steps: Step 401: Obtain the image to be denoised and the target parameter value set of the denoising parameters.

[0163] The target parameter value set is determined using any of the above-mentioned methods for determining denoising parameters.

[0164] Step 402: Denoise the image to be denoised based on the target parameter value group.

[0165] In some alternative implementations, the target parameter value set can be configured to the denoising module of the image processing device to drive the denoising algorithm to denoise the image to be denoised.

[0166] Based on the embodiments of this disclosure, by acquiring multiple denoised images corresponding to an undenoised image, a correspondence between different parameter configurations and denoising effects can be established; by determining the difference between any denoised image and an undenoised image to obtain a difference image, the impact of denoising processing on the undenoised image can be quantified more accurately; by determining the degree of dispersion of the difference image corresponding to any denoised image, the intensity of denoising can be measured, with a larger degree of dispersion indicating more effective denoising; by performing singular value decomposition on the difference image corresponding to any denoised image to obtain a set of singular values ​​and determining its singular values. The degree of dispersion can determine whether the denoising process excessively removes real signals (such as texture details and edge contours) from the image. The smaller the dispersion of singular values, the more noise is removed rather than useful information. By determining the target parameter set for image denoising based on the dispersion of differences, the dispersion of singular values, and the parameter value sets corresponding to multiple denoised images, we can screen out parameter value sets that effectively remove noise while preserving image details. This avoids the inefficiency and inconsistency of traditional human visual and subjective evaluations, and can improve the accuracy of determining denoising parameters. Furthermore, performing denoising processing on the image to be denoised based on the target parameter value set can improve the denoising effect.

[0167] Any of the image denoising methods provided in this disclosure can be executed by any suitable device with data processing capabilities, including but not limited to: terminal devices and servers. Alternatively, any of the image denoising methods provided in this disclosure can be executed by a processor, such as by a processor executing any of the image denoising methods mentioned in this disclosure by calling corresponding instructions stored in memory. Further details will not be elaborated below.

[0168] The implementation of any image denoising method provided in the embodiments of this disclosure can refer to any method for determining denoising parameters provided in the embodiments of this disclosure. Further details will not be elaborated here.

[0169] Exemplary device Figure 20 This is a schematic diagram of a denoising parameter determination apparatus provided in an exemplary embodiment of this disclosure. The denoising parameter determination apparatus of this disclosure can be used to implement the denoising parameter determination method of any of the above embodiments. Figure 20 As shown, the noise reduction parameter determination device in this embodiment includes: The first acquisition unit 510 is configured to: acquire multiple denoised images corresponding to the denoised image, and obtain the denoised image by performing denoising processing on the denoised image using multiple denoising parameters under the corresponding parameter value group for any denoised image; The first determining unit 520 is configured to: for any denoised image: Determine the difference between any denoised image and an undenoised image to obtain the difference image corresponding to any denoised image; Determine the degree of dispersion of the difference image corresponding to any denoised image; Perform singular value decomposition on the difference image corresponding to any denoised image to obtain the set of singular values ​​corresponding to any denoised image; determine the degree of singular value dispersion of the set of singular values ​​corresponding to any denoised image. The second determining unit 530 is configured to: determine a target parameter value set for image denoising based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value set corresponding to the multiple denoised images respectively.

[0170] Figure 21 This is a schematic diagram of the structure of a noise reduction parameter determination device provided in yet another exemplary embodiment of this disclosure. For example... Figure 21 As shown, in Figure 20 Based on the illustrated embodiment, in some possible implementations, the second determining unit 530 includes: The first determining subunit 531 is configured to: for any one of the multiple denoised images: Based on the degree of difference dispersion and the degree of singular value dispersion of any denoised image, the evaluation value of the denoising parameters corresponding to any denoised image can be obtained. Based on the denoising parameter evaluation values ​​corresponding to multiple denoised images and the parameter value groups corresponding to multiple denoised images, the target parameter value group for image denoising processing is determined.

[0171] In some alternative implementations, the first acquisition unit includes 510: The second determining subunit 511 is configured to: for any one of the M denoising parameters: From the range of any denoising parameter, determine N values ​​for the first parameter, where M and N are both positive integers; The third determining subunit 512 is configured to: determine a set of first parameter value groups based on N first parameter values ​​corresponding to any one of the M denoising parameters, wherein the first parameter value in each first parameter value group in the set of first parameter value groups corresponds one-to-one with the denoising parameters in the M denoising parameters, and any two first parameter value groups in the set of first parameter value groups contain at least one pair of different first parameter values ​​of the same denoising parameter; The first denoising subunit 513 is configured to: perform denoising processing on the undated image using any first parameter value group in the first parameter value group set, to obtain the first denoised image.

[0172] In some alternative implementations, the second determining subunit 511 includes: The first determining module 5111 is configured to: determine the sampling interval corresponding to any denoising parameter based on the value range of any denoising parameter and the preset selection quantity; The second determining module 5112 is configured to: determine N first parameter values ​​from the range of any denoising parameter based on the sampling interval.

[0173] In some alternative implementations, the second determining unit 530 includes: The third determining subunit 532 is configured to: determine an initial parameter value group from the parameter value groups corresponding to the multiple denoised images based on the degree of difference dispersion and the degree of singular value dispersion corresponding to the multiple denoised images respectively; The fourth determining subunit 533 is configured to: target any one of the preset M denoising parameters: Determine the initial parameter value corresponding to any denoising parameter from the initial parameter value set; From the range of any denoising parameter, determine X second parameter values ​​within the neighborhood of the initial parameter value, where X is a positive integer; The fifth determining subunit 534 is configured to: determine a set of second parameter value groups based on X second parameter values ​​corresponding to any one of the M denoising parameters, wherein the second parameter value in each second parameter value group in the set of second parameter value groups corresponds one-to-one with the denoising parameters in the M denoising parameters, and any two second parameter value groups in the set of second parameter value groups contain at least one pair of different second parameter values ​​of the same denoising parameter; The second denoising subunit 535 is configured to: for any second parameter value group in the set of second parameter value groups, perform denoising processing on the denoised image using any second parameter value group to obtain the second denoised image; The sixth determining subunit 536 is configured to: determine the target parameter value set for image denoising based on the difference dispersion degree, singular value dispersion degree corresponding to the multiple second denoised images respectively, and the parameter value set corresponding to the multiple second denoised images respectively.

[0174] In some alternative implementations, the first determining unit 520 includes: The seventh determining subunit 521 is configured to: determine the difference between the pixel value of any channel of any denoised image and the pixel value of the corresponding channel of any undated image, and obtain the difference image corresponding to any channel of any denoised image.

[0175] In some alternative implementations, the first determining unit 520 includes: The eighth determining subunit 522 is configured to: provide a difference image for any channel of any denoised image. Divide the difference image of any channel to obtain multiple image blocks for any channel; Singular value decomposition is performed on the pixel values ​​of any image block in any channel to obtain the set of singular values ​​of any image block in any channel of any denoised image.

[0176] In some alternative implementations, the first determining unit 520 includes: The ninth determining subunit 523 is configured to: for any image block of any channel of any denoised image: determine the degree of dispersion of the pixel values ​​of any image block of any channel, and obtain the degree of difference dispersion of any image block of any channel of any denoised image.

[0177] In some alternative implementations, the first determining unit 520 includes: The tenth determining subunit 524 is configured to: for any image block of any channel of any denoised image: determine the degree of singular value discreteness of the singular value set of any image block of any channel of any denoised image based on the singular value set of any image block of any denoised image.

[0178] In some alternative implementations, the second determining unit 530 includes: The eleventh determining subunit 537 is configured to: determine the target parameter value set for image denoising based on the difference dispersion degree, singular value dispersion degree corresponding to each channel of each image block of multiple denoised images, and the parameter value set corresponding to each of the multiple denoised images.

[0179] In some optional implementations, the number of undisturbed images is a preset number, wherein different undisturbed images correspond to different scenes; The second determining unit 530 includes: The twelfth determining subunit 538 is configured to: for any denoised image among multiple denoised images, based on the difference dispersion degree, singular value dispersion degree corresponding to the multiple denoised images corresponding to the denoised image, and the parameter value group corresponding to the multiple denoised images corresponding to the denoised image, determine the parameter value group for image denoising processing corresponding to the denoised image. The thirteenth determining subunit 539 is configured to: determine the target parameter value group based on the parameter value groups for image denoising processing corresponding to the multiple undated images respectively.

[0180] The exemplary embodiment of this device corresponds to the above-described method for determining denoising parameters in terms of implementation. The corresponding content of the two can be referenced, combined, and cited interchangeably, and will not be repeated here. The beneficial technical effects corresponding to the exemplary embodiment of this device can be found in the corresponding beneficial technical effects of the above-described method for determining denoising parameters, and will not be repeated here.

[0181] Figure 22 This is a schematic diagram of the structure of an image denoising apparatus provided in an exemplary embodiment of this disclosure. The image denoising apparatus of this disclosure can be used to implement the image denoising method of any of the above embodiments. Figure 22 As shown, the image denoising device in this embodiment includes: The second acquisition unit 610 is configured to acquire the image to be denoised and a target parameter value group of denoising parameters, wherein the target parameter value group is determined by any of the denoising parameter determining devices described above. The denoising unit 620 is configured to perform denoising processing on the image to be denoised based on the target parameter value group.

[0182] The exemplary embodiment of this device corresponds to the image denoising method described above in terms of implementation. The corresponding content of the two can be referenced, combined, and cited interchangeably, and will not be repeated here. The beneficial technical effects corresponding to the exemplary embodiment of this device can be found in the corresponding beneficial technical effects of the image denoising method described above, and will not be repeated here.

[0183] Exemplary electronic devices Figure 23 A structural diagram of an electronic device provided in an embodiment of this disclosure includes at least one processor 111 and a memory 112.

[0184] The processor 111 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0185] The memory 112 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 111 may execute one or more computer program instructions to implement the denoising parameter determination or image denoising method and / or other desired functions of the various embodiments of this disclosure described above.

[0186] In one example, the electronic device may also include an input device 113 and an output device 114, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0187] The input device 113 may also include, for example, a keyboard, a mouse, etc.

[0188] The output device 114 can output various information to the outside, including, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0189] Of course, for the sake of simplicity, Figure 23 Only some of the components of the electronic device relevant to this disclosure are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0190] Exemplary computer program products and computer-readable storage media In addition to the methods and apparatus described above, embodiments of this disclosure may also provide a computer program product, including computer program instructions, which, when executed by a processor, cause the processor to perform steps in the denoising parameter determination or image denoising method of the various embodiments of this disclosure described in the "Exemplary Methods" section above.

[0191] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. These programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0192] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform steps in the denoising parameter determination or image denoising method of the various embodiments of this disclosure described in the "Exemplary Methods" section above.

[0193] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, but is not limited to, systems, apparatuses, or devices that are electrical, magnetic, optical, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0194] The basic principles of this disclosure have been described above with reference to specific embodiments. However, the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0195] Various modifications and variations can be made to this disclosure without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this disclosure and their equivalents, this disclosure is also intended to include such modifications and variations.

Claims

1. A method for determining denoising parameters, comprising: Multiple denoised images corresponding to the denoised image are obtained, and each of the denoised images is used to denoise the denoised image under the corresponding parameter value group to obtain the denoised image. For any of the denoised images described above: Determine the difference between any of the denoised images and the undenoised images to obtain a difference image corresponding to any of the denoised images; Determine the degree of difference dispersion of the difference image corresponding to any of the denoised images; Perform singular value decomposition on the difference image corresponding to any of the denoised images to obtain the singular value set corresponding to any of the denoised images; determine the degree of singular value dispersion of the singular value set corresponding to any of the denoised images; Based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value group corresponding to the multiple denoised images, a target parameter value group for image denoising processing is determined.

2. The method according to claim 1, wherein, The step of determining a target parameter value set for image denoising based on the difference dispersion, the singular value dispersion, and the parameter value set corresponding to each of the denoised images includes: For any one of the denoised images: Based on the degree of difference dispersion and the degree of singular value dispersion corresponding to any of the denoised images, the evaluation value of the denoising parameters corresponding to any of the denoised images is obtained; Based on the denoising parameter evaluation values ​​corresponding to the multiple denoised images and the parameter value groups corresponding to the multiple denoised images, a target parameter value group for image denoising processing is determined.

3. The method according to claim 1, wherein, The step of obtaining multiple denoised images corresponding to the undenoised image includes: For any one of the M denoising parameters: From the range of values ​​of any of the denoising parameters, determine N values ​​for the first parameter, where M and N are both positive integers; Based on N first parameter values ​​corresponding to any one of the M denoising parameters, a set of first parameter value groups is determined. The first parameter value in each first parameter value group in the set of first parameter value groups corresponds one-to-one with the denoising parameters in the M denoising parameters. Any two first parameter value groups in the set of first parameter value groups contain at least one pair of different first parameter values ​​of the same denoising parameter. For any one of the first parameter value groups in the first parameter value group set, the denoised image is denoised using any one of the first parameter value groups to obtain a first denoised image.

4. The method according to claim 3, wherein, Determining N first parameter values ​​from the range of any of the denoising parameters includes: Based on the value range of any of the denoising parameters and the preset selection quantity, determine the sampling interval corresponding to any of the denoising parameters; Based on the sampling interval, N first parameter values ​​are determined from the range of any of the denoising parameters.

5. The method according to claim 3, wherein, The step of determining a target parameter value set for image denoising based on the difference dispersion, the singular value dispersion, and the parameter value set corresponding to each of the denoised images includes: Based on the degree of difference dispersion and the degree of singular value dispersion corresponding to the multiple denoised images respectively, an initial parameter value group is determined from the parameter value group corresponding to the multiple denoised images respectively; For any one of the preset M denoising parameters: Determine the initial parameter value corresponding to any of the denoising parameters from the initial parameter value group; From the range of values ​​of any of the denoising parameters, determine X second parameter values ​​within the neighborhood of the initial parameter value, where X is a positive integer; Based on X second parameter values ​​corresponding to any one of the M denoising parameters, a set of second parameter value groups is determined. The second parameter value in each second parameter value group in the set of second parameter value groups corresponds one-to-one with the denoising parameter in the M denoising parameters. Any two second parameter value groups in the set of second parameter value groups contain at least one pair of different second parameter values ​​of the same denoising parameter. For any one of the second parameter value groups in the second parameter value group set, the undenoised image is denoised using any one of the second parameter value groups to obtain a second denoised image; Based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value group corresponding to the multiple second denoised images respectively, a target parameter value group for image denoising processing is determined.

6. The method according to any one of claims 1-5, wherein, Determining the difference between any of the denoised images and the undenoised images includes: The difference between the pixel value of any channel of any denoised image and the pixel value of the corresponding channel of the undated image is determined to obtain the difference image corresponding to any channel of any denoised image.

7. The method according to claim 6, wherein, The step of performing singular value decomposition on the difference image corresponding to any of the denoised images to obtain the set of singular values ​​corresponding to any of the denoised images includes: The difference image for any channel of any of the denoised images: The difference image of any channel is divided to obtain multiple image blocks for any channel; Singular value decomposition is performed on the pixel values ​​of any image block in any channel to obtain the set of singular values ​​of any image block in any channel of any denoised image.

8. The method according to claim 7, wherein, Determining the degree of difference dispersion of the difference image corresponding to any of the denoised images includes: For any image block of any channel in any denoised image: determine the degree of dispersion of the pixel values ​​of any image block of any channel, and obtain the degree of difference dispersion of any image block of any channel in any denoised image.

9. The method according to claim 7, wherein, Determining the singular value discreteness of the singular value set corresponding to any of the denoised images includes: For any image block of any channel of any denoised image: Based on the set of singular values ​​of any image block of any channel of any denoised image, determine the degree of singular value dispersion of the set of singular values ​​of any image block of any channel of any denoised image.

10. The method according to claim 9, wherein, The step of determining a target parameter value set for image denoising based on the difference dispersion, the singular value dispersion, and the parameter value set corresponding to each of the denoised images includes: Based on the degree of difference dispersion, the degree of singular value dispersion corresponding to each channel of each image block of the multiple denoised images, and the parameter value group corresponding to each of the multiple denoised images, a target parameter value group for image denoising processing is determined.

11. The method according to any one of claims 1-5, wherein, The number of undenoised images is a preset number, where different undenoised images correspond to different scenes; The step of determining a target parameter value set for image denoising based on the difference dispersion, the singular value dispersion, and the parameter value set corresponding to each of the denoised images includes: For any one of the multiple denoised images, based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value group corresponding to the multiple denoised images corresponding to any one denoised image, a parameter value group for image denoising processing corresponding to any one denoised image is determined. Based on the parameter value groups for image denoising processing corresponding to the multiple undenoised images, a target parameter value group is determined.

12. An image denoising method, comprising: Obtain the image to be denoised and a target parameter value set of denoising parameters, wherein the target parameter value set is determined by the denoising parameter determination method according to any one of claims 1-11 above; The image to be denoised is denoised based on the target parameter value set.

13. A noise reduction parameter determination device, comprising: The first acquisition unit is configured to: acquire multiple denoised images corresponding to the denoised image, wherein any one of the denoised images is used to denoise the denoised image under a corresponding parameter value group to obtain the denoised image; The first determining unit is configured to: for any of the denoised images: Determine the difference between any of the denoised images and the undenoised images to obtain a difference image corresponding to any of the denoised images; Determine the degree of difference dispersion of the difference image corresponding to any of the denoised images; Perform singular value decomposition on the difference image corresponding to any of the denoised images to obtain the singular value set corresponding to any of the denoised images; determine the degree of singular value dispersion of the singular value set corresponding to any of the denoised images; The second determining unit is configured to: determine a target parameter value set for image denoising processing based on the degree of difference dispersion, the degree of singular value dispersion, and the parameter value set corresponding to the multiple denoised images respectively.

14. An image denoising apparatus, comprising: The second acquisition unit is configured to acquire the image to be denoised and a target parameter value group of denoising parameters, wherein the target parameter value group is determined by the denoising parameter determination device as described in claim 13. The denoising unit is configured to perform denoising processing on the image to be denoised based on the target parameter value set.

15. A computer-readable storage medium storing a computer program, which, when executed, performs the method according to any one of claims 1-12.

16. An electronic device, the electronic device comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1-12.

Images