Image denoising method and device

Abstract

The invention relates to an image denoising method and device, and the method comprises the steps: obtaining an image, and obtaining the brightness information before denoising from the image; performing spatial filtering on the brightness information before denoising to obtain the brightness information of the image after spatial denoising; obtaining first brightness residual information of the image according to the brightness information before denoising and the brightness information after airspace denoising; performing frequency domain filtering on the first brightness residual information to obtain second brightness residual information of the image; and obtaining denoised brightness information of the image according to the denoised brightness information of the spatial domain and the second brightness residual information. According to the method, a better image denoising effect is achieved by combining a spatial domain denoising method and a frequency domain denoising method, denoising is mainly executed based on brightness information, the denoising efficiency is improved, the denoising efficiency is improved by adopting haar wavelet decomposition to assist in denoising, the hardware cache requirement is reduced, and the method is suitable for large-scale popularization and application. And popularization and application in miniaturized image processing equipment are facilitated.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technologies, and particularly to an image denoising method and apparatus. Background Art

[0002] Image denoising is an important part of image processing. Image denoising methods have evolved through stages such as traditional spatial domain denoising, frequency domain filtering denoising based on Fourier transform and discrete cosine transform, denoising methods based on variational methods and partial differential equations simulating heat convection, multi-scale denoising using wavelets and multi-scale geometric transforms (super wavelets), and AI (artificial intelligence) denoising based on CNN (deep convolutional neural network).

[0003] In traditional image denoising methods, spatial domain filtering and frequency domain filtering each have their own advantages and disadvantages. Spatial domain filtering processes the image in the original image space, usually achieving image denoising effects through convolution operations of the image matrix with a template (filter). However, since the spatial domain cannot intuitively distinguish between noise and details, image details may be damaged during denoising. Frequency domain denoising refers to converting the original image from the spatial domain to the frequency domain, performing image denoising in the frequency domain, and then converting the frequency domain image back to the spatial domain image. The disadvantages are that on the one hand, it is more complex to implement, and on the other hand, noise is likely to remain. Summary of the Invention

[0004] In view of this, the present disclosure provides an image denoising method and apparatus, which combines spatial domain filtering and frequency domain filtering to achieve better image denoising effects.

[0005] The technical solution of the present disclosure is implemented as follows:

[0006] An image denoising method, comprising:

[0007] Obtaining an image and obtaining the pre-denoising luminance information from the image;

[0008] Performing spatial domain filtering on the pre-denoising luminance information of the image to obtain the post-spatial domain denoising luminance information of the image;

[0009] Obtaining the first luminance residual information of the image according to the pre-denoising luminance information of the image and the post-spatial domain denoising luminance information of the image;

[0010] Performing frequency domain filtering on the first luminance residual information of the image to obtain the second luminance residual information of the image;

[0011] Obtaining the post-denoising luminance information of the image according to the post-spatial domain denoising luminance information of the image and the second luminance residual information of the image.

[0012] In a possible implementation manner, performing spatial domain filtering on the pre-denoising luminance information of the image to obtain the post-spatial domain denoising luminance information of the image includes the following steps performed on each pixel point in the image:

[0013] In the luminance plane of the image, obtaining the luminance frequency division operator of the current pixel point through Haar wavelet decomposition, where the current pixel point is any pixel point in the image;

[0014] According to the mapping relationship between the luminance frequency division operator and a preset first denoising intensity, obtaining the first denoising intensity of the current pixel point;

[0015] According to the first denoising intensity of the current pixel point, performing spatial domain filtering on the luminance value of the current pixel point to obtain the post-spatial domain denoising luminance value of the current pixel point.

[0016] In a possible implementation manner, the obtaining the luminance frequency division operator of the current pixel point by performing Haar wavelet decomposition in the luminance plane of the image includes:

[0017] In the luminance plane of the image, performing Haar wavelet decomposition on the luminance values of the pixel points within the 2 k ×2 k neighborhood range around the current pixel point to obtain 3 groups of 2 k-1 ×2 k-1 -dimensional high-frequency wavelet coefficients and 1 group of 2 k-1 ×2 k-1 -dimensional low-frequency wavelet coefficients, where k≥2;

[0018] According to the 3 groups of 2 k-1 ×2 k-1 -dimensional high-frequency wavelet coefficients, obtaining the high-frequency statistical value associated with 2 k-1 ×2 k-1 -dimensional;

[0019] When k - 1>0, performing Haar wavelet decomposition on the 1 group of 2 k-1 ×2 k-1 -dimensional low-frequency wavelet coefficients to obtain 3 groups of 2 k-2 ×2 k-2 -dimensional high-frequency wavelet coefficients and 1 group of 2 k-2 ×2 k-2 -dimensional low-frequency wavelet coefficients, and according to the 3 groups of 2 k-2 ×2 k-2 -dimensional high-frequency wavelet coefficients, obtaining the high-frequency statistical value associated with 2 k-2 ×2 k-2 -dimensional;

[0020] When k - 2 > 0, for the 1 group of 2 k-2 ×2 k-2 dimensional low - frequency wavelet coefficients, perform Haar wavelet decomposition to obtain 3 groups of 2 k-3 ×2 k-3 dimensional high - frequency wavelet coefficients and 1 group of 2 k-3 ×2 k-3 dimensional low - frequency wavelet coefficients. According to the 3 groups of 2 k-3 ×2 k-3 dimensional high - frequency wavelet coefficients, obtain the high - frequency statistical value associated with 2 k-3 ×2 k-3 dimensions;

[0021] And so on, until obtaining the high - frequency statistical value associated with 1×1 dimension;

[0022] According to the high - frequency statistical values from the high - frequency statistical value associated with 2 k-1 ×2 k-1 dimensions to the high - frequency statistical value associated with 1×1 dimension, obtain the luminance frequency - division operator of the current pixel point.

[0023] In a possible implementation manner, the obtaining of the high - frequency statistical value associated with 2 k-1 ×2 k-1 dimensions according to the 3 groups of 2 k-1 ×2 k-1 dimensional high - frequency wavelet coefficients, and the obtaining of the high - frequency statistical value associated with 2 k-2 ×2 k-2 dimensions according to the 3 groups of 2 k-2 ×2 k-2 dimensional high - frequency wavelet coefficients, and the obtaining of the high - frequency statistical value associated with 2 k-3 ×2 k-3 dimensions according to the 3 groups of 2 k-3 ×2 k-3 dimensional high - frequency wavelet coefficients all include:

[0024] Among the 3 groups of 2 m ×2 m dimensional high - frequency wavelet coefficients, by comparing each same - position value, obtain the maximum value at each same - position of the 3 groups of 2 m ×2 m dimensional high - frequency wavelet coefficients;

[0025] According to the maximum value at each same - position of the 3 groups of 2 m ×2 m dimensional high - frequency wavelet coefficients, obtain the high - frequency statistical value associated with 2 m ×2 m dimensions;

[0026] Among them, in the case of 2 k-1 ×2 k-1 dimensions, m = k - 1, in 2 k-2 ×2 k-2 dimensions, m = k - 2, in 2 k-3 ×2 k-3 dimensions, m = k - 3.

[0027] In a possible implementation manner, the obtaining of the brightness frequency division operator of the current pixel point according to the high-frequency statistical values associated with 2 k-1 ×2 k-1 dimensions to the high-frequency statistical values associated with 1×1 dimensions includes:

[0028] Determining the average value of the high-frequency statistical values from the high-frequency statistical values associated with 2 k-1 ×2 k-1 dimensions to the high-frequency statistical values associated with 1×1 dimensions as the brightness frequency division operator of the current pixel point.

[0029] In a possible implementation manner, the obtaining of the spatially denoised brightness value of the current pixel point by performing spatial domain filtering on the brightness value of the current pixel point according to the first denoising intensity of the current pixel point includes:

[0030] Adjusting a preset initialization filter according to the first denoising intensity within a preset filtering window to obtain a spatial domain filter for performing spatial domain filtering on the current pixel point;

[0031] Performing spatial domain filtering on the current pixel point by using the spatial domain filter to obtain the spatially denoised brightness value of the current pixel point.

[0032] In a possible implementation manner, the obtaining of the first brightness residual information of the image according to the brightness information before denoising of the image and the spatially denoised brightness information of the image includes the following steps performed on each pixel point in the image:

[0033] Determining the difference between the brightness value before denoising of the current pixel point and the spatially denoised brightness value of the current pixel point as the first brightness residual value of the current pixel point, where the current pixel point is any pixel point in the image.

[0034] In a possible implementation manner, the obtaining of the second brightness residual information of the image by performing frequency domain filtering on the first brightness residual information of the image includes the following steps performed on each pixel point in the image:

[0035] In the first luminance residual plane composed of the first luminance residual information of the image, through Haar wavelet decomposition, the high-frequency luminance residual wavelet coefficients and low-frequency luminance residual wavelet coefficients of the current pixel are obtained;

[0036] According to the mapping relationship between the luminance frequency division operator of the current pixel and a preset second denoising intensity, the second denoising intensity of the current pixel is obtained;

[0037] Using the second denoising intensity of the current pixel to shrink the high-frequency luminance residual wavelet coefficients of the current pixel, the shrunk high-frequency luminance residual wavelet coefficients of the current pixel are obtained;

[0038] According to the shrunk high-frequency luminance residual wavelet coefficients and low-frequency luminance residual wavelet coefficients of the current pixel, perform frequency domain reconstruction on the current pixel to obtain the second luminance residual value of the current pixel;

[0039] Wherein, the second luminance residual information of the image is composed of the second luminance residual values of all pixels in the image.

[0040] In a possible implementation manner, the image denoising method further includes:

[0041] Obtain the pre-denoising chrominance information from the image;

[0042] Perform spatial domain filtering on the pre-denoising chrominance information of the image to obtain the spatially denoised chrominance information of the image;

[0043] Determine the spatially denoised chrominance information as the post-denoising chrominance information of the image.

[0044] An image denoising device includes:

[0045] An image information extraction module, configured to execute obtaining an image and obtaining pre-denoising luminance information from the image;

[0046] A spatial domain filtering module, configured to execute performing spatial domain filtering on the pre-denoising luminance information of the image to obtain the spatially post-denoised luminance information of the image;

[0047] A first luminance residual obtaining module, configured to execute obtaining the first luminance residual information of the image according to the pre-denoising luminance information of the image and the spatially post-denoised luminance information of the image;

[0048] A second luminance residual obtaining module, configured to execute performing frequency domain filtering on the first luminance residual information of the image to obtain the second luminance residual information of the image;

[0049] A denoised image obtaining module, configured to obtain denoised luminance information of the image according to the luminance information after spatial domain denoising of the image and the second luminance residual information of the image.

[0050] As can be seen from the above solution, the image denoising method and device of the present disclosure combine the methods of spatial domain denoising and frequency domain denoising to achieve a better image denoising effect. During the denoising process, denoising is mainly performed based on the luminance information of the image, thereby capturing the feature that the luminance of the noise points is significantly different from the luminance of other surrounding pixel points, which can help improve the denoising efficiency. According to the technical solution of the present disclosure, during the denoising process, the means of Haar wavelet decomposition is mainly used to assist spatial domain denoising and frequency domain denoising. Therefore, compared with denoising means such as Fourier transform, the denoising efficiency can be significantly improved, and it helps to reduce the hardware cache requirements during the denoising process, which is conducive to the popularization and application in miniaturized image processing devices. Description of the Drawings

[0051] Figure 1 is a flowchart of an image denoising method shown according to a schematic embodiment;

[0052] Figure 2 is a schematic diagram of a spatial domain filtering process shown according to a schematic embodiment;

[0053] Figure 3 is a schematic diagram of a process of obtaining a luminance frequency division operator shown according to a schematic embodiment;

[0054] Figure 4 is a schematic diagram of a graphical expression process of a frequency division operator shown according to a schematic embodiment;

[0055] Figure 5 is a flowchart of a process of obtaining a high-frequency statistical value according to high-frequency wavelet coefficients shown according to a schematic embodiment;

[0056] Figure 6 is a schematic diagram of a mapping relationship between a luminance frequency division operator and a first denoising intensity shown according to a schematic embodiment;

[0057] Figure 7 is a flowchart of a process of performing spatial domain filtering on the luminance value of a current pixel point shown according to a schematic embodiment;

[0058] Figure 8 A flowchart of a process of performing frequency domain filtering on the first luminance residual information of an image shown according to a schematic embodiment;

[0059] Figure 9 is a schematic diagram of a mapping relationship between a luminance frequency division operator and a second denoising intensity shown according to a schematic embodiment;

[0060] Figure 10 It is a schematic process diagram of chromaticity denoising shown according to a schematic embodiment;

[0061] Figure 11 It is a schematic diagram of the steps of a specific application scenario of an image denoising method shown according to a schematic embodiment;

[0062] Figure 12 It is a schematic flow diagram of a specific application scenario of an image denoising method shown according to a schematic embodiment;

[0063] Figure 13 It is a schematic diagram of the logical structure of an image denoising device shown according to a schematic embodiment;

[0064] Figure 14 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. Specific Embodiments

[0065] In order to make the objectives, technical solutions and advantages of the present disclosure clearer, the following examples are given with reference to the accompanying drawings to further elaborate on the present disclosure in detail.

[0066] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0067] Figure 1 It is a flow chart of an image denoising method shown according to a schematic embodiment. As Figure 1 shown, the image denoising method mainly includes the following steps 101 to step 105.

[0068] Step 101, obtain an image and obtain the luminance information before denoising from the image.

[0069] In the schematic embodiment, the image can be a YUV format image. In the YUV format image, the Y channel information is the luminance information. In step 101, if the obtained image is a YUV format image, the luminance information before denoising obtained from the image is the Y channel information before denoising. Since the image is composed of multiple pixel points, based on this, in the schematic embodiment, the luminance information before denoising includes the luminance values of all pixel points in the image.

[0070] Since the noise in the image is mainly manifested as the brightness difference between it and other surrounding pixels, in the exemplary embodiment, the image denoising method of the present disclosure mainly performs denoising processing on the brightness information in the image. Other information such as U (the blue projection part of the chrominance component) and V (the red projection part of the chrominance component) information can be denoised or not. In the subsequent embodiments of the present disclosure, a method for denoising U and V information is introduced. Combining it with the method for denoising the brightness information in the embodiments of the present disclosure can achieve a better denoising effect.

[0071] In addition to the YUV format, existing image formats also include multiple formats such as the RGB format. If it is necessary to use the image denoising method of the embodiments of the present disclosure for images in other formats, in the exemplary embodiment, according to the relevant conversion relationship, before performing the method steps of the embodiments of the present disclosure, the image in other formats can be converted into a YUV format image, and after denoising, it can be converted into other required formats as needed.

[0072] In the exemplary embodiment, the image obtained in step 101 can be a preprocessed image. The preprocessing can include operations such as geometric correction, brightness correction, and color correction of the image, and can also include image convolution or window operations. For the subsequent spatial domain filtering and frequency domain filtering steps, the preprocessing can also include an edge expansion operation at the positions of the four boundary points of the image.

[0073] Step 102: Perform spatial domain filtering on the brightness information of the image before denoising to obtain the brightness information of the image after spatial domain denoising.

[0074] Among them, various existing spatial domain filtering methods can be used for spatial domain filtering. In the exemplary embodiment, the present disclosure provides a spatial domain filtering method. Figure 2 is a schematic diagram of the spatial domain filtering process shown according to an exemplary embodiment, as Figure 2 shown. In the exemplary embodiment, step 102 can include the following steps 201 to 203 performed on each pixel point in the image.

[0075] Step 201: In the brightness plane of the image, obtain the brightness frequency division operator of the current pixel point through Haar wavelet decomposition, where the current pixel point is any pixel point in the image.

[0076] Among them, Haar wavelet decomposition, that is, Haar wavelet transform, is a method of decomposing an image into sub-signals of different frequencies, including approximation components and detail components. Through Haar wavelet decomposition, multi-scale analysis of the image can be achieved, and features at different scales can be extracted. Haar wavelet decomposition is a simple wavelet transform. Since no information is lost during this process, the original image can be reconstructed from the recorded data. In the embodiments of the present disclosure, using the idea of obtaining sub-signals of different frequencies through Haar wavelet decomposition, pixel points in the luminance plane that can reflect significant differences in luminance frequency from the surrounding are obtained. Usually, if the luminance frequency of a pixel point is significantly different from that of other surrounding pixel points, it indicates that the pixel point belongs to noise, and noise reduction processing can be performed on it. In the embodiments of the present disclosure, the luminance frequency division operator obtained through Haar wavelet decomposition is used to characterize the (luminance) frequency intensity of the pixel point.

[0077] In the exemplary embodiment, Haar wavelet decomposition needs to perform calculations on the pixel points within the neighborhood range of the pixel point. For the convenience of computer calculation, the neighborhood range needs to meet the condition of containing 2 k ×2 k pixel points, where k≥2. Figure 3 is a schematic diagram of the process of obtaining the luminance frequency division operator shown according to an exemplary embodiment. As Figure 3 shown, step 201 may specifically include the following steps 301 to 305.

[0078] Step 301: In the luminance plane of the image, perform Haar wavelet decomposition on the luminance values of the pixel points within the 2 k ×2 k neighborhood range around the current pixel point to obtain 3 sets of 2 k-1 ×2 k-1 dimensional high-frequency wavelet coefficients and 1 set of 2 k-1 ×2 k-1 dimensional low-frequency wavelet coefficients, where k≥2;

[0079] Step 302: According to the 3 sets of 2 k-1 ×2 k-1 dimensional high-frequency wavelet coefficients, obtain the high-frequency statistical value associated with 2 k-1 ×2 k-1 dimensions;

[0080] Step 303: When k - 1>0, perform Haar wavelet decomposition on 1 set of 2 k-1 ×2 k-1 dimensional low-frequency wavelet coefficients to obtain 3 sets of 2 k-2 ×2 k-2 dimensional high-frequency wavelet coefficients and 1 set of 2 k-2 ×2 k-2Low-frequency wavelet coefficients of the dimension, according to 3 groups of 2 k -2 ×2 k-2 High-frequency wavelet coefficients of the dimension, to obtain high-frequency statistical values associated with 2 k-2 ×2 k-2 Dimension;

[0081] Step 304, when k - 2 > 0, perform Haar wavelet decomposition on the low-frequency wavelet coefficients of 1 group of 2 k-2 ×2 k-2 Dimension, to obtain 3 groups of 2 k-3 ×2 k-3 High-frequency wavelet coefficients of the dimension and 1 group of 2 k-3 ×2 k-3 Low-frequency wavelet coefficients of the dimension, according to 3 groups of 2 k -3 ×2 k-3 High-frequency wavelet coefficients of the dimension, to obtain high-frequency statistical values associated with 2 k-3 ×2 k-3 Dimension; and so on, until high-frequency statistical values associated with 1×1 dimension are obtained;

[0082] Step 305, according to the high-frequency statistical values from those associated with 2 k-1 ×2 k-1 Dimension to those associated with 1×1 dimension, obtain the brightness frequency division operator of the current pixel point.

[0083] Figure 4 is a schematic diagram of the graphical expression process of the frequency division operator shown according to an exemplary embodiment. The following combines Figure 4 to further illustrate the above steps 301 to 305.

[0084] In Figure 4 In the shown embodiment, the leftmost side is the pixel points within the 8×8 neighborhood around the current pixel point. Each small square represents a pixel point, and the small square with a grid is the current pixel point. Corresponding to steps 301 to 305, k = 3, that is, 2 k = 2 3 = 8.

[0085] As Figure 4 shown, in step 301, in the luminance plane of the image, perform Haar wavelet decomposition on the luminance values of the pixel points within the 8×8 neighborhood around the current pixel point, to obtain 3 groups of high-frequency wavelet coefficients of 4×4 dimension (i.e., 2 2 ×2 2 Dimension) and 1 group of high-frequency wavelet coefficients of 4×4 dimension (i.e., 2 2 ×2 2The low-frequency wavelet coefficients of the 0th layer (dimension), i.e., the wavelet decomposition result of the 0th layer. Among them, the three groups of high-frequency wavelet coefficients of 4×4 dimension are 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, and 4×4 dimension HH wavelet coefficients, and the one group of low-frequency wavelet coefficients of 4×4 dimension is 4×4 dimension LL wavelet coefficients.

[0086] Among them, the 4×4 dimension LH wavelet coefficients represent the horizontal characteristics of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point, and it reflects the fluctuation of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point in the horizontal direction; the 4×4 dimension HL wavelet coefficients represent the vertical characteristics of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point, and it reflects the fluctuation of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point in the vertical direction; the 4×4 dimension HH wavelet coefficients represent the diagonal characteristics of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point, and it reflects the fluctuation of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point in the diagonal direction; the 4×4 dimension LL wavelet coefficients represent the characteristics of the low-frequency part of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point, representing the general outline of the change of the luminance values of the pixels in the 8×8 neighborhood around the current pixel point.

[0087] The Haar wavelet decomposition (or called Haar wavelet transform) is one of the simple and widely used wavelet transform methods. For the specific implementation details of the Haar wavelet transform, reference can be made to the prior art implementation, which will not be elaborated here.

[0088] Figure 4 The embodiment shown corresponds to the case of k = 3. After step 301, the wavelet decomposition result of the 0th layer is obtained. As described above, the wavelet decomposition result of the 0th layer includes 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, 4×4 dimension HH wavelet coefficients, and 4×4 dimension LL wavelet coefficients. Among them, the obtained one group of 2 k-1 ×2 k-1 dimension low-frequency wavelet coefficients (4×4 dimension LL wavelet coefficients) is 2 3-1 = 2 2The low-frequency wavelet coefficients of the dimension, that is, k - 1 = 3 - 1 = 2 > 0, belong to the case where k - 1 > 0 in step 303. In this case, in step 303, Haar wavelet decomposition is performed on a group of low-frequency wavelet coefficients of 4×4 dimension, that is, Haar wavelet decomposition is performed on the 4×4 dimension LL wavelet coefficients, and three groups of high-frequency wavelet coefficients of 2×2 dimension and one group of low-frequency wavelet coefficients of 2×2 dimension are obtained, that is, the wavelet decomposition result of the first layer. Among them, the three groups of high-frequency wavelet coefficients of 2×2 dimension are the 2×2 dimension LH wavelet coefficients, the 2×2 dimension HL wavelet coefficients, and the 2×2 dimension HH wavelet coefficients respectively, and the one group of low-frequency wavelet coefficients of 2×2 dimension is the 2×2 dimension LL wavelet coefficients.

[0089] In step 303, when k - 1 > 0, for one group of 2 k-1 ×2 k-1 dimension low-frequency wavelet coefficients, Haar wavelet decomposition is performed, aiming to continue to extract the high-frequency components from the 2 k-1 ×2 k-1 dimension wavelet coefficients. In the Figure 4 illustrated embodiment, the 2×2 dimension LH wavelet coefficients represent the characteristics of the 4×4 dimension LL wavelet coefficients in the horizontal direction, and it reflects the fluctuation of the 4×4 dimension LL wavelet coefficients in the horizontal direction; the 2×2 dimension HL wavelet coefficients represent the characteristics of the 4×4 dimension LL wavelet coefficients in the vertical direction, and it reflects the fluctuation of the 4×4 dimension LL wavelet coefficients in the vertical direction; the 2×2 dimension HH wavelet coefficients represent the characteristics of the 4×4 dimension LL wavelet coefficients in the diagonal direction, and it reflects the fluctuation of the 4×4 dimension LL wavelet coefficients in the diagonal direction; the 2×2 dimension LL wavelet coefficients represent the characteristics of the low-frequency part of the 4×4 dimension LL wavelet coefficients, representing the general outline of the change of the 4×4 dimension LL wavelet coefficients.

[0090] Continue to refer to Figure 4 the case of k = 3 shown. After step 303, the wavelet decomposition result of the first layer is obtained. As described above, the wavelet decomposition result of the first layer includes the 2×2 dimension LH wavelet coefficients, the 2×2 dimension HL wavelet coefficients, the 2×2 dimension HH wavelet coefficients, and the 2×2 dimension LL wavelet coefficients. Among them, the obtained one group of 2 k-2 ×2 k-2 dimension low-frequency wavelet coefficients (2×2 dimension LL wavelet coefficients) is 2 3-2 = 2 1For the low-frequency wavelet coefficients of the dimension, that is, k - 2 = 3 - 2 = 1 > 0, it belongs to the case where k - 1 > 0 in step 304. In this case, in step 304, Haar wavelet decomposition is performed on 1 group of low-frequency wavelet coefficients of 2×2 dimension, that is, Haar wavelet decomposition is performed on the 2×2 dimension LL wavelet coefficients, resulting in 3 groups of high-frequency wavelet coefficients of 1×1 dimension and 1 group of low-frequency wavelet coefficients of 1×1 dimension, that is, the wavelet decomposition result of the second layer. Among them, the 3 groups of high-frequency wavelet coefficients of 1×1 dimension are the 1×1 dimension LH wavelet coefficients, the 1×1 dimension HL wavelet coefficients, and the 1×1 dimension HH wavelet coefficients respectively, and the 1 group of low-frequency wavelet coefficients of 1×1 dimension is the 1×1 dimension LL wavelet coefficients. Among them, the 1×1 dimension only contains 1 number. Since the 1×1 dimension cannot be further decomposed by Haar wavelet, the Haar wavelet decomposition stops here.

[0091] In step 304, when k - 2 > 0, for 1 group of k-2 ×2 k-2 low-frequency wavelet coefficients of dimension, the purpose is to continue to extract the high-frequency components from the 2 k-2 ×2 k-2 dimension wavelet coefficients. In the Figure 4 illustrated embodiment, the 1×1 dimension LH wavelet coefficients represent the characteristics of the 2×2 dimension LL wavelet coefficients in the horizontal direction, which reflect the fluctuation of the 2×2 dimension LL wavelet coefficients in the horizontal direction; the 1×1 dimension HL wavelet coefficients represent the characteristics of the 2×2 dimension LL wavelet coefficients in the vertical direction, which reflect the fluctuation of the 2×2 dimension LL wavelet coefficients in the vertical direction; the 1×1 dimension HH wavelet coefficients represent the characteristics of the 2×2 dimension LL wavelet coefficients in the diagonal direction, which reflect the fluctuation of the 2×2 dimension LL wavelet coefficients in the diagonal direction; the 1×1 dimension LL wavelet coefficients represent the characteristics of the low-frequency part of the 2×2 dimension LL wavelet coefficients, representing the general outline of the change of the 2×2 dimension LL wavelet coefficients.

[0092] In the illustrative embodiment, in step 302, based on 3 groups of 2 k-1 ×2 k-1 dimension high-frequency wavelet coefficients, obtaining the high-frequency statistical values associated with 2 k-1 ×2 k-1 dimension, in step 303, based on 3 groups of 2 k-2 ×2 k-2 dimension high-frequency wavelet coefficients, obtaining the high-frequency statistical values associated with 2 k-2 ×2 k-2 dimension, and in step 304, based on 3 groups of 2 k-3 ×2 k-3 dimension high-frequency wavelet coefficients, obtaining the high-frequency statistical values associated with 2 k-3 ×2 k-3For the high-frequency statistical values of all dimensions, the respective high-frequency statistical values in steps 302, 303, and 304 are obtained using the same method.

[0093] Figure 5 It is a schematic flowchart showing the process of obtaining high-frequency statistical values from high-frequency wavelet coefficients according to an exemplary embodiment, as Figure 5 shown. For the 3 groups of 2 k-1 ×2 k-1 -dimensional high-frequency wavelet coefficients in steps 302, 303, and 304, the high-frequency statistical values associated with 2 k-1 ×2 k-1 dimensions are obtained, and for the 3 groups of 2 k-2 ×2 k-2 -dimensional high-frequency wavelet coefficients, the high-frequency statistical values associated with 2 k-2 ×2 k-2 dimensions are obtained, and for the 3 groups of 2 k-3 ×2 k-3 -dimensional high-frequency wavelet coefficients, the high-frequency statistical values associated with 2 k-3 ×2 k-3 dimensions are all obtained using the following process from step 501 to step 502.

[0094] Step 501: Among the 3 groups of 2 m ×2 m -dimensional high-frequency wavelet coefficients, by comparing each same position, the maximum value at each same position in the 3 groups of 2 m ×2 m -dimensional high-frequency wavelet coefficients is obtained;

[0095] Step 502: Based on the maximum value at each same position in the 3 groups of 2 m ×2 m -dimensional high-frequency wavelet coefficients, the high-frequency statistical values associated with 2 m ×2 m dimensions are obtained.

[0096] Among them, in the case of 2 k-1 ×2 k-1 dimensions, m = k - 1 (corresponding to the 3 groups of 2 k-1 ×2 k-1 -dimensional high-frequency wavelet coefficients obtained in step 301), in the case of 2 k-2 ×2 k-2 dimensions, m = k - 2 (corresponding to the 3 groups of 2 k-2 ×2 k-2 -dimensional high-frequency wavelet coefficients obtained in step 303), and in the case of 2 k-3 ×2 k-3 dimensions, m = k - 3 (corresponding to the 3 groups of 2k-3 ×2 k-3 (high-frequency wavelet coefficients of the dimension).

[0097] For the specific implementation of the above steps 501 to 502, refer to Figure 4 the graphical expression process of the frequency division operator shown in Figure 4 As shown, in the wavelet decomposition result of the 0th layer, among the 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, and 4×4 dimension HH wavelet coefficients, the values are taken in the order from left to right and from top to bottom. Assume that the values of each element in the 4×4 dimension LH wavelet coefficients are a11, a12, a13, a14, a21, a22... a43, a44 in sequence, the values of each element in the 4×4 dimension HL wavelet coefficients are b11, b12, b13, b14, b21, b22... b43, b44 in sequence, and the values of each element in the 4×4 dimension HH wavelet coefficients are c11, c12, c13, c14, c21, c22... c43, c44 in sequence. Then a11, b11, and c11 are the same positions (both in the 1st row and 1st column position) of the 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, and 4×4 dimension HH wavelet coefficients, a12, b12, and c12 are the same positions (both in the 1st row and 2nd column position) of the 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, and 4×4 dimension HH wavelet coefficients, and so on. Then in step 501, the maximum value at each same position in the 1st row and 1st column of the 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, and 4×4 dimension HH wavelet coefficients is obtained from a11, b11, and c11. In this way, in step 501, the maximum values from the 1st row and 1st column to the 4th row and 4th column in the 4×4 dimension LH wavelet coefficients, 4×4 dimension HL wavelet coefficients, and 4×4 dimension HH wavelet coefficients are obtained. The maximum values obtained in this way represent the maximum values of the high-frequency features in the 4×4 dimension wavelet coefficients.

[0098] As Figure 4As shown, in the wavelet decomposition result of the first layer, among the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients, the values are taken in the order from left to right and from top to bottom. Assume that the values of each element in the 2×2 - dimensional LH wavelet coefficients are d11, d12, d21, d22 in sequence, the values of each element in the 2×2 - dimensional HL wavelet coefficients are e11, e12, e21, e22 in sequence, and the values of each element in the 2×2 - dimensional HH wavelet coefficients are f11, f12, f21, f22 in sequence. Then d11, e11, and f11 are in the same position (both in the first row and the first column position) of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients, d12, e12, and f12 are in the same position (both in the first row and the second column position) of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients, d21, e21, and f21 are in the same position (both in the second row and the first column position) of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients, and d22, e22, and f22 are in the same position (both in the second row and the second column position) of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients. Then in step 501, the maximum value at the first row and the first column position of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients is obtained through d11, e11, and f11, the maximum value at the first row and the second column position of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients is obtained through d12, e12, and f12, the maximum value at the second row and the first column position of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients is obtained through d21, e21, and f21, and the maximum value at the second row and the second column position of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients is obtained through d22, e22, and f22. The maximum values at each same position of the 2×2 - dimensional LH wavelet coefficients, 2×2 - dimensional HL wavelet coefficients, and 2×2 - dimensional HH wavelet coefficients obtained in this way characterize the maximum values of the high - frequency features in the 2×2 - dimensional wavelet coefficients.

[0099] As Figure 4 shown, in the wavelet decomposition result of the second layer, among the 1×1 - dimensional LH wavelet coefficients, 1×1 - dimensional HL wavelet coefficients, and 1×1 - dimensional HH wavelet coefficients, there is only one element. So, only the maximum value of the 1×1 - dimensional LH wavelet coefficients, 1×1 - dimensional HL wavelet coefficients, and 1×1 - dimensional HH wavelet coefficients needs to be taken. The maximum value obtained in this way characterizes the maximum value of the high - frequency features in the 1×1 - dimensional wavelet coefficients.

[0100] In the illustrative embodiment, step 502 may specifically include: determining the average value of the maximum values at the same positions of each of the high-frequency wavelet coefficients of 3 groups of 2 m ×2 m dimensions as the high-frequency statistical value associated with 2 m ×2 m dimensions.

[0101] As Figure 4 shown, after obtaining the maximum values from the 1st row and 1st column to the 4th row and 4th column in the 4×4-dimensional LH wavelet coefficients, 4×4-dimensional HL wavelet coefficients, and 4×4-dimensional HH wavelet coefficients obtained in step 501, in step 502, the average value obtained by averaging the maximum values from the 1st row and 1st column to the 4th row and 4th column is determined as the high-frequency statistical value associated with 4×4 dimensions, that is, the high-frequency statistical value corresponding to the wavelet decomposition result of the 0th layer, which can be denoted as haar0.

[0102] As Figure 4 shown, after obtaining the maximum values from the 1st row and 1st column to the 2nd row and 2nd column in the 2×2-dimensional LH wavelet coefficients, 2×2-dimensional HL wavelet coefficients, and 2×2-dimensional HH wavelet coefficients obtained in step 501, in step 502, the average value obtained by averaging the maximum values from the 1st row and 1st column to the 2nd row and 2nd column is determined as the high-frequency statistical value associated with 2×2 dimensions, that is, the high-frequency statistical value corresponding to the wavelet decomposition result of the 1st layer, which can be denoted as haar1.

[0103] As Figure 4 shown, after obtaining the maximum value in the 1×1-dimensional LH wavelet coefficients, 1×1-dimensional HL wavelet coefficients, and 1×1-dimensional HH wavelet coefficients obtained in step 501, in step 502, since there is only 1 such maximum value and the average value is the maximum value itself, the maximum value is determined in step 502 as the high-frequency statistical value associated with 1×1 dimensions, that is, the high-frequency statistical value corresponding to the wavelet decomposition result of the 2nd layer, which can be denoted as haar2.

[0104] In the illustrative embodiment, step 305 may specifically include: determining the average value from the high-frequency statistical value associated with 2 k-1 ×2 k-1 dimensions to the high-frequency statistical value associated with 1×1 dimensions as the brightness frequency division operator of the current pixel point.

[0105] For example Figure 4 in the embodiment shown, the brightness frequency division operator of the current pixel point can be obtained by the following formula:

[0106] freq = (haar0 + haar1 + haar2) / 3

[0107] where freq is the luminance frequency division operator of the current pixel point.

[0108] Step 202: Obtain the first denoising strength of the current pixel point according to the mapping relationship between the luminance frequency division operator and a preset first denoising strength.

[0109] In the illustrative embodiment, the first denoising strength is related to the weight of the filter used. Therefore, the weight of the corresponding filter can be obtained from the first denoising strength obtained in step 202. Figure 6 is a schematic diagram of the mapping relationship between the luminance frequency division operator and the first denoising strength shown according to an illustrative embodiment. In the illustrative embodiment, according to Figure 6 the shown mapping relationship, when the luminance frequency division operator is in the range of 0 to x0, the first denoising strength is r3; when the luminance frequency division operator is in the range of x0 to x1, the first denoising strength linearly decreases from r3 to r2; when the luminance frequency division operator is in the range of x1 to x2, the first denoising strength is r2; when the luminance frequency division operator is in the range of x2 to x3, the first denoising strength linearly decreases from r2 to r1; when the luminance frequency division operator is in the range of x3 and above, the first denoising strength is r1. Generally speaking, the smaller the luminance frequency division operator, the larger the corresponding first denoising strength, and the larger the luminance frequency division operator, the smaller the corresponding first denoising strength. A region with a smaller luminance frequency division operator means that the change in luminance is relatively flat, and there may be some isolated noise points. In this case, using a larger first denoising strength can reduce the luminance difference between the isolated noise points and the surrounding pixel points; a region with a larger luminance frequency division operator means that the change in luminance is relatively drastic, indicating that it contains more image details. In this case, using a smaller first denoising strength can avoid eliminating too many image details.

[0110] Step 203: Perform spatial domain filtering on the luminance value of the current pixel point according to the first denoising strength of the current pixel point to obtain the luminance value after spatial domain denoising of the current pixel point. The luminance information after spatial domain denoising of the image is composed of the luminance values after spatial domain denoising of all pixel points in the image.

[0111] In the illustrative embodiment, the spatial domain filtering can adopt existing filtering methods, such as Gaussian filtering, bilateral filtering, or non-local mean filtering, etc.

[0112] In the illustrative embodiment, when performing spatial domain filtering on the luminance value of the current pixel point, a filtering window with a size greater than or equal to 3 is used, that is, a filtering window centered on the current pixel point and containing at least 3×3 neighboring pixel points is used for spatial domain filtering. The larger the filtering window, the better the denoising effect, but the corresponding hardware implementation cost also increases. The appropriate size of the filtering window can be selected according to actual requirements and design costs.

[0113] Figure 7It is a schematic flowchart of performing spatial domain filtering on the brightness value of the current pixel point shown according to a schematic embodiment. As Figure 7 shown, in the schematic embodiment, step 203 may specifically include the following steps 701 to step 702.

[0114] Step 701: Within a preset filtering window, adjust a preset initial filter according to a first denoising intensity to obtain a spatial domain filter for performing spatial domain filtering on the current pixel point.

[0115] In the schematic embodiment, the size of the filtering window may be 3×3, 5×5 or larger. In the schematic embodiment, the adjustment of the initial filter is to adjust the weights of the initial filter. By adjusting the weights of the initial filter, the denoising intensity of the obtained spatial domain filter can approach or be equal to the first denoising intensity.

[0116] Step 702: Use the spatial domain filter to perform spatial domain filtering on the current pixel point to obtain the brightness value of the current pixel point after spatial domain denoising.

[0117] In the schematic embodiment, spatial domain filtering can be implemented by methods such as Gaussian filtering, bilateral filtering, and non-local means filtering. Among them, the spatial domain filters used corresponding to different filtering methods are different. For example, when using the Gaussian filtering method, the spatial domain filter is a Gaussian filter, and the Gaussian filter is a Gaussian kernel (weight matrix). Gaussian filtering uses the Gaussian kernel to perform weighted averaging on the pixel points in the neighborhood of the current pixel point, and the calculation of the weights only considers the distance between pixel points and does not consider the difference in pixel values (such as brightness values). For example, when using the bilateral filtering method, the spatial domain filter is a bilateral filter, and the bilateral filter is a bilateral kernel, which considers both the spatial distance between pixel points and the difference in pixel values (such as brightness values). In addition to the distance between pixel points, the calculation of the weights also considers the similarity between pixel points, that is, the difference in pixel values. Therefore, bilateral filtering can preserve edge information and details.

[0118] Step 103: Obtain the first brightness residual information of the image according to the brightness information of the image before denoising and the brightness information of the image after spatial domain denoising.

[0119] In the schematic embodiment, step 103 may specifically include the following steps performed on each pixel point in the image:

[0120] Determine the difference between the brightness value of the current pixel point before denoising and the brightness value of the current pixel point after spatial domain denoising as the first brightness residual value of the current pixel point, where the current pixel point is any pixel point in the image.

[0121] In the embodiments of the present disclosure, the pre-denoising luminance information of the image is composed of the pre-denoising luminance values of all pixel points in the image, the post-spatial-domain denoising luminance information of the image is composed of the post-spatial-domain denoising luminance values of all pixel points in the image, and the first luminance residual information of the image is composed of the first luminance residual values of all pixel points in the image.

[0122] Step 104: Perform frequency-domain filtering on the first luminance residual information of the image to obtain the second luminance residual information of the image.

[0123] In the exemplary embodiments, existing frequency-domain filtering methods can be adopted, such as Fourier transform filtering, Haar filtering, etc. The embodiments of the present disclosure provide a method for performing frequency-domain filtering using Haar wavelet decomposition, and the specific details are as follows.

[0124] Figure 8 According to the schematic flowchart of performing frequency-domain filtering on the first luminance residual information of the image shown in an exemplary embodiment, as Figure 8 shown, in the exemplary embodiment, step 104 may specifically include the following steps 801 to 804 performed on each pixel point in the image.

[0125] Step 801: In the first luminance residual plane formed by the first luminance residual information of the image, perform Haar wavelet decomposition to obtain the high-frequency luminance residual wavelet coefficients and the low-frequency luminance residual wavelet coefficients of the current pixel point.

[0126] In the exemplary embodiment, step 801 can be implemented by the method of step 301 for obtaining high-frequency wavelet coefficients and low-frequency wavelet coefficients as described above. For example, in the first luminance residual plane, perform Haar wavelet decomposition on the first luminance residual values of the pixel points within the 2 k ×2 k neighborhood range around the current pixel point to obtain 3 groups of 2 k-1 ×2 k-1 dimensional high-frequency wavelet coefficients and 1 group of 2 k-1 ×2 k-1 dimensional low-frequency wavelet coefficients, where k≥2. In a specific example, k = 3. Refer to the schematic diagram of the graphic expression process of the frequency division operator shown in Figure 4 to understand the process of obtaining the high-frequency luminance residual wavelet coefficients and the low-frequency luminance residual wavelet coefficients of the current pixel point in step 801. For example, Figure 4If the 8×8 neighborhood range around the current pixel on the left side in the middle is regarded as the first luminance residual plane, then the wavelet decomposition result of the 0th layer obtained through wavelet decomposition is the high-frequency luminance residual wavelet coefficient and the low-frequency luminance residual wavelet coefficient of the current pixel. Among them, the high-frequency luminance residual wavelet coefficient of the current pixel includes the 4×4 dimensional LH luminance residual wavelet coefficient, the 4×4 dimensional HL luminance residual wavelet coefficient, and the 4×4 dimensional HH luminance residual wavelet coefficient.

[0127] Step 802: Obtain the second denoising intensity of the current pixel according to the mapping relationship between the luminance frequency division operator of the current pixel and the preset second denoising intensity.

[0128] Figure 9 is a schematic diagram of the mapping relationship between the luminance frequency division operator and the second denoising intensity shown according to an exemplary embodiment. Similar to Figure 6 the mapping relationship between the luminance frequency division operator and the first denoising intensity shown, the change trend between the luminance frequency division operator and the second denoising intensity is that the smaller the luminance frequency division operator, the greater the corresponding second denoising intensity, and the larger the luminance frequency division operator, the smaller the corresponding second denoising intensity. This is because a region with a smaller luminance frequency division operator means that the change in luminance is relatively flat, and the isolated noise points eliminated by the previous spatial domain filtering can be further weakened rather than strengthened in the frequency domain filtering. While a region with a larger luminance frequency division operator means that it contains more image details, and image details may be lost after the previous spatial domain filtering, and should not be further weakened in the frequency domain filtering but can be strengthened to restore the image details.

[0129] In the exemplary embodiment, according to Figure 9 the mapping relationship shown, when the luminance frequency division operator is in the range of 0 to x0’, the second denoising intensity is r3’. When the luminance frequency division operator is in the range of x0’ to x1’, the second denoising intensity linearly decreases from r3’ to r2’. When the luminance frequency division operator is in the range of x1’ to x2’, the second denoising intensity is r2’. When the luminance frequency division operator is in the range of x2’ to x3’, the second denoising intensity linearly decreases from r2’ to r1’. When the luminance frequency division operator is in the range of x3’ and above, the second denoising intensity is r1’. It should be noted that Figure 9 and Figure 6 the mapping relationship shown is only an exemplary illustration. In practical applications, the mapping relationship can be adaptively set according to needs.

[0130] In the exemplary embodiment, in cooperation with subsequent steps, the second denoising intensity can be expressed as a shrinkage coefficient.

[0131] Step 803: Shrink the high-frequency luminance residual wavelet coefficient of the current pixel by using the second denoising intensity of the current pixel to obtain the shrunk high-frequency luminance residual wavelet coefficient of the current pixel.

[0132] In an exemplary embodiment, step 803 may specifically include multiplying the high-frequency brightness residual wavelet coefficient of the current pixel by the shrinkage coefficient to determine the high-frequency brightness residual wavelet coefficient of the current pixel after shrinkage. Figure 4 As shown, the high-frequency brightness residual wavelet coefficients of the current pixel point include 4×4-dimensional LH brightness residual wavelet coefficients, 4×4-dimensional HL brightness residual wavelet coefficients and 4×4-dimensional HH brightness residual wavelet coefficients. In step 803, each element of the 4×4-dimensional LH brightness residual wavelet coefficients is multiplied by the contraction coefficient to obtain the contracted 4×4-dimensional LH brightness residual wavelet coefficients, each element of the 4×4-dimensional HL brightness residual wavelet coefficients is multiplied by the contraction coefficient to obtain the contracted 4×4-dimensional HL brightness residual wavelet coefficients, and each element of the 4×4-dimensional HH brightness residual wavelet coefficients is multiplied by the contraction coefficient to obtain the contracted 4×4-dimensional HH brightness residual wavelet coefficients.

[0133] It should be noted that the low-frequency brightness residual wavelet coefficients are not processed because they represent the characteristics of the low-frequency part, while the denoising process is performed on the high-frequency characteristics, and the low-frequency characteristic part can remain unchanged.

[0134] Step 804: reconstruct the current pixel in the frequency domain according to the shrunk high-frequency brightness residual wavelet coefficient and the low-frequency brightness residual wavelet coefficient of the current pixel to obtain a second brightness residual value of the current pixel.

[0135] The second brightness residual information of the image is composed of the second brightness residual values ​​of all pixels in the image.

[0136] Among them, frequency domain reconstruction refers to the inverse process of the haar wavelet decomposition of step 801 above. Because the current pixel is subjected to frequency domain denoising after steps 801 to 803, the brightness residual value (i.e., the second brightness residual value) after the current pixel is restored by the frequency domain reconstruction of step 804 is a brightness residual value after the denoising effect is further improved. In an exemplary embodiment, frequency domain reconstruction may refer to wavelet reconstruction. In step 804, the second brightness residual value of the current pixel is obtained by haar wavelet reconstruction based on the high-frequency brightness residual wavelet coefficients and low-frequency brightness residual wavelet coefficients of the current pixel after contraction.

[0137] Step 105 : Obtain denoised brightness information of the image according to the denoised brightness information of the image in the spatial domain and the second brightness residual information of the image.

[0138] The brightness information of the denoised image is the sum of the brightness information of the image after spatial denoising and the second brightness residual information of the image.

[0139] In a schematic embodiment, step 105 may include adding the denoised luminance value of each pixel point in the image and the second luminance residual value to obtain the denoised luminance value of each pixel point. The luminance information of the denoised image is composed of the denoised luminance values of all pixel points.

[0140] Since the noise in the image is mainly presented in the obvious difference between the luminance value and other surrounding pixel points, therefore, the above dual denoising means of spatial domain denoising and frequency domain denoising can significantly eliminate luminance noise. In a YUV format image, when the U and V values of all pixel points are 0, the image appears as a black-and-white image (grayscale image), and each pixel point in the image is presented by the luminance (Y value). When the U and V values of all pixel points are not all 0, each pixel point in the image is presented not only by the luminance (Y value) but also by the color (determined by the U and V values). For noise, it is mainly the most obvious in the luminance (Y value) expression and relatively weak in the color expression. Therefore, in an alternative embodiment, for a color YUV format image, the U channel and the V channel may not be denoised. However, in order to achieve a better denoising effect and eliminate the noise of obvious color differences, the U channel and the V channel may also be denoised. However, because the noise is relatively weak in the color expression, only spatial domain denoising may be performed on the U channel and the V channel without performing frequency domain denoising. Thereby, the consumption of software and hardware resources in the image denoising process can be reduced and the denoising effect can be improved, and compared with also performing spatial domain denoising and frequency domain denoising on the U channel and the V channel, the difference in the denoising effect is very small.

[0141] Figure 10 is a schematic diagram of the chrominance denoising process shown according to a schematic embodiment, as Figure 10 shown, in a schematic embodiment, the image denoising method of the embodiments of the present disclosure may further include the following steps 1001 to step 1003.

[0142] Step 1001: Obtain the chrominance information before denoising from the image.

[0143] Among them, the chrominance information before denoising includes the U channel information before denoising and the V channel information before denoising.

[0144] Step 1002: Perform spatial domain filtering on the chrominance information before denoising of the image to obtain the spatial domain denoised chrominance information of the image.

[0145] In a schematic embodiment, performing spatial domain filtering on the chrominance information before denoising of the image may be the same as the process of performing spatial domain filtering on the luminance information before denoising of the image in step 102 above. It should be noted that step 1002 includes performing spatial domain filtering on the U channel information before denoising and the V channel information before denoising of the image respectively to obtain the U channel information after spatial domain denoising and the V channel information after spatial domain denoising of the image.

[0146] Step 1003: Determine the chrominance information after spatial domain denoising as the chrominance information of the denoised image.

[0147] In an illustrative embodiment, step 1003 includes respectively determining the U-channel information after spatial domain denoising and the V-channel information after spatial domain denoising as the U-channel information of the denoised image and the V-channel information of the denoised image.

[0148] In an illustrative embodiment, a sliding window method can be used to traverse each pixel point in the image point by point for denoising operations, including obtaining a luminance frequency division operator, spatial domain filtering, frequency domain filtering, etc. for each pixel point point by point.

[0149] The image denoising method of the embodiments of the present disclosure combines spatial domain denoising and frequency domain denoising methods to achieve a better image denoising effect. During the denoising process, denoising is mainly performed based on the luminance information of the image, thereby capturing the feature that the luminance of the noise points is significantly different from the luminance of other surrounding pixel points, which can help improve the denoising efficiency. According to the technical solution of the embodiments of the present disclosure, during the denoising process, the means of Haar wavelet decomposition is mainly used to assist spatial domain denoising and frequency domain denoising. Therefore, compared with denoising methods such as Fourier transform means, the denoising efficiency can be significantly improved, and it helps to reduce the hardware cache requirements during the denoising process, which is conducive to popularization and application in miniaturized image processing devices.

[0150] The following describes the implementation process of the image denoising method of the embodiments of the present disclosure in combination with a specific application scenario.

[0151] Figure 11 is a schematic diagram of the steps of a specific application scenario of the image denoising method shown according to an illustrative embodiment. Figure 12 is a schematic flowchart of a specific application scenario of the image denoising method shown according to an illustrative embodiment. This application scenario is described by taking a YUV400 image as an example, as Figure 11 and Figure 12 shown. This application scenario mainly includes the following steps 1101 to step 1108.

[0152] Step 1101: Obtain the original image, preprocess the original image to obtain a YUV format image, and then proceed to step 1102.

[0153] In an illustrative embodiment, the preprocessing may include operations such as geometric correction, luminance correction, color correction, etc. of the original image, and may also include image convolution or window operations. In order to process the edge pixel points of the image in subsequent steps, the preprocessing may also include an edge extension operation at the positions of the four boundary points of the image. In an illustrative embodiment, if the original image format is not the YUV format, the preprocessing also includes an image format conversion process.

[0154] In this application scenario, the YUV format image is YUV400, which does not contain U and V channel information. Therefore, the denoising of U and V channel information will not be elaborated here. For the denoising of U and V channel information, refer to the spatial domain denoising process of Y channel information.

[0155] In step 1101, a processed YUV format image is obtained, and the Y channel values (including the Y values of each pixel) of the processed image are stored in the array image[i, j], where i and j are the row number and column number of the pixel. The array image[i, j] is the luminance information of the image.

[0156] Step 1102: Perform local texture feature analysis on the luminance of each pixel in the image to obtain the luminance frequency division operator for each pixel, and then proceed to step 1103.

[0157] Among them, the luminance frequency division operator reflects the local texture feature of the pixel. The larger the luminance frequency division operator, the more details of the local texture feature of the pixel; conversely, the smaller the luminance frequency division operator, the fewer details of the local texture feature of the pixel.

[0158] In the illustrative embodiment, the local texture feature analysis of the luminance of each pixel in step 1102 is implemented through Haar wavelet decomposition. For the specific implementation process, refer to the relevant description of step 201 above, and it will not be elaborated here.

[0159] Step 1103: Traverse each pixel's luminance in the image point by point, perform 5×5 bilateral filtering denoising to obtain a spatial domain filtered image, and then proceed to step 1104.

[0160] Among them, bilateral filtering is spatial domain filtering, and the filtering intensity is adaptively controlled according to the luminance frequency division operator of the current pixel. In the illustrative embodiment, the filtering intensity (equivalent to the first denoising intensity) can be adaptively controlled according to Figure 6 the shown mapping relationship. The spatial domain filtered image can be represented by the array MNR[i, j]. The spatial domain filtered image is equivalent to the luminance information after spatial domain denoising in the above embodiment.

[0161] Step 1104: Obtain a spatial domain residual image based on the denoising result and the luminance information of the image, and then proceed to step 1105.

[0162] Among them, the spatial domain residual image is represented by the array res[i, j], then step 1104 is expressed by the formula:

[0163] res[i, j] = image[i, j] - MNR[i, j]

[0164] Among them, the spatial domain residual image is equivalent to the first luminance residual information in the above description.

[0165] Step 1105: Perform Haar wavelet decomposition on the spatial domain residual image to obtain the high-frequency wavelet coefficients and low-frequency coefficients of each pixel point in the spatial domain residual image, and then perform Step 1106.

[0166] In the exemplary embodiment, in Step 1105, Haar wavelet decomposition is performed on each pixel point in the spatial domain residual image with an 8×8 window to obtain three groups of high-frequency wavelet coefficients LH, HL, HH with a dimension of 4×4 and one group of low-frequency wavelet coefficients LL with a dimension of 4×4 for each pixel point.

[0167] In the exemplary embodiment, in order to perform Haar wavelet decomposition on the edge pixel points of the spatial domain residual image, in Step 1105, it may further include an edge extension operation performed on the spatial domain residual image before Haar wavelet decomposition.

[0168] Step 1106: Shrink the high-frequency wavelet coefficients of each pixel point according to the luminance frequency division operator of each pixel point, and then perform Step 1107.

[0169] In the exemplary embodiment, in Step 1106, it is possible to Figure 9 obtain the shrinkage coefficient (the second denoising intensity) corresponding to the luminance frequency division operator of each pixel point according to the mapping relationship shown, and then shrink the high-frequency wavelet coefficients according to the obtained shrinkage coefficient to obtain the shrunk high-frequency wavelet coefficients of each pixel point. Continuing with the Haar wavelet decomposition with an 8×8 window performed in the above exemplary embodiment, in Step 1106, for each pixel point, the three groups of high-frequency wavelet coefficients LH, HL, HH with a dimension of 4×4 are shrunk according to the shrinkage coefficient, and the low-frequency wavelet coefficients LL with a dimension of 4×4 remain unchanged. As Figure 9 shown, the mapping relationship is that the larger the luminance frequency division operator, the smaller the shrinkage coefficient (the second denoising intensity). In the exemplary embodiment, the means of shrinking the three groups of high-frequency wavelet coefficients LH, HL, HH with a dimension of 4×4 includes: multiplying the shrinkage coefficient by the three groups of high-frequency wavelet coefficients LH, HL, HH with a dimension of 4×4 to obtain the shrunk high-frequency wavelet coefficients LH, the shrunk high-frequency wavelet coefficients HL, and the shrunk high-frequency wavelet coefficients HH.

[0170] Step 1107: Perform frequency domain reconstruction on the image according to the shrunk high-frequency wavelet coefficients and the unchanged low-frequency wavelet coefficients of each pixel point to obtain a residual filtered image, and then perform Step 1108.

[0171] In a schematic embodiment, the execution process of frequency-domain reconstruction is specifically the inverse process of Haar wavelet decomposition, i.e., the inverse Haar transform. The resulting residual filtered image is the second luminance residual information in the above description. The residual filtered image can be represented by the array res_denoise[i,j].

[0172] The process from step 1105 to step 1107 is a frequency-domain filtering process for the spatial-domain residual image.

[0173] Step 1108: Superimpose the spatially filtered image and the residual filtered image to obtain the denoised image.

[0174] In a schematic embodiment, if the denoised image is represented by the array image_denoise[i,j], then step 1108 can obtain the denoised image through the following formula:

[0175] image_denoise[i,j] = MNR[i,j] + res_denoise[i,j]

[0176] Thus, the denoising process for this specific application scenario is completed.

[0177] Figure 13 It is a schematic diagram of the logical structure of an image denoising device shown according to a schematic embodiment. As Figure 13 shown, the image denoising device mainly includes an image information extraction module 1301, a spatial-domain filtering module 1302, a first luminance residual obtaining module 1303, a second luminance residual obtaining module 1304, and a denoised image obtaining module 1305.

[0178] Among them, the image information extraction module 1301 is configured to execute obtaining an image and obtaining the pre-denoising luminance information from the image. The spatial-domain filtering module 1302 is configured to execute spatial-domain filtering on the pre-denoising luminance information of the image to obtain the post-spatial-domain denoising luminance information of the image. The first luminance residual obtaining module 1303 is configured to execute obtaining the first luminance residual information of the image according to the pre-denoising luminance information of the image and the post-spatial-domain denoising luminance information of the image. The second luminance residual obtaining module 1304 is configured to execute frequency-domain filtering on the first luminance residual information of the image to obtain the second luminance residual information of the image. The denoised image obtaining module 1305 is configured to execute obtaining the post-denoising luminance information of the image according to the post-spatial-domain denoising luminance information of the image and the second luminance residual information of the image, where the luminance information of the post-denoised image is the sum of the post-spatial-domain denoising luminance information of the image and the second luminance residual information of the image.

[0179] In a schematic embodiment, the spatial-domain filtering module 1302 includes the following sub-modules that perform operations on each pixel point in the image:

[0180] The frequency division operator obtaining sub-module is configured to perform, in the luminance plane of the image, obtaining the luminance frequency division operator of the current pixel point through Haar wavelet decomposition, where the current pixel point is any pixel point in the image;

[0181] The first denoising intensity obtaining sub-module is configured to perform obtaining the first denoising intensity of the current pixel point according to the mapping relationship between the luminance frequency division operator and a preset first denoising intensity;

[0182] The spatial domain filtering sub-module is configured to perform spatially filtering the luminance value of the current pixel point according to the first denoising intensity of the current pixel point to obtain the spatially denoised luminance value of the current pixel point, and the spatially denoised luminance information of the image is composed of the spatially denoised luminance values of all pixel points in the image.

[0183] In the illustrative embodiment, the frequency division operator obtaining sub-module is further configured to perform:

[0184] In the luminance plane of the image, performing Haar wavelet decomposition on the luminance values of the pixel points within a 2 k ×2 k neighborhood range around the current pixel point to obtain 3 groups of 2 k-1 ×2 k-1 -dimensional high-frequency wavelet coefficients and 1 group of 2 k-1 ×2 k-1 -dimensional low-frequency wavelet coefficients, where k≥2;

[0185] According to the 3 groups of 2 k-1 ×2 k-1 -dimensional high-frequency wavelet coefficients, obtaining the high-frequency statistical value associated with 2 k-1 ×2 k-1 -dimensions;

[0186] In the case where k - 1>0, performing Haar wavelet decomposition on 1 group of 2 k-1 ×2 k-1 -dimensional low-frequency wavelet coefficients to obtain 3 groups of 2 k-2 ×2 k-2 -dimensional high-frequency wavelet coefficients and 1 group of 2 k-2 ×2 k-2 -dimensional low-frequency wavelet coefficients, and according to the 3 groups of 2 k-2 ×2 k-2 -dimensional high-frequency wavelet coefficients, obtaining the high-frequency statistical value associated with 2 k-2 ×2 k-2 -dimensions;

[0187] In the case where k - 2>0, performing Haar wavelet decomposition on 1 group of 2 k-2 ×2 k-2 -dimensional low-frequency wavelet coefficients to obtain 3 groups of 2k-3 ×2 k-3 The high-frequency wavelet coefficients of the dimension and 1 group of 2 k-3 ×2 k-3 The low-frequency wavelet coefficients of the dimension, according to 3 groups of 2 k-3 ×2 k-3 The high-frequency wavelet coefficients of the dimension, obtain the correlation with 2 k-3 ×2 k-3 The high-frequency statistical value of the dimension;

[0188] And so on, until obtaining the high-frequency statistical value associated with the 1×1 dimension;

[0189] According to the correlation from 2 k-1 ×2 k-1 The high-frequency statistical value of the dimension to the high-frequency statistical value associated with the 1×1 dimension, obtain the luminance frequency division operator of the current pixel point.

[0190] In the exemplary embodiment, the frequency division operator obtaining sub-module is further configured to execute:

[0191] Among the high-frequency wavelet coefficients of 3 groups of 2 m ×2 m The dimension, by comparing each same position, obtain 3 groups of 2 m ×2 m The maximum value at each same position in the high-frequency wavelet coefficients of the dimension;

[0192] According to 3 groups of 2 m ×2 m The maximum value at each same position in the high-frequency wavelet coefficients of the dimension, obtain the correlation with 2 m ×2 m The high-frequency statistical value of the dimension;

[0193] Wherein, in the case of 2 k-1 ×2 k-1 The dimension, m = k - 1, in 2 k-2 ×2 k-2 The dimension, m = k - 2, in 2 k-3 ×2 k-3 The dimension, m = k - 3.

[0194] In the exemplary embodiment, the frequency division operator obtaining sub-module is further configured to execute:

[0195] Take the average value of the high-frequency statistical values from the correlation with 2 k-1 ×2 k-1 The dimension to the high-frequency statistical value associated with the 1×1 dimension, and determine it as the luminance frequency division operator of the current pixel point.

[0196] In the exemplary embodiment, the spatial domain filtering sub-module includes:

[0197] An adjustment sub-module, configured to perform an adjustment on a preset initialization filter according to a first denoising intensity within a preset filtering window to obtain a spatial domain filter for performing spatial domain filtering on the current pixel point;

[0198] A filtering sub-module, configured to perform spatial domain filtering on the current pixel point by using the spatial domain filter to obtain the brightness value after spatial domain denoising of the current pixel point.

[0199] In a schematic embodiment, the first brightness residual obtaining module 1303 is further configured to perform the following operations on each pixel point in the image:

[0200] Determine the difference between the brightness value before denoising of the current pixel point and the brightness value after spatial domain denoising of the current pixel point as the first brightness residual value of the current pixel point, where the current pixel point is any pixel point in the image;

[0201] Wherein, the brightness information before denoising of the image is composed of the brightness values before denoising of all pixel points in the image, the brightness information after spatial domain denoising of the image is composed of the brightness values after spatial domain denoising of all pixel points in the image, and the first brightness residual information of the image is composed of the first brightness residual values of all pixel points in the image.

[0202] In a schematic embodiment, the second brightness residual obtaining module 1304 further includes the following sub-modules for performing operations on each pixel point in the image:

[0203] A wavelet decomposition sub-module, configured to perform Haar wavelet decomposition in the first brightness residual plane formed by the first brightness residual information of the image to obtain the high-frequency brightness residual wavelet coefficients and low-frequency brightness residual wavelet coefficients of the current pixel point;

[0204] A second denoising intensity obtaining sub-module, configured to obtain the second denoising intensity of the current pixel point according to the mapping relationship between the brightness frequency division operator of the current pixel point and a preset second denoising intensity;

[0205] A wavelet coefficient shrinking sub-module, configured to perform shrinking on the high-frequency brightness residual wavelet coefficients of the current pixel point by using the second denoising intensity of the current pixel point to obtain the shrunk high-frequency brightness residual wavelet coefficients of the current pixel point;

[0206] A frequency domain reconstruction sub-module, configured to perform frequency domain reconstruction on the current pixel point according to the shrunk high-frequency brightness residual wavelet coefficients and low-frequency brightness residual wavelet coefficients of the current pixel point to obtain the second brightness residual value of the current pixel point;

[0207] Wherein, the second brightness residual information of the image is composed of the second brightness residual values of all pixel points in the image.

[0208] In an exemplary embodiment, the image denoising device further includes:

[0209] A chrominance acquisition module configured to obtain chrominance information before denoising from an image;

[0210] A chrominance spatial domain filtering module configured to perform spatial domain filtering on the chrominance information before denoising of the image to obtain spatial domain denoised chrominance information of the image;

[0211] A denoised chrominance determination module configured to determine the spatial domain denoised chrominance information as the chrominance information after denoising of the image.

[0212] All of the above optional technical solutions can be combined arbitrarily to form optional embodiments of the present disclosure, which will not be elaborated one by one here.

[0213] Regarding the image denoising device in the above embodiments, the specific manners in which each unit performs operations have been described in detail in the embodiments related to the image denoising method, and will not be elaborated in detail here.

[0214] It should be noted that: the above embodiments are only illustrated by the division of the above functional modules. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0215] Figure 14 It is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. In some embodiments, the electronic device is a server. The electronic device 1400 may vary greatly due to configuration or performance differences, and may include one or more processors (Central Processing Units, CPUs) 1401 and one or more memories 1402. Among them, at least one program code is stored in the memory 1402, and the at least one program code is loaded and executed by the processor 1401 to implement the image denoising method provided by each of the above embodiments. Of course, the electronic device 1400 may also have components such as wired or wireless network interfaces, keyboards, and input / output interfaces for input / output. The electronic device 1400 may also include other components for implementing device functions, which will not be elaborated here.

[0216] In an exemplary embodiment, a computer-readable storage medium including at least one instruction is also provided, such as a memory including at least one instruction. The at least one instruction can be executed by a processor in a computer device to complete the image denoising method in the above embodiments.

[0217] Optionally, the above computer-readable storage medium may be a non-transitory computer-readable storage medium. For example, the non-transitory computer-readable storage medium may include ROM (Read-Only Memory), RAM (Random-Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic tape, floppy disk, and optical data storage device, etc.

[0218] The foregoing is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included within the scope of protection of the present disclosure.

Claims

1. An image denoising method, comprising: Obtain an image and get the brightness information before denoising from the image; Perform spatial domain filtering on the brightness information before denoising of the image to obtain the brightness information after spatial domain denoising of the image; Obtain the first brightness residual information of the image according to the brightness information before denoising of the image and the brightness information after spatial domain denoising of the image; Perform frequency domain filtering on the first brightness residual information of the image to obtain the second brightness residual information of the image; Obtain the brightness information after denoising of the image according to the brightness information after spatial domain denoising of the image and the second brightness residual information of the image.

2. The image denoising method according to claim 1, wherein The step of performing spatial domain filtering on the brightness information before denoising of the image to obtain the brightness information after spatial domain denoising of the image includes the following steps performed on each pixel point in the image: In the brightness plane of the image, obtain the brightness frequency division operator of the current pixel point through Haar wavelet decomposition, where the current pixel point is any pixel point in the image; Obtain the first denoising intensity of the current pixel point according to the mapping relationship between the brightness frequency division operator and a preset first denoising intensity; Perform spatial domain filtering on the brightness value of the current pixel point according to the first denoising intensity of the current pixel point to obtain the brightness value after spatial domain denoising of the current pixel point.

3. The image denoising method according to claim 2, wherein The step of obtaining the brightness frequency division operator of the current pixel point by performing Haar wavelet decomposition in the brightness plane of the image includes: In the luminance plane of the said image, perform Haar wavelet decomposition on the luminance values of the pixel points within the range of the 2 k ×2 k neighborhood around the current pixel point, to obtain 3 sets of high-frequency wavelet coefficients of 2 k-1 ×2 k-1 dimensions and 1 set of low-frequency wavelet coefficients of 2 k-1 ×2 k-1 dimensions, where k≥2; According to the high-frequency wavelet coefficients of the three groups of 2 k-1 ×2 k-1 dimensions, obtain the high-frequency statistical values associated with 2 k-1 ×2 k-1 dimensions; In the case where k - 1>0, for the 1 group of 2 k-1 ×2 k-1 dimensional low-frequency wavelet coefficients, perform Haar wavelet decomposition to obtain 3 groups of 2 k-2 ×2 k-2 dimensional high-frequency wavelet coefficients and 1 group of 2 k-2 ×2 k-2 dimensional low-frequency wavelet coefficients. According to the 3 groups of 2 k-2 ×2 k-2 dimensional high-frequency wavelet coefficients, obtain the high-frequency statistical value associated with 2 k-2 ×2 k-2 dimensions; In the case of k - 2 > 0, for the 1 group of 2 k-2 ×2 k-2 dimensional low - frequency wavelet coefficients, perform Haar wavelet decomposition to obtain 3 groups of 2 k-3 ×2 k-3 dimensional high - frequency wavelet coefficients and 1 group of 2 k-3 ×2 k-3 dimensional low - frequency wavelet coefficients. According to the 3 groups of 2 k-3 ×2 k-3 dimensional high - frequency wavelet coefficients, obtain the high - frequency statistical value associated with 2 k-3 ×2 k-3 dimensions; And so on until obtaining the high-frequency statistical value associated with the 1×1 dimension; According to the high-frequency statistical value associated with 2 k-1 ×2 k-1 dimension to the high-frequency statistical value associated with 1×1 dimension, the brightness frequency division operator of the current pixel point is obtained.

4. The image denoising method according to claim 3, wherein The high-frequency wavelet coefficients of the said 3 groups of 2 k-1 ×2 k-1 dimensions are used to obtain the high-frequency statistical values associated with 2 k-1 ×2 k-1 dimensions. Moreover, the high-frequency wavelet coefficients of the said 3 groups of 2 k-2 ×2 k-2 dimensions are used to obtain the high-frequency statistical values associated with 2 k-2 ×2 k-2 dimensions. Additionally, the high-frequency wavelet coefficients of the said 3 groups of 2 k-3 ×2 k-3 dimensions are used to obtain the high-frequency statistical values associated with 2 k-3 ×2 k-3 dimensions, all of which include: Among the high-frequency wavelet coefficients of 3 groups of 2 m ×2 m dimensions, by comparing each same position, 3 groups of 2 m ×2 m the maximum value at each same position among the high-frequency wavelet coefficients of dimensions; According to the maximum value at the same position of each of the high-frequency wavelet coefficients of the three groups of 2 m ×2 m dimensions, the high-frequency statistical value associated with 2 m ×2 m dimensions is obtained; Among them, at 2 k-1 ×2 k-1 dimensions, m = k - 1, at 2 k-2 ×2 k-2 dimensions, m = k - 2, at 2 k-3 ×2 k-3 dimensions, m = k - 3.

5. The image denoising method according to claim 3, wherein The one obtained from the high-frequency statistical value associated with 2 k-1 ×2 k-1 dimension to the high-frequency statistical value associated with the 1×1 dimension to obtain the luminance frequency division operator of the current pixel point, comprising: The average value from the high-frequency statistical value associated with 2 k-1 ×2 k-1 to the high-frequency statistical value associated with 1×1 dimension is determined as the brightness frequency division operator of the current pixel point.

6. The image denoising method according to claim 2, wherein The step of performing spatial domain filtering on the brightness value of the current pixel point according to the first denoising intensity of the current pixel point to obtain the brightness value after spatial domain denoising of the current pixel point includes: In a preset filtering window, adjust a preset initial filter according to the first denoising intensity to obtain a spatial domain filter for performing spatial domain filtering on the current pixel point; Perform spatial domain filtering on the current pixel point by using the spatial domain filter to obtain the brightness value after spatial domain denoising of the current pixel point.

7. The image denoising method according to claim 1, wherein The step of obtaining the first brightness residual information of the image according to the brightness information before denoising of the image and the brightness information after spatial domain denoising of the image includes the following steps performed on each pixel point in the image: Determine the difference between the brightness value before denoising of the current pixel point and the brightness value after spatial domain denoising of the current pixel point as the first brightness residual value of the current pixel point, where the current pixel point is any pixel point in the image.

8. The image denoising method according to claim 2, wherein The step of performing frequency domain filtering on the first brightness residual information of the image to obtain the second brightness residual information of the image includes the following steps performed on each pixel point in the image: In the first brightness residual plane formed by the first brightness residual information of the image, obtain the high-frequency brightness residual wavelet coefficient and the low-frequency brightness residual wavelet coefficient of the current pixel point through Haar wavelet decomposition; Obtain the second denoising intensity of the current pixel point according to the mapping relationship between the brightness frequency division operator of the current pixel point and a preset second denoising intensity; Shrink the high-frequency luminance residual wavelet coefficients of the current pixel using the second denoising intensity of the current pixel to obtain the shrunk high-frequency luminance residual wavelet coefficients of the current pixel; Perform frequency-domain reconstruction on the current pixel according to the shrunk high-frequency luminance residual wavelet coefficients and the low-frequency luminance residual wavelet coefficients of the current pixel to obtain the second luminance residual value of the current pixel; Among them, the second luminance residual information of the image is composed of the second luminance residual values of all pixels in the image.

9. The image denoising method according to claim 1, wherein, The image denoising method further includes: Obtain the pre-denoising chrominance information from the image; Perform spatial domain filtering on the pre-denoising chrominance information of the image to obtain the spatially denoised chrominance information of the image; Determine the spatially denoised chrominance information as the post-denoising chrominance information of the image.

10. An image denoising device, wherein, Includes: An image information extraction module, configured to execute obtaining an image and obtaining pre-denoising luminance information from the image; A spatial domain filtering module, configured to execute performing spatial domain filtering on the pre-denoising luminance information of the image to obtain the spatially post-denoised luminance information of the image; A first luminance residual obtaining module, configured to execute obtaining the first luminance residual information of the image according to the pre-denoising luminance information of the image and the spatially post-denoised luminance information of the image; A second luminance residual obtaining module, configured to execute performing frequency domain filtering on the first luminance residual information of the image to obtain the second luminance residual information of the image; A denoised image obtaining module, configured to execute obtaining the post-denoising luminance information of the image according to the spatially post-denoised luminance information of the image and the second luminance residual information of the image.

Images