Image spatial domain noise reduction method and device

Abstract

The application discloses an image spatial domain noise reduction method and device, the method comprises the following steps: establishing a multi-scale pyramid according to an image to be reduced, and establishing a directional filtering pyramid by using the intermediate result of the Gaussian pyramid, and fusing and reconstructing the image to be reduced according to the non-local mean filtering result of each pyramid. It can be seen that the application can effectively avoid the situation that the scene adaptability is not enough by using the multi-scale pyramid, and the problem that the texture edge of the image after noise reduction is excessively smoothed or edge color noise residues are generated is effectively avoided by adding the directional filtering pyramid, the overall definition is considered while the noise reduction effect is ensured, and the efficiency and accuracy of the image noise reduction work are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to an image spatial domain denoising method. BACKGROUND

[0002] With the rapid development of the digital era, digital images have gradually become an important part of people's production and life, but digital images will inevitably be contaminated by noise in the process of acquisition, transmission, etc. Since people's demand for high-quality digital images is increasingly strong, image denoising has become a key link in image processing.

[0003] At present, the most representative image denoising method is a non-local filter, such as Non-Local Means (NLM) and Block Mathing 3D (BM3D), which utilizes the self-similarity feature of the image and obtains image denoising results by weighted average of pixel points or pixel blocks. However, the existing image denoising method has a relatively unsatisfactory result in processing the image edge, and it is difficult to be compatible with the denoising effect of the flat texture area. In addition, although it can effectively remove low-frequency noise, it will also lose the high-frequency details of the image, resulting in a decrease in image clarity, that is, the current image denoising method cannot guarantee both the image denoising effect and the image clarity. Therefore, it is particularly important to propose an image denoising method that takes into account the overall denoising effect and clarity requirements. SUMMARY

[0004] The technical problem to be solved by the present application is to provide an image spatial domain denoising method and device, which can improve the clarity of the denoised image and ensure the overall denoising effect of the image.

[0005] To solve the above technical problems, the present application discloses an image spatial domain denoising method in the first aspect, which comprises:

[0006] Obtaining a to-be-denoised image, establishing a multi-scale pyramid of the to-be-denoised image, the multi-scale pyramid comprising a Gaussian pyramid and a Laplacian pyramid;

[0007] Constructing a directional filter pyramid of the to-be-denoised image according to the Gaussian pyramid;

[0008] Determining point information corresponding to each pixel point in the directional filter pyramid, the point information comprising category information corresponding to the pixel point, the category information being used to indicate whether the pixel point is a flat point or an edge texture point;

[0009] performing non-local mean filtering on all pixel points in each layer of the Laplacian pyramid layer by layer to obtain filtering results of the pixel points in each layer of the Laplacian pyramid, and determining the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid;

[0010] performing the non-local mean filtering on each of the edge texture points in each layer of the directional filter pyramid layer by layer to obtain filtering results of the edge texture points in each layer of the directional filter pyramid, and determining the number of similar points corresponding to each of the edge texture points in each layer of the directional filter pyramid;

[0011] performing a fusion reconstruction operation according to the classification of each pixel point in the directional filter pyramid, the filtering results of the pixel points in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering results of the edge texture points in each layer of the directional filter pyramid, and the number of similar points corresponding to each of the edge texture points in each layer of the directional filter pyramid, to obtain a denoised image corresponding to the image to be denoised.

[0012] As an optional implementation, in the first aspect of the present application, the construction of the directional filter pyramid of the image to be denoised according to the Gaussian pyramid comprises:

[0013] obtaining each directional filter result of the directional filter pyramid from the intermediate results of the Gaussian pyramid;

[0014] constructing a residual image of the directional filter pyramid according to each directional filter result of the directional filter pyramid;

[0015] wherein each directional filter result of the directional filter pyramid is represented as:

[0016] OF l,m =I l *f m ,

[0017] wherein OF l,m is a filtering result of a pixel point in the lth layer of the directional filter pyramid in the mth direction, I l is an image in the lth layer, l is the number of image layers, f m is a directional filter corresponding to the mth direction, and * is an image convolution operation;

[0018] and the residual image of the directional filter pyramid is represented as:

[0019]

[0020] wherein OR l,mfor a residual map of the directional filter pyramid, upsample() is an operator of an up-sampling operation.

[0021] As an optional implementation, in the first aspect of the present application, for each pixel point in the directional filter pyramid, if the pixel point is the edge texture point, the point information corresponding to the pixel point further comprises the main direction information of the pixel point and the edge intensity information of the pixel point; if the pixel point is the flat point, the point information corresponding to the pixel point further comprises the edge intensity information of the pixel point.

[0022] The determination of the point information corresponding to each pixel point in the directional filter pyramid comprises:

[0023] For each pixel point in the directional filter pyramid, the residual values of the pixel point in different directional residual maps of the directional filter pyramid are compared to obtain a maximum residual value corresponding to the pixel point.

[0024] The maximum residual value is compared with a preset edge low threshold value:

[0025] When the maximum residual value is less than the preset edge low threshold value, it is determined that the category of the pixel point is used to represent that the pixel point is a flat point.

[0026] When the maximum residual value is greater than or equal to the preset edge low threshold value, it is determined that the category of the pixel point is used to represent that the pixel point is an edge texture point, and the main direction information of the edge texture point is determined, the main direction information being the direction corresponding to the maximum residual value.

[0027] The edge intensity information of the pixel point is determined according to the maximum residual value and a preset edge high threshold value.

[0028] As an optional implementation, in the first aspect of the present application, the calculation formula of the edge intensity information of the pixel point is:

[0029]

[0030] wherein, edge str is the edge intensity information of the pixel point, th_hi is the preset edge high threshold value, mx_OR l is the maximum residual value corresponding to the pixel point.

[0031] As an optional implementation, in the first aspect of the present application, the non-local mean filtering is performed on each edge texture point in each layer of the directional filter pyramid point by point, comprising:

[0032] For each of the edge texture points in the directional filtering pyramid, the non-local mean filtering is performed on the residual map point by point according to the main direction information corresponding to the edge texture point in the directional filtering pyramid.

[0033] As an optional implementation, in the first aspect of the present application, the fusion reconstruction operation is performed according to the classification of each pixel point in the directional filtering pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the directional filtering pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid, to obtain the denoised image corresponding to the image to be denoised, including:

[0034] For each of the flat points in the directional filtering pyramid, the filtering result of the target pixel point in each layer of the Laplacian pyramid that matches the position of the flat point is accumulated layer by layer according to the structure of the Gaussian pyramid and the Laplacian pyramid, to reconstruct the denoised result corresponding to the flat point.

[0035] For each of the edge texture points in the directional filtering pyramid, the filtering result of each of the pixel points in each layer of the Laplacian pyramid that matches the position of the edge texture point is fused with the filtering result of each of the edge texture points in each layer of the directional filtering pyramid according to the number of similar points corresponding to each of the pixel points in each layer of the Laplacian pyramid, the number of similar points corresponding to each of the edge texture points in each layer of the directional filtering pyramid, and the determined edge intensity information of each pixel point in the directional filtering pyramid, to reconstruct the denoised result corresponding to the edge texture point.

[0036] The denoised result corresponding to the flat point and the denoised result corresponding to the edge texture point are integrated to obtain the denoised image corresponding to the image to be denoised.

[0037] As an optional implementation, in the first aspect of the present application, the formula for reconstructing the denoised result corresponding to the edge texture point is:

[0038]

[0039] wherein out is the denoised result corresponding to the edge texture point, count or is the number of similar points corresponding to each of the edge texture points in each layer of the directional filtering pyramid, f or is the filtering result of each of the edge texture points in each layer of the directional filtering pyramid, count lpThe number of similar points corresponding to each pixel in each layer of the Laplacian pyramid is f lp The filtering result of each pixel in each layer of the Laplacian pyramid is edge str The edge intensity information of all the pixels in the determined direction filtering pyramid is l, and the number of image layers is l.

[0040] The second aspect of the present application discloses an image spatial domain noise reduction device, the device comprises:

[0041] The acquisition module is used for acquiring a to-be-reduced-noise image, and establishing a multi-scale pyramid of the to-be-reduced-noise image, wherein the multi-scale pyramid comprises a Gaussian pyramid and a Laplacian pyramid;

[0042] The construction module is used for constructing a direction filtering pyramid of the to-be-reduced-noise image according to the Gaussian pyramid;

[0043] The determination module is used for determining point information corresponding to each pixel in the direction filtering pyramid, wherein the point information comprises category information corresponding to the pixel, and the category information is used for indicating whether the pixel is a flat point or an edge texture point;

[0044] The filtering module is used for performing non-local mean filtering on all the pixels in each layer of the Laplacian pyramid point by point, obtaining a filtering result of each pixel in each layer of the Laplacian pyramid, and determining the number of similar points corresponding to each pixel in each layer of the Laplacian pyramid;

[0045] The filtering module is also used for performing the non-local mean filtering on each edge texture point in each layer of the direction filtering pyramid point by point, obtaining a filtering result of each edge texture point in each layer of the direction filtering pyramid, and determining the number of similar points corresponding to each edge texture point in each layer of the direction filtering pyramid;

[0046] The fusion reconstruction module is used for performing a fusion reconstruction operation according to the classification of each pixel in the direction filtering pyramid, the filtering result of each pixel in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the direction filtering pyramid, and the number of similar points corresponding to each edge texture point in each layer of the direction filtering pyramid, to obtain a reduced-noise image corresponding to the to-be-reduced-noise image.

[0047] As an optional implementation, in the second aspect of the present application, the specific implementation of the construction module for constructing the direction filtering pyramid of the to-be-reduced-noise image according to the Gaussian pyramid is as follows:

[0048] acquiring each directional filtering result of the directional filtering pyramid via intermediate results of the Gaussian pyramid;

[0049] constructing a residual map of the directional filtering pyramid according to each directional filtering result of the directional filtering pyramid;

[0050] wherein each directional filtering result of the directional filtering pyramid is represented as:

[0051] OF l,m = I l *f m ,

[0052] wherein OF l,m is a filtering result of a pixel point in the lth layer of the directional filtering pyramid in the mth direction, I l is an image of the lth layer, l is a layer number of the image, f m is a directional filter corresponding to the mth direction, and * is an image convolution operation;

[0053] and the residual map of the directional filtering pyramid is represented as:

[0054]

[0055] wherein OR l,m is the residual map of the directional filtering pyramid, and upsample() is an operator of an up-sampling operation.

[0056] As an optional implementation, in the second aspect of the present application, for each pixel point in the directional filtering pyramid, if the pixel point is the edge texture point, the point information corresponding to the pixel point further comprises main direction information of the pixel point and edge intensity information of the pixel point; if the pixel point is the flat point, the point information corresponding to the pixel point further comprises edge intensity information of the pixel point.

[0057] wherein the specific implementation of the determining module to determine the point information corresponding to each pixel point in the directional filtering pyramid is:

[0058] for each pixel point in the directional filtering pyramid, comparing residual values of the pixel point in different directional residual maps of the directional filtering pyramid to obtain a maximum residual value corresponding to the pixel point;

[0059] comparing the maximum residual value with a preset edge low threshold value:

[0060] when the maximum residual value is less than the preset edge low threshold value, determining the category of the pixel point to represent that the pixel point is a flat point;

[0061] when the maximum residual value is greater than or equal to the preset edge low threshold, determining the category of the pixel point to represent that the pixel point is an edge texture point, and determining the main direction information of the edge texture point, the main direction information being a direction corresponding to the maximum residual value;

[0062] determining the edge strength information of the pixel point according to the maximum residual value and a preset edge high threshold.

[0063] As an optional implementation, in the second aspect of the present application, the calculation formula of the edge strength information of the pixel point is:

[0064]

[0065] wherein, edge str is the edge strength information of the pixel point, th_hi is the preset edge high threshold, mx_OR l is the maximum residual value corresponding to the pixel point.

[0066] As an optional implementation, in the second aspect of the present application, the specific implementation of the non-local mean filtering performed by the filtering module on each edge texture point in each layer of the directional filter pyramid is:

[0067] for all edge texture points in each layer of the directional filter pyramid, according to the main direction information corresponding to the edge texture point, selecting a residual image in the directional filter pyramid to perform the non-local mean filtering on the edge texture point by point.

[0068] As an optional implementation, in the second aspect of the present application, the specific implementation of the fusion reconstruction module performing a fusion reconstruction operation according to the category of each pixel point in the directional filter pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the directional filter pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filter pyramid to obtain the denoised image corresponding to the image to be denoised is:

[0069] for each flat point in the directional filter pyramid, according to the structure of the Gaussian pyramid and the Laplacian pyramid, layer by layer accumulating the filtering result of the target pixel point in each layer of the Laplacian pyramid matched with the position of the flat point to reconstruct the denoised result corresponding to the flat point;

[0070] For each of the edge texture points in the direction filter pyramid, according to the number of similar points corresponding to each of the pixel points in each layer of the Laplacian pyramid, the number of similar points corresponding to each of the edge texture points in each layer of the direction filter pyramid and the determined edge intensity information of each pixel point in the direction filter pyramid, the filtering results of each of the pixel points in each layer of the Laplacian pyramid matching the position of the edge texture point and the filtering results of each of the edge texture points in each layer of the direction filter pyramid are fused to reconstruct the denoised result corresponding to the edge texture point.

[0071] The denoised results corresponding to the flat points and the denoised results corresponding to the edge texture points are integrated to obtain a denoised image corresponding to the image to be denoised.

[0072] As an optional implementation, in the second aspect of the present application, the formula for reconstructing the denoised result corresponding to the edge texture point is:

[0073]

[0074] Wherein, out is the denoised result corresponding to the edge texture point, count or is the number of similar points corresponding to each of the edge texture points in each layer of the direction filter pyramid, f or is the filtering result of each of the edge texture points in each layer of the direction filter pyramid, count lp is the number of similar points corresponding to each of the pixel points in each layer of the Laplacian pyramid, f lp is the filtering result of each of the pixel points in each layer of the Laplacian pyramid, edge str is the determined edge intensity information of all pixel points in the direction filter pyramid, and l is the number of image layers.

[0075] The third aspect of the present application discloses another image spatial domain noise reduction device, which comprises:

[0076] A memory storing executable program codes;

[0077] A processor coupled with the memory;

[0078] The processor invokes the executable program codes stored in the memory to execute the image spatial domain noise reduction method disclosed in the first aspect of the present application.

[0079] The fourth aspect of the present application discloses a computer storage medium storing computer instructions, which are invoked to execute the image spatial domain noise reduction method disclosed in the first aspect of the present application.

[0080] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0081] In the embodiment of the present application, a multi-scale pyramid is established according to the image to be denoised, a directional filtering pyramid is established by using the intermediate result of the Gaussian pyramid, and the image to be denoised is fused and reconstructed according to the non-local mean filtering result of each pyramid. It can be seen that the implementation of the present application can effectively avoid the case that the scene adaptability is not enough by using the multi-scale pyramid, and the addition of the directional filtering pyramid effectively avoids the problems that the texture edge of the image after denoising is excessively smoothed or edge color noise residues are generated, which guarantees the denoising effect, takes into account the overall clarity, and improves the efficiency and accuracy of the image denoising work. BRIEF DESCRIPTION OF DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0083] Figure 1 is a flowchart of an image spatial domain denoising method disclosed by the embodiment of the present application;

[0084] Figure 2 is a flowchart of another image spatial domain denoising method disclosed by the embodiment of the present application;

[0085] Figure 3 is a structural schematic diagram of an image spatial domain denoising device disclosed by the embodiment of the present application;

[0086] Figure 4 is a structural schematic diagram of another image spatial domain denoising device disclosed by the embodiment of the present application. DETAILED DESCRIPTION

[0087] In order to make the person skilled in the art better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0088] The terms "first", "second", and the like in the description and claims of the present application and the above drawings are used to distinguish between similar objects, not to describe a particular sequential order. Furthermore, the terms "comprises", "comprising", "includes", "including" and the like are intended to cover non-exclusive inclusions, such as one or more steps or units can be included in the processes, methods, articles, or apparatuses without the listed steps or units being exclusive, but optionally in addition to other steps or units inherent to such processes, methods, articles, or apparatuses.

[0089] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive or alternative embodiments. It is expressly understood that the embodiments described herein are combinable with each other.

[0090] The application discloses an image spatial domain noise reduction method and device, which can improve the definition of the image after noise reduction and ensure the overall noise reduction effect of the image. The following will be described in detail.

[0091] Embodiment one

[0092] Please refer to Figure 1 , Figure 1 is a flowchart of an image spatial domain noise reduction method disclosed by the embodiment of the application. Wherein, Figure 1 The described image spatial domain noise reduction method can be applied to an image processing system, wherein the image processing system can include any one of an image processing device, an intelligent server or an intelligent platform for controlling the image processing device, wherein the image processing device can include an image noise reduction device, the intelligent server includes a local server or a cloud server, and the image processing device can establish a communication connection with a user equipment, wherein the user equipment includes but is not limited to at least one of a smart phone (Android phone, iOS phone, etc.), a smart phone number watch, a tablet computer, a palm computer, a vehicle-mounted computer, a desktop computer, a netbook, a personal digital assistant (PDA), a smart navigator and a mobile internet device (MID), etc., and the embodiment of the application is not limited. As shown in the figure, Figure 1 The image spatial domain noise reduction method can include the following operations:

[0093] 101, acquire a to-be-noise-reduced image, establish a multi-scale pyramid of the to-be-noise-reduced image, and the multi-scale pyramid includes a Gaussian pyramid and a Laplacian pyramid.

[0094] In the embodiment of the present application, a multi-scale pyramid of the image to be denoised is obtained through a series of Gaussian smoothing and down-sampling operations, wherein the Gaussian pyramid contains low-frequency features of the image to be denoised, and the Laplacian pyramid contains high-frequency information of more noise and partial edge texture of the image to be denoised.

[0095] In the embodiment of the present application, the multi-scale pyramid can also include a DOG pyramid, a wavelet or a curved wave, etc., which is used to replace the existing pyramid, and the present application is not limited thereto.

[0096] In the embodiment of the present application, a Gaussian low-pass filter with different scales (n layers, n≥2) is used to convolve with the image to obtain a multi-level Gaussian pyramid and a Laplacian pyramid. For example, given an image I, an n-layer Gaussian pyramid {G l} of the image is constructed as follows:

[0097]

[0098]

[0099] wherein * is an image convolution operation, l∈0,1,...,n, downsample(·) is an operation of down-sampling the image; and a Laplacian pyramid {L l} is calculated as follows:

[0100]

[0101] wherein upsample(·) is an operator of using a Gaussian kernel to up-sample (double the size of the image) the image at each layer, and according to the formula, the top layer (l=n) of the Gaussian pyramid is consistent with the top layer of the Laplacian pyramid.

[0102] 102. Construct a directional filter pyramid of the image to be denoised according to the Gaussian pyramid.

[0103] In the embodiment of the present application, the intermediate results of the Gaussian pyramid are used to obtain the results of directional filtering, and at the same time, the operation of the Laplacian pyramid is imitated to subtract the filtering results of each layer of the directional filter pyramid from the input original image of the previous layer to obtain the high-frequency residual of each layer of directional filtering. The Gaussian filter has the characteristic of isotropy, but it itself contains filtering processing in each direction, and the directional filtering results can be simply obtained by outputting the intermediate results for a certain caching processing while obtaining the filtering results of each layer of the Gaussian pyramid. Taking a four-directional filter as an example, its expression is as follows:

[0104]

[0105] The directional filtering results OFl,m and a residual map OR l,m The calculation process is as follows:

[0106]

[0107] 103, determine the point information corresponding to each pixel point in the direction filtering pyramid, the point information includes the category information corresponding to the pixel point, and the category information is used to indicate that the pixel point is a flat point or an edge texture point.

[0108] In the embodiment of the application, the maximum value of each pixel point in the direction filtering pyramid in different direction filtering residual maps is counted, and compared with a preset edge low threshold, the pixel point lower than the edge low threshold is marked as a flat point, the pixel point higher than the edge low threshold is marked as an edge texture point, and the direction where the residual maximum value is located is recorded as the main direction of the pixel point, wherein, optionally, the edge low threshold can be a preset fixed threshold, or an adaptive threshold, such as a threshold calculated according to noise curve and / or gradient difference information, which can be selected according to actual conditions, and the application does not make any limitation.

[0109] 104, performing non-local mean filtering on all pixel points in each layer of the Laplacian pyramid point by point to obtain the filtering result of each pixel point in each layer of the Laplacian pyramid, and determining the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid.

[0110] In the embodiment of the application, NLM (non-local mean filtering) filtering is performed on all pixel points in each layer of the Laplacian pyramid, it is assumed that the gray value of the pixel point i is f(i), a center window P(i) with a window size of r and a search window with a window size of t (t>s) are extracted respectively with the pixel point as the center, the pixel point j is selected by traversing in the search area, and the matching window P(j) with the same size as the center window is selected with the pixel point as the center, if the difference between the sum of the gray values of multiple pixel points in P(j) and the sum of the gray values of multiple pixel points in P(i) is less than a preset similarity threshold, it is considered that the matching window has a higher similarity with the center window, and the pixel point j is marked as a similar point.

[0111] In the embodiment of the application, first, the Gaussian weighted Euclidean distance d(i,j) of the center window P(i) and the similar window P(j) is calculated as follows:

[0112] d(i,j)=gs1*P(i)-P(j)2,

[0113] Secondly, the weight coefficient w(i,j) of the center window P(i) and the similar window P(j) is calculated as follows:

[0114]

[0115] Wherein, h is used for controlling the smooth degree, and gs1 is a preset Gaussian function.

[0116] Then, the number of similar points in the search window is counted, and a weighted average operation is performed on the similar points, and the specific formula is as follows:

[0117]

[0118] Finally, the NLM filtering is performed on each layer of the Laplacian pyramid, and the filtering results of each pixel point after filtering and the number of corresponding similar points are recorded.

[0119] 105. The non-local mean filtering is performed on each edge texture point in each layer of the directional filtering pyramid, to obtain the filtering result of each edge texture point in each layer of the directional filtering pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid is determined.

[0120] In the embodiment of the application, for each edge texture point in each layer of the directional filtering pyramid, the residual image corresponding to the main direction of the edge texture point is selected to perform NLM filtering on the edge texture point, wherein, optionally, according to preset requirements or actual conditions, the residual image corresponding to other directions of the edge texture point can be selected to perform NLM filtering, taking a four-direction filter as an example, the maximum value of the edge texture point in the four-direction residual image is counted, and the direction is defined as the main direction with the most obvious feature, and when the NLM filtering is performed on the edge texture point, the residual image corresponding to the main direction can be used, or the residual image corresponding to the other three directions can be used, which is not limited in the application.

[0121] 106. The fusion reconstruction operation is performed according to the classification of each pixel point in the directional filtering pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the directional filtering pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid, to obtain the denoised image corresponding to the to-be-denoised image.

[0122] In the embodiment of the application, for the flat point in the directional filtering pyramid, the result after NLM filtering of each layer can be directly accumulated layer by layer according to the structure of the Gaussian pyramid and the Laplacian pyramid, to reconstruct the denoised result, and for the edge texture point in the directional filtering pyramid, the NLM filtering results of each edge texture point in the directional filtering pyramid and the Laplacian pyramid can be fused and reconstructed according to the number of similar points corresponding to each edge texture point and the edge intensity value.

[0123] It can be seen that the image spatial domain noise reduction method described in the embodiment of the present application can establish a multi-scale pyramid according to the image to be reduced, and establish a directional filtering pyramid using the intermediate result of the Gaussian pyramid, and fuse and reconstruct the image to be reduced according to the non-local mean filtering result of each pyramid. The use of the multi-scale pyramid effectively avoids the situation that the scene adaptability is not enough, and the addition of the directional filtering pyramid effectively avoids the problem that the texture edge of the image after noise reduction is excessively smoothed or edge color noise is generated. In addition, the overall clarity is taken into account while ensuring the noise reduction effect, thereby improving the efficiency and accuracy of the image noise reduction work.

[0124] In an optional embodiment, for each pixel point in the directional filtering pyramid, if the pixel point is an edge texture point, the point information corresponding to the pixel point further includes the main direction information of the pixel point and the edge intensity information of the pixel point; if the pixel point is a flat point, the point information corresponding to the pixel point further includes the edge intensity information of the pixel point.

[0125] The method can further include the following operation:

[0126] For each pixel point in the directional filtering pyramid, the residual values of the pixel point in the residual error maps in different directions of the directional filtering pyramid are compared to obtain the maximum residual value corresponding to the pixel point.

[0127] The maximum residual value is compared with a preset edge low threshold value:

[0128] When the maximum residual value is less than the preset edge low threshold value, it is determined that the category of the pixel point is used to represent that the pixel point is a flat point.

[0129] When the maximum residual value is greater than or equal to the preset edge low threshold value, it is determined that the category of the pixel point is used to represent that the pixel point is an edge texture point, and the main direction information of the edge texture point is determined, and the main direction information is the direction corresponding to the maximum residual value.

[0130] The edge intensity information of the pixel point is determined according to the maximum residual value and a preset edge high threshold value.

[0131] In the optional embodiment, the maximum values of each pixel point in different directional filtering residual error maps in each layer of the directional filtering pyramid are counted, and compared with a preset edge low threshold value. The pixel points lower than the edge low threshold value are marked as flat points, the pixel points higher than the edge low threshold value are marked as edge texture points, and the direction of the maximum residual value is recorded as the main direction of the pixel point. Optionally, the edge low threshold value can be a preset fixed threshold value, or an adaptive threshold value, such as a threshold value calculated according to noise curve and / or gradient difference information. The actual situation can be selected, and the present embodiment is not limited.

[0132] In the optional embodiment, the formula for calculating the maximum value of each pixel in each layer of the directional filter pyramid in the different directional filter residual error maps can be:

[0133] mx_OR l = MAX(OR l,0 ,OR l,1 ,...,OR l,m ),

[0134] The formula for determining the edge texture points and flat points can be:

[0135]

[0136] The formula for calculating the edge strength of the pixel point by comparing it with the edge high threshold value can be:

[0137]

[0138] It can be seen that by implementing the optional embodiment, for each pixel point in the directional filter pyramid, the residual error value of the pixel point in the different directional residual error maps of the directional filter pyramid is compared to obtain the maximum residual error value corresponding to the pixel point. The maximum residual error value is compared with the preset edge low threshold value. When the maximum residual error value is less than the preset edge low threshold value, the category of the pixel point is determined to indicate that the pixel point is a flat point. When the maximum residual error value is greater than or equal to the preset edge low threshold value, the category of the pixel point is determined to indicate that the pixel point is an edge texture point, and the main direction information of the edge texture point is determined. The main direction information is the direction corresponding to the maximum residual error value. The edge strength information of the pixel point is determined according to the maximum residual error value and the preset edge high threshold value, which improves the clarity of the denoised image and effectively avoids the problems of over-smoothing of the texture edge or residual edge color noise of the denoised image.

[0139] Embodiment Two

[0140] Please refer to Figure 2 , Figure 2 is a flowchart of another image spatial domain noise reduction method disclosed in the embodiments of the present application. In the method, Figure 2The described image spatial domain noise reduction method can be applied to an image processing system, which can include any one of an image processing device, an intelligent server or an intelligent platform for controlling the image processing device, wherein the image processing device can include an image noise reduction device, the intelligent server includes a local server or a cloud server, and the image processing device can establish a communication connection with a user equipment, which includes at least one of a smart phone (an Android phone, an iOS phone, etc.), a smart phone number watch, a tablet computer, a palm computer, a vehicle-mounted computer, a desktop computer, a netbook, a personal digital assistant (PDA), a smart navigator, and a mobile internet device (MID), etc., without any limitation. As shown in the following formula (1), the image spatial domain noise reduction method can include the following operations: Figure 2

[0141] 201. An image to be denoised is acquired, and a multi-scale pyramid of the image to be denoised is established, which includes a Gaussian pyramid and a Laplacian pyramid.

[0142] 202. Each directional filtering result of a directional filtering pyramid is acquired through an intermediate result of the Gaussian pyramid.

[0143] In the embodiment of the present application, the intermediate result of the Gaussian pyramid is used to acquire the result of directional filtering, and the directional filtering pyramid, like the Gaussian-Laplacian pyramid, contains a low-frequency filtering result and a residual image, wherein the filtering result is acquired by buffering and calculating some intermediate results of the current layer of the Gaussian pyramid.

[0144] 203. A residual image of the directional filtering pyramid is constructed according to each directional filtering result of the directional filtering pyramid.

[0145] In the embodiment of the present application, the filtering result of each layer of the directional filtering pyramid is subtracted from the input original image of the previous layer to acquire the high-frequency residual of each layer of directional filtering, which imitates the operation of the Laplacian pyramid.

[0146] 204. Point information corresponding to each pixel point in the directional filtering pyramid is determined, and the point information includes category information corresponding to the pixel point, which is used to represent whether the pixel point is a flat point or an edge-texture point.

[0147] 205. Non-local mean filtering is performed on all pixel points in each layer of the Laplacian pyramid point by point to obtain the filtering result of each pixel point in each layer of the Laplacian pyramid, and the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid is determined.

[0148] ​206. Perform nonlocal mean filtering on each edge texture point in each layer of the directional filtering pyramid to obtain the filtering result of each edge texture point in each layer of the directional filtering pyramid, and determine the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid.

[0149] 207. Based on the classification of each pixel in the directional filtering pyramid, the filtering results of each pixel in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel in each layer of the Laplacian pyramid, the filtering results of each edge texture point in each layer of the directional filtering pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid, perform a fusion and reconstruction operation to obtain the denoised image corresponding to the image to be denoised.

[0150] For further descriptions of steps 201 and 204-207 in this embodiment of the invention, please refer to the detailed description of steps 101 and 103-106 in Embodiment 1. These descriptions will not be repeated in this embodiment of the invention.

[0151] As can be seen, the image spatial domain denoising method described in the embodiments of the present invention can establish a multi-scale pyramid based on the image to be denoised, and establish a directional filtering pyramid using the intermediate results of the Gaussian pyramid. The image to be denoised is then fused and reconstructed based on the non-local mean filtering results of each pyramid. The introduction of the intermediate results of the Gaussian pyramid to calculate the directional filtering pyramid improves the computational efficiency, thereby improving the denoising efficiency of the image to be denoised. At the same time, the addition of the directional filtering pyramid effectively avoids the problem of excessive smoothing of texture edges or residual edge color noise in the image after denoising. While ensuring the denoising effect, it also takes into account the overall clarity, thus improving the efficiency and accuracy of image denoising.

[0152] In an optional embodiment, the formula for calculating the edge intensity information of a pixel is:

[0153]

[0154] Among them, edge str This represents the edge intensity information of a pixel, where th_hi is the preset edge height threshold, and mx_OR is the edge intensity information of the pixel. l This represents the maximum residual value corresponding to the pixel.

[0155] In this optional embodiment, the edge intensity information of a pixel is calculated by comparing the maximum residual value of the pixel with a preset edge high threshold. Optionally, the edge high threshold can be a preset fixed threshold or an adaptive threshold, such as a threshold calculated based on information such as noise curves and / or gradient differences. It can be selected according to the actual situation, and this embodiment does not limit it.

[0156] It can be seen that the optional embodiment can calculate the edge intensity information of the pixel point according to the maximum residual value corresponding to the pixel point and the preset edge high threshold, and improve the accuracy and reliability of the determined edge intensity information.

[0157] In another optional embodiment, the point-by-point non-local mean filtering of each edge texture point in each layer of the directional filtering pyramid can include the following operations:

[0158] For all edge texture points in each layer of the directional filtering pyramid, the residual map is selected in the directional filtering pyramid for point-by-point non-local mean filtering according to the main direction information corresponding to the edge texture point.

[0159] In the optional embodiment, for the edge texture point, the residual map is selected in the directional filtering pyramid for point-by-point non-local mean filtering according to the main direction information of the edge texture point, and the filtering window and the filtering process are similar to the operation of the Laplacian pyramid, that is, the NLM (non-local mean filtering) filtering is performed on all edge texture points in each layer of the directional filtering pyramid, assuming that the gray value of the edge texture point i is f(i), a center window P(i) with a window size of r and a search window with a window size of t (t>s) are extracted respectively with the edge texture point as the center, the pixel point j is selected by traversing the search area, and the matching window P(j) with the same size as the center window is selected with the pixel point as the center. If the difference between the sum of the gray values of multiple pixel points in the matching window and the sum of the gray values of multiple edge texture points in the center window is less than a preset similarity threshold, it is considered that the matching window has a higher similarity with the center window, and the pixel point j is marked as a similar point.

[0160] It can be seen that the optional embodiment can select the residual map in the directional filtering pyramid for point-by-point non-local mean filtering according to the main direction information corresponding to the edge texture point, effectively reduce the calculation amount of the non-local mean filtering, improve the calculation efficiency, and effectively avoid the problems of over-smoothing of the edge texture and residual edge color noise in the denoised image after denoising due to the addition of the directional filtering pyramid.

[0161] In yet another optional embodiment, the fusion reconstruction operation is performed according to the classification of each pixel point in the directional filtering pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the directional filtering pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid, and the denoised image corresponding to the to-be-denoised image can include the following operations:

[0162] For each flat point in the directional filtering pyramid, according to the structures of the Gaussian pyramid and the Laplacian pyramid, the filtering results of the target pixel points in each layer of the Laplacian pyramid that match the position of the flat point are accumulated layer by layer to reconstruct the denoised result corresponding to the flat point;

[0163] For each edge texture point in the directional filtering pyramid, according to the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid and the determined edge intensity information of each pixel point in the directional filtering pyramid, the filtering results of the pixel points in each layer of the Laplacian pyramid that match the position of the edge texture point are fused with the filtering results of the edge texture points in each layer of the directional filtering pyramid to reconstruct the denoised result corresponding to the edge texture point.

[0164] The denoised results corresponding to the flat points and the denoised results corresponding to the edge texture points are integrated to obtain the denoised image corresponding to the image to be denoised.

[0165] In the optional embodiment, for the flat points in the directional filtering pyramid, the results of the NLM filtering of each layer can be directly accumulated layer by layer according to the structures of the Gaussian pyramid and the Laplacian pyramid to reconstruct the denoised result; for the edge texture points in the directional filtering pyramid, the NLM filtering results of the edge texture points in the directional filtering pyramid and the Laplacian pyramid can be fused according to the number of similar points corresponding to each edge texture point and the edge intensity value to reconstruct.

[0166] It can be seen that the optional embodiment can perform different denoising fusion reconstruction strategies for the flat points and the edge texture points in the directional filtering pyramid, can effectively balance the denoising effect and the clarity requirement, and improves the efficiency and accuracy of image denoising work.

[0167] In yet another optional embodiment, the formula for reconstructing the denoised result corresponding to the edge texture point is:

[0168]

[0169] wherein out is the denoised result corresponding to the edge texture point, count or is the number of similar points corresponding to each edge texture point in each layer of the directional filtering pyramid, f or is the filtering result of each edge texture point in each layer of the directional filtering pyramid, count lp is the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, f lp is the filtering result of each pixel point in each layer of the Laplacian pyramid, edge strThe edge intensity information of all pixel points in the determined direction filtering pyramid is determined, and l is the number of image layers.

[0170] In the optional embodiment, the top layer (l=n) of the pyramid is weighted according to the number of similar points to ensure that the top layer is superimposed on the real direction feature, and the remaining layers are controlled by the edge intensity to balance the noise reduction effect and the clarity requirement.

[0171] It can be seen that the optional embodiment can perform different fusion reconstruction strategies on the edge texture points of different layers, and can quantize the reconstruction process according to the formula, which not only ensures the noise reduction effect and the clarity requirement of the reconstructed image, but also improves the efficiency and quantifiability of the fusion reconstruction.

[0172] Embodiment three

[0173] Please refer to Figure 3 , Figure 3 is a structural schematic diagram of an image spatial domain noise reduction device disclosed by the embodiment of the application. Wherein, Figure 3 The image spatial domain noise reduction device described can be applied to an image processing system, wherein the image processing system can include any one of an image processing device, an intelligent server or an intelligent platform for controlling the image processing device, wherein the image processing device can include an image noise reduction device, the intelligent server includes a local server or a cloud server, and the image processing device can establish a communication connection with a user equipment, wherein the user equipment includes but is not limited to at least one of a smart phone (an Android phone, an iOS phone, etc.), a smart phone number watch, a tablet computer, a palm computer, a vehicle-mounted computer, a desktop computer, a netbook, a personal digital assistant (PDA), a smart navigator and a mobile internet device (MID) and the like, and the embodiment of the application is not limited. As shown in the Figure 3 The image spatial domain noise reduction method device can include:

[0174] The acquisition module 301 is configured to acquire a to-be-noise-reduced image, and establish a multi-scale pyramid of the to-be-noise-reduced image, wherein the multi-scale pyramid includes a Gaussian pyramid and a Laplacian pyramid.

[0175] The construction module 302 is configured to construct a direction filtering pyramid of the to-be-noise-reduced image according to the Gaussian pyramid.

[0176] The determination module 303 is configured to determine point information corresponding to each pixel point in the direction filtering pyramid, wherein the point information includes category information corresponding to the pixel point, and the category information is used to indicate whether the pixel point is a flat point or an edge texture point.

[0177] The filtering module 304 is configured to perform non-local mean filtering on all pixel points in each layer of the Laplacian pyramid layer by layer to obtain filtering results of the pixel points in each layer of the Laplacian pyramid, and determine the number of similar points corresponding to the pixel points in each layer of the Laplacian pyramid.

[0178] The filtering module 304 is further configured to perform non-local mean filtering on each edge texture point in each layer of the directional filtering pyramid layer by layer to obtain filtering results of the edge texture points in each layer of the directional filtering pyramid, and determine the number of similar points corresponding to the edge texture points in each layer of the directional filtering pyramid.

[0179] The fusion reconstruction module 305 is configured to perform a fusion reconstruction operation according to the classification of each pixel point in the directional filtering pyramid, the filtering results of the pixel points in each layer of the Laplacian pyramid, the number of similar points corresponding to the pixel points in each layer of the Laplacian pyramid, the filtering results of the edge texture points in each layer of the directional filtering pyramid, and the number of similar points corresponding to the edge texture points in each layer of the directional filtering pyramid, to obtain a denoised image corresponding to the image to be denoised.

[0180] It can be seen that the image spatial domain denoising device described above can establish a multi-scale pyramid according to the image to be denoised, establish a directional filtering pyramid using the intermediate results of the Gaussian pyramid, and perform a fusion reconstruction operation on the image to be denoised according to the non-local mean filtering results of the pyramids. Figure 3 The image spatial domain denoising device described above can establish a multi-scale pyramid according to the image to be denoised, establish a directional filtering pyramid using the intermediate results of the Gaussian pyramid, and perform a fusion reconstruction operation on the image to be denoised according to the non-local mean filtering results of the pyramids.

[0181] In an optional embodiment, the specific manner in which the construction module 302 constructs the directional filtering pyramid of the image to be denoised according to the Gaussian pyramid includes:

[0182] The directional filtering results of the directional filtering pyramid are obtained through the intermediate results of the Gaussian pyramid.

[0183] A residual image of the directional filtering pyramid is constructed according to the directional filtering results of the directional filtering pyramid.

[0184] The directional filtering results of the directional filtering pyramid are represented as:

[0185] OF l,m = I l *f m ,

[0186] wherein OF l,m is the filtering result of a pixel point in the lth layer of the directional filtering pyramid in the mth direction, Il is the image of the first layer, l is the image layer number, f m is the direction filter corresponding to the mth direction, * is the image convolution operation;

[0187] And the residual map of the direction filter pyramid is represented as:

[0188]

[0189] Wherein, OR l,m is the residual map of the direction filter pyramid, and upsample() is the operator of the up-sampling operation.

[0190] It can be seen that the implementation Figure 3 The described image spatial domain noise reduction device can establish a multi-scale pyramid according to the image to be denoised, and establish a direction filter pyramid by using the intermediate result of the Gaussian pyramid, and fuse and reconstruct the image to be denoised according to the non-local mean filtering result of each pyramid, and introduce the intermediate result of the Gaussian pyramid to calculate the direction filter pyramid, thereby improving the operation efficiency, and further improving the denoising efficiency of the image to be denoised. At the same time, the addition of the direction filter pyramid effectively avoids the problem that the texture edge of the denoised image is excessively smoothed or edge color noise is left, ensures the denoising effect, takes into account the overall clarity, and improves the efficiency and accuracy of the image denoising work.

[0191] In another optional embodiment, for each pixel point in the direction filter pyramid, if the pixel point is an edge texture point, the point information corresponding to the pixel point further includes the main direction information of the pixel point and the edge intensity information of the pixel point; if the pixel point is a flat point, the point information corresponding to the pixel point further includes the edge intensity information of the pixel point.

[0192] Wherein, the specific manner in which the determining module 303 determines the point information corresponding to each pixel point in the direction filter pyramid includes:

[0193] For each pixel point in the direction filter pyramid, the residual values of the pixel point in the different direction residual maps of the direction filter pyramid are compared to obtain the maximum residual value corresponding to the pixel point;

[0194] The maximum residual value is compared with a preset edge low threshold value:

[0195] When the maximum residual value is less than the preset edge low threshold value, the category of the pixel point is determined to indicate that the pixel point is a flat point;

[0196] When the maximum residual value is greater than or equal to the preset edge low threshold value, the category of the pixel point is determined to indicate that the pixel point is an edge texture point, and the main direction information of the edge texture point is determined, the main direction information being the direction corresponding to the maximum residual value;

[0197] The edge intensity information of the pixel point is determined according to the maximum residual value and a preset high edge threshold value.

[0198] It can be seen that, in the implementation Figure 3 The described image spatial domain noise reduction device can compare the residual values of each pixel point in the direction filter pyramid in different direction residual maps of the direction filter pyramid, obtain a maximum residual value corresponding to the pixel point, compare the maximum residual value with a preset low edge threshold value, determine the category of the pixel point to indicate that the pixel point is a flat point when the maximum residual value is less than the preset low edge threshold value, determine the category of the pixel point to indicate that the pixel point is an edge texture point when the maximum residual value is greater than or equal to the preset low edge threshold value, and determine the main direction information of the edge texture point, which is the direction corresponding to the maximum residual value. The edge intensity information of the pixel point is determined according to the maximum residual value and a preset high edge threshold value, which improves the definition of the noise-reduced image and effectively avoids the problems of over-smoothing of texture edges or residual edge color noise in the noise-reduced image.

[0199] In another optional embodiment, the calculation formula of the edge intensity information of the pixel point is:

[0200]

[0201] Wherein, edge str is the edge intensity information of the pixel point, th_hi is the preset high edge threshold value, mx_OR l is the maximum residual value corresponding to the pixel point.

[0202] It can be seen that, in the implementation Figure 3 The described image spatial domain noise reduction device can calculate the edge intensity information of the pixel point according to the maximum residual value corresponding to the pixel point and the preset high edge threshold value, which improves the accuracy and reliability of the determined edge intensity information.

[0203] In another optional embodiment, the specific manner in which the filtering module 304 performs non-local mean filtering on each edge texture point in each layer of the direction filter pyramid point by point includes:

[0204] For all edge texture points in each layer of the direction filter pyramid, according to the main direction information corresponding to the edge texture point, select the residual map in the direction filter pyramid to perform non-local mean filtering point by point.

[0205] It can be seen that, in the implementation Figure 3The described image spatial domain noise reduction device can select residual maps in the direction filter pyramid for point-by-point non-local mean filtering according to the main direction information corresponding to the edge texture points, effectively reduces the calculation amount of the non-local mean filtering, improves the calculation efficiency, and effectively avoids the problems of over-smoothing of the edge texture or residual edge color noise of the image after noise reduction by adding the direction filter pyramid.

[0206] In yet another optional embodiment, the fusion reconstruction module 305 performs a fusion reconstruction operation according to the classification of each pixel point in the direction filter pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the direction filter pyramid, and the number of similar points corresponding to each edge texture point in each layer of the direction filter pyramid, to obtain a specific manner of the image to be denoised corresponding to the image after noise reduction.

[0207] For each flat point in the direction filter pyramid, according to the structure of the Gaussian pyramid and the Laplacian pyramid, the filtering results of the target pixel points matched with the flat point position in each layer of the Laplacian pyramid are accumulated layer by layer to reconstruct the noise reduction result corresponding to the flat point.

[0208] For each edge texture point in the direction filter pyramid, according to the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each edge texture point in each layer of the direction filter pyramid, and the edge intensity information of each pixel point in the direction filter pyramid determined, the filtering results of the pixel points matched with the edge texture point position in each layer of the Laplacian pyramid and the filtering results of the edge texture points in each layer of the direction filter pyramid are fused to reconstruct the noise reduction result corresponding to the edge texture point.

[0209] The noise reduction results corresponding to the flat points and the noise reduction results corresponding to the edge texture points are integrated to obtain the image after noise reduction corresponding to the image to be denoised.

[0210] It can be seen that the implementation Figure 3 The described image spatial domain noise reduction device can perform different noise reduction fusion reconstruction strategies for flat points and edge texture points in the direction filter pyramid, effectively balance the noise reduction effect and clarity requirements, and improve the efficiency and accuracy of image noise reduction work.

[0211] In yet another optional embodiment, the formula for reconstructing the noise reduction result corresponding to the edge texture point is:

[0212]

[0213] Wherein, out is the noise reduction result corresponding to the edge texture point, count orf is the number of similar points corresponding to each edge texture point of each layer of the directional filtering pyramid, or count is the filtering result of each edge texture point in each layer of the directional filtering pyramid, lp f is the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, lp edge is the filtering result of each pixel point in each layer of the Laplacian pyramid, str l is the edge intensity information of all pixel points in the determined directional filtering pyramid, and l is the number of layers of the image.

[0214] It can be seen that the implementation Figure 4 The described image spatial domain noise reduction device can perform different fusion reconstruction strategies on edge texture points of different layers, can quantize the reconstruction process according to the formula, not only ensures the noise reduction effect and definition requirements of the reconstructed image, but also improves the efficiency and quantifiability of the fusion reconstruction.

[0215] Embodiment four

[0216] Please refer to Figure 4 , Figure 4 is another structural schematic diagram of an image spatial domain noise reduction device disclosed by the embodiment of the present application. As ​ shown, the image spatial domain noise reduction device can include:

[0217] a memory 401 storing executable program codes;

[0218] a processor 402 coupled with the memory 401;

[0219] The processor 402 invokes the executable program codes stored in the memory 401 to execute the steps in the image spatial domain noise reduction method described in the embodiment one or the embodiment two of the present application.

[0220] Embodiment five

[0221] The embodiment of the present application discloses a computer storage medium, which stores computer instructions. When the computer instructions are invoked, the steps in the image spatial domain noise reduction method described in the embodiment one or the embodiment two of the present application are executed.

[0222] Embodiment six

[0223] The embodiment of the present application discloses a computer program product, which includes a non-transitory computer readable storage medium storing a computer program, and the computer program is operable to make a computer execute the steps in the image spatial domain noise reduction method described in the embodiment one or the embodiment two.

[0224] The apparatus embodiments described above are only illustrative, wherein the modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules, i.e., can be located in one place or distributed on multiple network modules. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0225] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magnetic disk storage, a magnetic tape storage, or any other computer readable medium that can be used to carry or store data.

[0226] Finally, it should be noted that: the image spatial domain noise reduction method and device disclosed by the embodiments of the present application are only the preferred embodiments of the present application, and are used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that; it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for image spatial domain noise reduction, characterized in that, The method comprises: acquiring a to-be-noise-reduced image, establishing a multi-scale pyramid of the to-be-noise-reduced image, the multi-scale pyramid comprising a Gaussian pyramid and a Laplacian pyramid; constructing a directional filter pyramid of the to-be-noise-reduced image according to the Gaussian pyramid; determining point information corresponding to each pixel point in the directional filter pyramid, the point information comprising category information corresponding to the pixel point, the category information being used to indicate whether the pixel point is a flat point or an edge texture point; performing non-local mean filtering on all pixel points in each layer of the Laplacian pyramid point by point to obtain a filtering result of each pixel point in each layer of the Laplacian pyramid, and determining a number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid; performing the non-local mean filtering on each edge texture point in each layer of the directional filter pyramid point by point to obtain a filtering result of each edge texture point in each layer of the directional filter pyramid, and determining a number of similar points corresponding to each edge texture point in each layer of the directional filter pyramid; performing a fusion reconstruction operation according to the classification of each pixel point in the directional filter pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each edge texture point in each layer of the directional filter pyramid, and the number of similar points corresponding to each edge texture point in each layer of the directional filter pyramid to obtain a noise-reduced image corresponding to the to-be-noise-reduced image.

2. The image spatial domain noise reduction method of claim 1, wherein, The method comprises: obtaining each directional filter result of the directional filter pyramid through an intermediate result of the Gaussian pyramid; constructing a residual map of the directional filter pyramid according to each directional filter result of the directional filter pyramid; wherein each directional filter result of the directional filter pyramid is expressed as: OF l,m = I l f m , wherein, OF l,m is the filtering result of the pixel point in the lth layer of the directional filter pyramid in the mth direction, I l is the image of the lth layer, and l is the image layer number, f m is the directional filter corresponding to the mth direction, and * is the image convolution operation; and the residual map of the directional filter pyramid is expressed as: where OR l,m is the residual map of the directional filter pyramid, and upsample() is the operator of the up-sampling operation.

3. The image spatial domain noise reduction method of claim 1 or 2, characterized in that, For each pixel point in the directional filter pyramid, if the pixel point is the edge texture point, the point information corresponding to the pixel point further comprises main direction information of the pixel point and edge intensity information of the pixel point; if the pixel point is the flat point, the point information corresponding to the pixel point further comprises edge intensity information of the pixel point. The method comprises: for each pixel point in the directional filter pyramid, comparing residual values of the pixel point in different directional residual maps of the directional filter pyramid to obtain a maximum residual value corresponding to the pixel point; comparing the maximum residual value with a preset edge low threshold value: when the maximum residual value is less than the preset edge low threshold value, it is determined that the category of the pixel point is used to indicate that the pixel point is a flat point; When the maximum residual value is greater than or equal to the preset edge low threshold, it is determined that the category of the pixel point is used to represent that the pixel point is an edge texture point, and main direction information of the edge texture point is determined, the main direction information being a direction corresponding to the maximum residual value; Edge strength information of the pixel point is determined according to the maximum residual value and a preset edge high threshold.

4. The image spatial domain noise reduction method of claim 3, wherein, A calculation formula of the edge strength information of the pixel point is: wherein edge str is the edge intensity information of the pixel point, th_hi is the preset high edge threshold, mx_OR l is the maximum residual value corresponding to the pixel point.

5. The image spatial domain noise reduction method according to claim 3 or 4, characterized in that, The non-local mean filtering is performed on each of the edge texture points in each layer of the directional filtering pyramid, including: For all the edge texture points in each layer of the directional filtering pyramid, the non-local mean filtering is performed on each of the edge texture points according to the main direction information corresponding to the edge texture point and a residual image selected in the directional filtering pyramid.

6. The image spatial domain noise reduction method of claim 5, wherein, The fusion reconstruction operation is performed according to the category of each of the pixel points in the directional filtering pyramid, the filtering result of each pixel point in each layer of the Laplacian pyramid, the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, the filtering result of each of the edge texture points in each layer of the directional filtering pyramid, and the number of similar points corresponding to each of the edge texture points in each layer of the directional filtering pyramid, to obtain a denoised image corresponding to the image to be denoised, including: For each of the flat points in the directional filtering pyramid, the filtering result of a target pixel point matched in position with the flat point in each layer of the Laplacian pyramid is accumulated layer by layer according to the structure of the Gaussian pyramid and the Laplacian pyramid, to reconstruct a denoised result corresponding to the flat point; For each of the edge texture points in the directional filtering pyramid, the filtering result of each of the pixel points matched in position with the edge texture point in each layer of the Laplacian pyramid and the filtering result of each of the edge texture points in each layer of the directional filtering pyramid are fused according to the number of similar points corresponding to each of the pixel points in each layer of the Laplacian pyramid, the number of similar points corresponding to each of the edge texture points in each layer of the directional filtering pyramid, and the edge strength information of each of the pixel points in the directional filtering pyramid, to reconstruct a denoised result corresponding to the edge texture point; The denoised result corresponding to the flat point and the denoised result corresponding to the edge texture point are integrated, to obtain a denoised image corresponding to the image to be denoised.

7. The image spatial domain noise reduction method of claim 6, wherein, A formula for reconstructing the denoised result corresponding to the edge texture point is: wherein, out is the result after noise reduction corresponding to the edge texture point, count or is the number of similar points corresponding to each edge texture point of each layer of the directional filter pyramid, f or is the filtering result of each edge texture point in each layer of the directional filter pyramid, count lp is the number of similar points corresponding to each pixel point in each layer of the Laplacian pyramid, f lp is the filtering result of each pixel point in each layer of the Laplacian pyramid, edge str is the determined edge intensity information of all pixel points in the directional filter pyramid, and l is the number of image layers.

8. An image spatial domain noise reduction apparatus, characterized by comprising: The apparatus includes: An acquisition module is configured to acquire an image to be denoised, and establish a multi-scale pyramid of the image to be denoised, the multi-scale pyramid including a Gaussian pyramid and a Laplacian pyramid; A construction module is configured to construct a directional filtering pyramid of the image to be denoised according to the Gaussian pyramid; A determination module is configured to determine point information corresponding to each of the pixel points in the directional filtering pyramid, the point information including category information corresponding to the pixel point, the category information being used to represent that the pixel point is a flat point or an edge texture point; The filtering module is configured to perform non-local mean filtering on all pixel points in each layer of the Laplacian pyramid layer by layer to obtain filtering results of the pixel points in each layer of the Laplacian pyramid, and determine the number of similar points corresponding to the pixel points in each layer of the Laplacian pyramid. The filtering module is further configured to perform the non-local mean filtering on each of the edge texture points in each layer of the directional filtering pyramid layer by layer to obtain filtering results of the edge texture points in each layer of the directional filtering pyramid, and determine the number of similar points corresponding to the edge texture points in each layer of the directional filtering pyramid. The fusion reconstruction module is configured to perform a fusion reconstruction operation according to the classification of each pixel point in the directional filtering pyramid, the filtering results of the pixel points in each layer of the Laplacian pyramid, the number of similar points corresponding to the pixel points in each layer of the Laplacian pyramid, the filtering results of the edge texture points in each layer of the directional filtering pyramid, and the number of similar points corresponding to the edge texture points in each layer of the directional filtering pyramid, to obtain a denoised image corresponding to the image to be denoised.

9. An image spatial domain noise reduction apparatus, characterized by comprising: The apparatus comprises: a memory storing executable program codes; a processor coupled to the memory; the processor invokes the executable program codes stored in the memory to execute the image spatial domain denoising method according to any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which are invoked to execute the image spatial domain denoising method according to any one of claims 1-7.

Images