Three-dimensional recursive filtering method and device for image space-time domain joint noise reduction

Abstract

The application provides a three-dimensional recursive filtering method and device for image space-time domain joint noise reduction. The image space-time domain joint noise reduction method of the application is different from the traditional three-dimensional recursive filtering method which is directly connected in series by three one-dimensional filters. Instead, the application integrates the information of time domain recursion in the process of point-by-point space domain recursion to improve the effect of space domain filtering. Then, the information of space domain recursion is integrated in the process of point-by-point time domain filtering to improve the effect of time domain filtering. Such repeated recursion traverses the whole image. Under the same level of operation logic and storage cost, compared with the traditional space-time domain series and space-time domain parallel methods, the application can better improve the image definition and signal-to-noise ratio while reducing the phenomena of motion trail lack and raindrop noise.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a three-dimensional recursive filtering method and apparatus for joint spatiotemporal domain noise reduction of images. Background Technology

[0002] One crucial aspect of image processing is noise reduction to achieve cleaner and clearer images. This typically involves assuming that images generate additive random noise during sensor acquisition and then using this noise's zero-mean value for weighted removal. For video noise reduction, the two main methods are temporal denoising and spatial denoising. Temporal denoising utilizes temporal correlation and historical information, such as the content of the previous frame, to denoise static areas of the current frame. Spatial denoising employs two methods: one is to denoise the current image by utilizing spatially similar content within the current image, and the other is to use domain transformations such as discrete cosine transform and discrete wavelet transform to filter different frequency domains and then transform back to reduce image noise.

[0003] Temporal denoising can improve the clarity of static areas while removing noise, but it cannot denoise moving areas. Furthermore, the trade-off between motion blur loss and raindrop noise is difficult to handle. Either the motion blur loss is caused in order to suppress raindrop noise, or raindrop noise is caused in order to prevent motion blur loss.

[0004] Although spatial noise reduction methods become stronger as the receptive field increases, there is always a trade-off between noise reduction intensity and sharpness. Either the spatial noise reduction intensity is weakened in order to maintain sharpness, or some sharpness is sacrificed in order to improve the image signal-to-noise ratio.

[0005] Spatiotemporal denoising typically employs a combination approach. For example, it can be done by first performing temporal denoising and then concatenating the results of temporal denoising to the moving and stationary regions with different intensities of spatial denoising. Alternatively, it can be done by first performing spatial denoising and then concatenating the results of spatial denoising to temporal denoising. Or, it can be done by performing temporal denoising on the original image and simultaneously performing spatial denoising on the original image, and then weighting the results in a parallel approach.

[0006] Generally, cascaded denoising yields better results than parallel denoising. This is because the preceding denoising stage in a cascaded process improves the image's signal-to-noise ratio (SNR), thus reducing the difficulty of subsequent denoising stages. For example, if spatial denoising is preceded by temporal denoising preprocessing, the SNR of static areas increases, reducing noise and improving sharpness. Performing spatial denoising at this point allows for better differentiation between texture and noise, resolving the trade-off between sharpness and noise reduction. Similarly, if temporal denoising is preceded by spatial denoising preprocessing, the SNR of the current image increases. Since raindrop noise is caused by noise being mistaken for motion, reducing noise reduces the raindrop noise phenomenon, making motion more noticeable, thus better addressing the trade-off between motion blur and raindrop noise. However, in a cascaded structure, one of the two steps must be performed first; the problems inherent in the filtering performed first remain unresolved.

[0007] A recursive filter is a type of IIR filter, also known as a feedback filter. Similar to traditional filtering methods, its output depends on the input and the filter response function. However, unlike traditional filters, recursive filters also depend on past outputs. Complex recursive filters based on recursive structures include Kalman filters, particle filters, and Gaussian recursive filters. One-dimensional recursive filters are used for time-domain noise reduction, two-dimensional recursive filters for spatial-domain noise reduction, and three-dimensional recursive filters for both spatiotemporal-domain noise reduction. For high-dimensional recursive filters, they are often decomposed into a series of multiple one-dimensional filters. For example, for two-dimensional recursive filtering, a horizontal one-dimensional recursive filter is performed first, followed by a vertical one-dimensional recursive filter. For three-dimensional recursive filtering, a time-domain one-dimensional recursive filter is added in series before or after the two-dimensional recursive filter. However, this decomposition method still suffers from the problem of not mitigating the issues associated with the previous spatiotemporal-domain series connection. Summary of the Invention

[0008] The purpose of this invention is to provide a three-dimensional recursive filtering method and apparatus for joint temporal and spatial domain noise reduction of images, so as to solve the problem that the filtering problem of the former is not improved by the series connection in the traditional series structure of temporal and spatial domain filtering.

[0009] To address the aforementioned technical problems, based on one aspect of the present invention, a three-dimensional recursive filtering method for joint spatiotemporal domain noise reduction of images is provided, comprising:

[0010] Step 1: Determine the points to be processed in the current image and the reference points in the reference image at the same location based on the recursive filtering direction;

[0011] Step 2: Perform spatial recursive filtering on the difference between the point to be processed in the current image and the reference point in the reference image to obtain the difference filtering result and spatial filtering coefficients.

[0012] Step 3: Perform temporal recursive filtering on the points to be processed in the current image based on the difference filtering results to obtain the updated points to be processed in the current image;

[0013] Step 4: Calculate the directional gradient of the point to be processed in the current image and the directional gradient of the reference point in the reference image;

[0014] Step 5: Based on the spatial filtering coefficients, perform temporal recursive filtering on the directional gradient of the point to be processed in the current image to obtain the temporal filtering result;

[0015] Step 6: Based on the temporal filtering result, perform spatial recursive filtering on the updated current image points to be processed to obtain the current spatial filtering result;

[0016] Step 7: Perform spatial recursive filtering on the reference points of the reference image based on the directional gradient of the reference points to obtain the reference spatial filtering result;

[0017] Step 8: Update the difference filtering result based on the current spatial domain filtering result and the reference spatial domain filtering result.

[0018] Optionally, the method further includes:

[0019] Step 9: Determine whether all points to be processed in the current image along the recursive filtering direction have completed steps 1 to 8; if yes, proceed to step 10; otherwise, find the next point to be processed in the current image and proceed to steps 1 to 8.

[0020] Step 10: Based on the updated difference filtering results of all unprocessed points in the current image, perform temporal recursive filtering on the current image and output the final filtering result.

[0021] Optionally, the reference image is the previous frame of the current image from which the final filtered result has been output.

[0022] Based on another aspect of the present invention, the present invention also provides a three-dimensional recursive filtering device for joint spatiotemporal denoising of images, comprising:

[0023] The data input unit is used to acquire the current image and the reference image, and determine the point to be processed in the current image and the reference point of the reference image at the same position according to the recursive filtering direction;

[0024] The image difference gradient calculation unit is used to calculate the difference between the point to be processed in the current image and the reference point in the reference image, and to calculate the directional gradient of the point to be processed in the current image and the directional gradient of the reference point in the reference image.

[0025] The image difference spatial recursive filtering unit is used to perform spatial recursive filtering on the difference between the point to be processed in the current image and the reference point in the reference image to obtain the difference filtering result and the spatial filtering coefficient.

[0026] The image orientation gradient temporal recursive filtering unit is used to perform temporal recursive filtering on the orientation gradient of the point to be processed in the current image based on the spatial filtering coefficients, so as to obtain the temporal filtering result.

[0027] The image temporal recursive unit is used to perform temporal recursive filtering on the points to be processed in the current image based on the difference filtering result to obtain the updated points to be processed in the current image.

[0028] The image spatial recursive filtering unit is used to perform spatial recursive filtering on the updated current image point to be processed based on the temporal filtering result to obtain the current spatial filtering result, and to perform spatial recursive filtering on the reference point of the reference image based on the directional gradient of the reference point of the reference image to obtain the reference spatial filtering result.

[0029] An image difference update unit is used to update the difference filtering result based on the current spatial domain filtering result and the reference spatial domain filtering result.

[0030] Optionally, the device further includes a judgment unit and a temporal filtering post-processing unit. The judgment unit is used to determine whether all points to be processed in the current image have updated the difference filtering result in the recursive filtering direction. If so, the temporal filtering post-processing unit performs temporal recursive filtering processing on the current image based on the updated difference filtering result of all points to be processed in the current image and outputs the final filtering result. If not, the judgment unit finds the next point to be processed in the current image and inputs it to the data input unit.

[0031] As described above, the three-dimensional recursive filtering method for image spatiotemporal joint denoising of the present invention differs from the traditional approach of breaking down three-dimensional recursive filtering into three one-dimensional filters directly connected in series. Instead, it improves the effect of spatial filtering by incorporating temporal recursive information during the point-by-point spatial recursion process, and then improves the effect of temporal filtering by incorporating spatial recursive information during the point-by-point temporal filtering process. This process is repeated recursively to traverse the entire image. Under the same level of computational logic and storage overhead, the present invention can better improve image clarity and signal-to-noise ratio compared with traditional spatiotemporal serial and parallel spatiotemporal methods, while reducing the absence of motion blur and raindrop noise.

[0032] It should be noted that the three-dimensional recursive filtering device for image spatiotemporal domain joint denoising and the three-dimensional recursive filtering method for image spatiotemporal domain joint denoising are based on the same inventive concept and have the same or corresponding specific technical features. Therefore, the three-dimensional recursive filtering device for image spatiotemporal domain joint denoising also has the technical effects of the three-dimensional recursive filtering method for image spatiotemporal domain joint denoising, which will not be repeated here. Attached Figure Description

[0033] Those skilled in the art will understand that the accompanying drawings are provided to better understand the invention and do not constitute any limitation on the scope of the invention. Wherein:

[0034] Figure 1 This is a schematic diagram of a three-dimensional recursive filtering method for joint spatiotemporal denoising of images according to an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of a three-dimensional recursive filtering device for image spatiotemporal joint noise reduction according to an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, advantages, and features of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the drawings are all in a very simplified form and are not drawn to scale, and are only used to facilitate and clarify the explanation of the embodiments of this invention. Furthermore, the structures shown in the drawings are often part of the actual structures. In particular, different figures may emphasize different aspects and may sometimes use different scales.

[0037] As used in this invention, the singular forms “a,” “an,” and “the” include plural objects; the term “or” is generally used to mean “and / or”; the term “a number” is generally used to mean “at least one”; and the term “at least two” is generally used to mean “two or more”. Furthermore, the terms “first,” “second,” and “third” are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as “first,” “second,” or “third” may explicitly or implicitly include one or at least two of that feature. “One end” and “the other end,” as well as “proximal end” and “distal end,” generally refer to two corresponding parts, including not only endpoints. The terms “installed,” “connected,” and “joined” should be interpreted broadly, for example, as a fixed connection, a detachable connection, or an integral part; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two elements or an interaction between two elements. Furthermore, as used in this invention, the phrase "one element is disposed on another element" generally only indicates that there is a connection, coupling, cooperation, or transmission relationship between the two elements, and the connection, coupling, cooperation, or transmission between the two elements can be direct or indirect through an intermediate element. It should not be construed as indicating or implying a spatial positional relationship between the two elements, i.e., one element can be located arbitrarily inside, outside, above, below, or to one side of the other element, unless otherwise explicitly stated. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0038] Figure 1 This is a schematic diagram of a three-dimensional recursive filtering method for joint spatiotemporal denoising of images according to an embodiment of the present invention. (See attached diagram.) Figure 1 In one embodiment of the present invention, a three-dimensional recursive filtering method for joint spatiotemporal domain noise reduction of images is provided. The method is used to perform three-dimensional recursive filtering on images. Specifically, the current image and a reference image are first obtained, the current image is traversed, and the image is subjected to three-dimensional recursive filtering through the following steps one to ten.

[0039] Step 1: Determine the point to be processed C(h1, w1, t1) in the current image and the reference point P(h1, w1, t0) in the reference image at the same location based on the recursive filtering direction. It's clear that "same location" here means the points to be processed in the current image and the reference image are at the same position. Understandably, the points to be processed in an image refer to pixels that require special attention or processing during image processing. These points may be noise points, edge points, feature points, etc., requiring specific processing to improve image quality or extract useful information.

[0040] Step 2: Calculate the difference between the point C(h1, w1, t1) to be processed in the current image and the reference point P(h1, w1, t0) in the reference image to obtain the difference Dif(h1, w1, t1), and perform spatial recursive filtering on the difference Dif(h1, w1, t1) to obtain the difference filtering result Dif_r(h1, w1, t1) and the spatial filtering coefficients.

[0041] Step 3: Perform time-domain recursive filtering on the point C(h1, w1, t1) to be processed in the current image based on the difference filtering result Dif_r(h1, w1, t1) to obtain the updated point C_t(h1, w1, t1) to be processed in the current image.

[0042] Step 4: Calculate the directional gradient Gc(h1, w1, t1) of the point to be processed in the current image and the directional gradient Gp(h1, w1, t0) of the reference point in the reference image.

[0043] Step 5: Based on the spatial filtering coefficients, perform temporal recursive filtering on the directional gradient Gc(h1, w1, t1) of the point to be processed in the current image to obtain the temporal filtering result Gc_t(h1, w1, t1).

[0044] Step 6: Based on the time-domain filtering result Gc_t(h1, w1, t1), perform spatial recursive filtering on the updated current image point C_t(h1, w1, t1) to obtain the current spatial filtering result C_tr(h1, w1, t1).

[0045] Step 7: Perform spatial recursive filtering on the reference point P(h1, w1, t0) of the reference image according to the directional gradient Gp(h1, w1, t0) of the reference point of the reference image to obtain the reference spatial filtering result P_r(h1, w1, t0).

[0046] Step 8: Update the difference filtering result according to the current spatial domain filtering result C_tr(h1, w1, t1) and the reference spatial domain filtering result P_r(h1, w1, t0) to obtain the updated difference filtering result Dif_rt(h1, w1, t1).

[0047] Step Nine: Determine whether all points to be processed in the current image along the recursive filtering direction have completed steps one through eight, that is, whether all points to be processed in the current image along the recursive filtering direction have undergone three-dimensional recursive filtering and obtained the updated difference filtering result; if yes, proceed to step ten; if no, find the next point to be processed in the recursive filtering direction of the current image and execute steps one through eight, thereby performing three-dimensional recursive filtering on the next point to be processed.

[0048] Step 10: Based on the updated difference filtering results of all points to be processed in the current image, perform temporal recursive filtering on the current image and output the final filtering result, thereby realizing the three-dimensional recursive filtering processing of the current image.

[0049] It should be noted that the reference image is generally selected as the image of the previous frame from which the final filtering result has been output; that is, the reference image is the image of the previous frame from which the three-dimensional recursive filtering process of this invention has been fully iterated. This selection of the reference image allows the most recently updated image to be used as the reference image for the current image, reducing the difference between the current image and the reference image, thus improving the filtering effect and the quality of image processing.

[0050] Thus, when updating the current image, the reference image, and the points to be processed for image differences, the spatial filtering information of the image differences is used to perform temporal filtering on the image gradient. Then, the temporally filtered image gradient is used to perform spatial filtering on the current image. Finally, the frame difference is filtered and updated based on the filtered current image and the reference image. This alternating temporal and spatial filtering continuously improves the signal-to-noise ratio (SNR) of the processed information. Furthermore, recursive filtering itself performs the above steps multiple times in different directions for each processing point. Ultimately, a high SNR output result can be produced, effectively handling scenarios with poor SNR. The three-dimensional recursive filtering method for image spatiotemporal joint denoising of this invention differs from the traditional approach of breaking down three-dimensional recursive filtering into three one-dimensional filters directly connected in series. Instead, it improves the effect of spatial filtering by incorporating temporal recursive information during point-by-point spatial recursion, and then improves the effect of temporal filtering by incorporating spatial recursive information during point-by-point temporal filtering. This process is repeated recursively to traverse the entire image. Under the same level of computational logic and storage overhead, this invention can better improve image clarity and signal-to-noise ratio compared to traditional spatiotemporal serial and parallel spatiotemporal methods, while reducing motion blur and raindrop noise.

[0051] Understandably, performing the three-dimensional recursive filtering process on the current image as described above includes processing in both the horizontal and vertical directions; that is, the recursive filtering direction is chosen to be either horizontal or vertical. The logical steps are the same for different recursive filtering directions, only the reference points of the selected reference images differ. The following example illustrates forward recursive filtering in the horizontal direction. Let the current image to be filtered be C0, the reference image to be filtered be P0, and the image difference to be filtered be Dif0. C1, P1, and Dif1 are the results of filtering C0, P0, and Dif0, respectively.

[0052] First, after determining the point to be processed C0(i,j) in the current image and the reference point P0(i,j) in the reference image, the formula for calculating the difference Dif0 between the point to be processed in the current image and the reference point in the reference image is as follows:

[0053] Dif0(i,j)=C0(i,j)-P0(i,j)

[0054] Spatial filtering requires determining the filtering reference points. Since it is a forward recursive filtering in the horizontal direction, for the current point to be filtered at position (i,j), the value at position (i,j-1) is the output result of the forward recursive filtering in the horizontal direction that has already been performed. Therefore, the reference points are C1(i,j-1), P1(i,j-1), and Dif1(i,j-1).

[0055] The spatial filtering coefficients Wd, representing the difference between the current image and the reference image, can be calculated using the following formula:

[0056]

[0057] sigma d Let be the filter intensity. Then, the spatial domain filtering result Dif0 representing the difference between the current image and the reference image at the current location to be processed is... R for:

[0058] Dif0 R =Dif0(i,j)*(1-Wd)+Dif1(i,j-1)*Wd

[0059] The gradients of the current image and the reference image in four directions are calculated. The gradient calculation uses a convolution kernel, which employs the Sobel operator, as detailed below:

[0060] Wg0 = [-1 1 0]

[0061] Wg1 = [0 1-1]

[0062]

[0063] This example demonstrates forward filtering in the horizontal direction. We select Wg0 to calculate the gradient, i.e., the gradient in the left direction. For the other three cases, we use the last three convolutional kernels to calculate the gradients sequentially. The formulas for calculating the left-direction gradients Gc0 and Gp0 of the current image and the reference image are as follows:

[0064] Gc0=C0(i,j)-C1(i,j-1)

[0065] Gp0 = P0(i,j) - P1(i,j-1)

[0066] The relationship between the left-direction gradient and the image difference between the current image and the reference image is easily seen from the following recursive formula:

[0067] Dif0(i,j)-Dif1(i,j-1)

[0068] =(C0(i,j)-P0(i,j))-(C1(i,j-1)-P1(i,j-1))

[0069] =(C0(i,j)-C1(i,j-1))-(P0(i,j)-P1(i,j-1))

[0070] =Gc0-Gp0

[0071] Therefore, the temporal filtering coefficients Wg for temporal filtering of the image gradient are equal to the spatial filtering coefficients Wd for the difference between the current image and the reference image. Thus, the temporal filtering result Gc0 of the left-direction gradient of the current image at the current processing position is... T for:

[0072] Gc0 T =Gc0(i,j)*(1-Wg)+Gc1(i,j-1)*Wg

[0073] Where Wg = Wd.

[0074] According to the output Dif0 R Calculate the temporal filtering result C0 of the current image at the current position to be processed. T The formula is as follows:

[0075]

[0076] C0 T =C0(i,j)*(1-Wct)+P0(i,j)*Wct

[0077] sigma ct This represents the filter strength.

[0078] Then based on the output Gc0 T The spatial filtering result C1(i,j) of the current image at the current position to be processed is calculated using the following formula:

[0079]

[0080] C1(i,j)=C0T*(1-Wcr)+C1(i,j-1)*Wcr

[0081] sigma cr This represents the filter strength.

[0082] Based on Gp0, calculate the spatial filtering result P1(i,j) of the reference image at the current location to be processed, as follows:

[0083]

[0084] P1(i,j)=P0(i,j)*(1-Wpr)+P1(i,j-1)*Wpr

[0085] Among them, sigma Pr This represents the filter strength.

[0086] The difference Dif1(i,j) between the filtered current image and the reference image at this location is calculated based on C1(i,j) and P1(i,j), as shown in the following formula:

[0087] Dif1(i,j) = C1(i,j) - P1(i,j)

[0088] Let j = j + 1, and repeat the above steps.

[0089] The operational logic for other recursive directions is the same as in the example above, only the recursive direction and input / output data differ. It's easy to see that the final outputs, Dif_final and C_final, are recursively generated by continuously referencing relevant information updated in the temporal and spatial domains. P_final is obtained from the reference image P0 through recursive filtering in the pure spatial domain. To ensure final clarity, the spatial recursive filtering strength of the reference image should be weaker than that of the current image. Once temporal filtering results in motion loss, subsequent recursion cannot compensate for this loss. Since the same point will undergo multiple temporal filtering iterations in different directions, the temporal filtering strength in step three of the recursive filtering of C_final should be as weak as possible. Considering the relatively weak temporal filtering strength in different recursive directions, and that each temporal filter uses image differences (not the final filtering result) to calculate the filtering coefficients, it is necessary to use the final filtering result of the image differences to calculate cleaner, more complete motion, and stronger temporal filtering coefficients as a finishing touch. This allows for the final temporal recursive filtering of the current image, resulting in the final filtered output. The following are the temporal filtering coefficients W. final The calculation formula is as follows:

[0090]

[0091] sigma final This represents the final time-domain filter strength.

[0092] The final filtered result output after joint spatiotemporal filtering is calculated using the following formula:

[0093] output = C_final * (1 - W) final )+P_final*W final

[0094] For example, if final is 3, following the above steps, first perform a horizontal forward filter on C0, P0, and Dif0 to obtain C1, P1, and Dif1. Then perform a horizontal reverse filter on C1, P1, and Dif1 to obtain C2, P2, and Dif2. Finally, perform a vertical forward filter on C2, P2, and Dif2 to obtain C3, P3, and Dif3.

[0095] Figure 2 This is a schematic diagram of a three-dimensional recursive filtering device for image spatiotemporal joint denoising according to an embodiment of the present invention. Based on the same inventive concept as the above-described three-dimensional recursive filtering method for image spatiotemporal joint denoising, the present invention also provides a three-dimensional recursive filtering device for image spatiotemporal joint denoising, see reference. Figure 2 The device includes:

[0096] The data input unit is used to acquire the current image and the reference image, and determine the point to be processed in the current image and the reference point of the reference image at the same position according to the recursive filtering direction;

[0097] The image difference gradient calculation unit is used to calculate the difference between the point to be processed in the current image and the reference point in the reference image, and to calculate the directional gradient of the point to be processed in the current image and the directional gradient of the reference point in the reference image.

[0098] The image difference spatial recursive filtering unit is used to perform spatial recursive filtering on the difference between the point to be processed in the current image and the reference point in the reference image to obtain the difference filtering result and the spatial filtering coefficient.

[0099] The image orientation gradient temporal recursive filtering unit is used to perform temporal recursive filtering on the orientation gradient of the point to be processed in the current image based on the spatial filtering coefficients, so as to obtain the temporal filtering result.

[0100] The image temporal recursive unit is used to perform temporal recursive filtering on the points to be processed in the current image based on the difference filtering result to obtain the updated points to be processed in the current image.

[0101] The image spatial recursive filtering unit is used to perform spatial recursive filtering on the updated current image point to be processed based on the temporal filtering result to obtain the current spatial filtering result, and to perform spatial recursive filtering on the reference point of the reference image based on the directional gradient of the reference point of the reference image to obtain the reference spatial filtering result.

[0102] An image difference update unit is used to update the difference filtering result based on the current spatial domain filtering result and the reference spatial domain filtering result.

[0103] Furthermore, the device also includes a judgment unit and a temporal filtering post-processing unit. The judgment unit is used to determine whether all the points to be processed in the current image have been updated with the difference filtering result in the recursive filtering direction. If so, the temporal filtering post-processing unit performs temporal recursive filtering processing on the current image based on the updated difference filtering result of all the points to be processed in the current image and outputs the final filtering result. If not, the judgment unit finds the next point to be processed in the current image and inputs it to the data input unit.

[0104] It should be noted that those skilled in the art can understand the three-dimensional recursive filtering device for image spatiotemporal domain joint denoising, as well as the cooperation and logical execution relationships between the various units in the device, based on the description of the three-dimensional recursive filtering method for image spatiotemporal domain joint denoising above. This will not be elaborated further here.

[0105] While the present invention has been disclosed above with reference to preferred embodiments, these embodiments are not intended to limit the invention. For any person skilled in the art, many possible variations and modifications can be made to the technical solutions of the present invention based on the disclosed technical content, or equivalent embodiments can be modified accordingly, without departing from the scope of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention, without departing from the content of the present invention, shall still fall within the scope of protection of the present invention.

Claims

1. A three-dimensional recursive filtering method for joint spatiotemporal domain noise reduction of images, characterized in that, include: Step 1: Determine the point to be processed in the current image and the reference point in the reference image at the same position according to the recursive filtering direction. The recursive filtering direction is either horizontal or vertical. Step 2: Perform spatial recursive filtering on the pixel differences between the point to be processed in the current image and the reference point in the reference image to obtain the difference filtering result and spatial filtering coefficients. Step 3: Perform temporal recursive filtering on the points to be processed in the current image based on the difference filtering results to obtain the updated points to be processed in the current image; Step 4: Calculate the directional gradient of the point to be processed in the current image and the directional gradient of the reference point in the reference image; Step 5: Based on the spatial filtering coefficients, perform temporal recursive filtering on the directional gradient of the point to be processed in the current image to obtain the temporal filtering result; Step 6: Based on the temporal filtering result, perform spatial recursive filtering on the updated current image points to be processed to obtain the current spatial filtering result; Step 7: Perform spatial recursive filtering on the reference points of the reference image based on the directional gradient of the reference points to obtain the reference spatial filtering result; Step 8: Update the difference filtering result based on the current spatial domain filtering result and the reference spatial domain filtering result; Step 9: Determine whether all points to be processed in the current image along the recursive filtering direction have completed steps 1 to 8; If yes, proceed to step ten; otherwise, find the next point to be processed in the current image and proceed to steps one through eight. Step 10: Based on the updated difference filtering results of all unprocessed points in the current image, perform temporal recursive filtering on the current image and output the final filtering result; The following formula is used to calculate the points to be processed in the current image. and reference point of reference image Differences between : The following formula is used to calculate the points to be processed in the current image. and reference point of reference image Differences between : The spatial filtering coefficients Wd in step two are calculated using the following formula: This represents the filter strength.

2. The three-dimensional recursive filtering method for joint spatiotemporal denoising of images according to claim 1, characterized in that, The reference image is the previous frame of the current image from which the final filtered result has been output.

3. A three-dimensional recursive filtering device for joint spatiotemporal noise reduction of images, characterized in that, include: The data input unit is used to acquire the current image and the reference image, and determine the point to be processed in the current image and the reference point in the reference image at the same position according to the recursive filtering direction, wherein the recursive filtering direction is either horizontal or vertical. The image difference gradient calculation unit is used to calculate the difference between the point to be processed in the current image and the reference point in the reference image, and to calculate the directional gradient of the point to be processed in the current image and the directional gradient of the reference point in the reference image. The image difference spatial recursive filtering unit is used to perform spatial recursive filtering on the difference between the point to be processed in the current image and the reference point in the reference image to obtain the difference filtering result and the spatial filtering coefficient. The image orientation gradient temporal recursive filtering unit is used to perform temporal recursive filtering on the orientation gradient of the point to be processed in the current image based on the spatial filtering coefficients, so as to obtain the temporal filtering result. The image temporal recursive unit is used to perform temporal recursive filtering on the points to be processed in the current image based on the difference filtering result to obtain the updated points to be processed in the current image. The image spatial recursive filtering unit is used to perform spatial recursive filtering on the updated current image point to be processed based on the temporal filtering result to obtain the current spatial filtering result, and to perform spatial recursive filtering on the reference point of the reference image based on the directional gradient of the reference point of the reference image to obtain the reference spatial filtering result. An image difference update unit is used to update the difference filtering result based on the current spatial domain filtering result and the reference spatial domain filtering result; The device further includes a judgment unit and a temporal filtering post-processing unit. The judgment unit is used to determine whether all the points to be processed in the current image have been updated with the difference filtering result in the recursive filtering direction. If so, the temporal filtering post-processing unit performs temporal recursive filtering on the current image based on the updated difference filtering result of all the points to be processed in the current image and outputs the final filtering result. If not, the judgment unit finds the next point to be processed in the current image and inputs it to the data input unit. The following formula is used to calculate the points to be processed in the current image. and reference point of the reference image Differences between : The following formula is used to calculate the points to be processed in the current image. and reference point of reference image Differences between : The spatial filtering coefficients Wd in step two are calculated using the following formula: This represents the filter strength.

Images