Stream measurement video noise suppression method based on space-time quadrilateral filter

Abstract

The invention relates to the technical field of video noise suppression, in particular to a flow measurement video noise suppression method based on a space-time quadrilateral filter. According to the method, a quadrilateral filter is constructed by combining a time similarity weight, a pulse weight and a bilateral filter; performing pixel fusion on the filtering window area on the target frame by combining the adjacent frame corresponding to the filtering window, then filtering the fused filtering window area on the target frame by adopting a quadrilateral filter to obtain a de-noised frame of the filtering window area, and then replacing each filtering window area on the target frame with the corresponding de-noised frame. Therefore, the denoised image frame is formed. According to the method, the defects that time and space information cannot be effectively combined and video frame time and space correlation cannot be reserved in the existing denoising technology are overcome, the time information of adjacent frames and the space structure characteristics of a target frame are combined, the time and space information of a water flow video is fully excavated and utilized, speckles and impulse noise are effectively smoothed, and the denoising effect is improved. Meanwhile, the edge and texture details of the image are reserved, and the noise reduction effect is remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of video noise suppression, in particular to a method for suppressing noise in a flow measurement video based on a spatiotemporal quadrilateral filter. Background Art

[0002] Video detection of water velocity is an advanced technology that combines video imaging and fluid dynamics analysis. It can continuously and high-resolution monitor water velocity under various hydrological conditions, and has the characteristics of all-weather and all-day. However, the noise problem in water flow video has caused great interference to the subsequent processing and application of the image, affecting the accuracy of water velocity measurement and the reliability of hydrological analysis.

[0003] Currently, many denoising algorithms are used to suppress and smooth noise in water flow videos, including the BM3D algorithm and algorithms based on partial differential equations (PDE). The BM3D algorithm performs well in reducing additive white Gaussian noise, but has limited effect in dealing with complex noise in water flow videos. The improved VBM3D filter by Li et al. processes adjacent image frames through temporal averaging. Although it improves the denoising effect, it does not fully utilize the temporal correlation between frames. The PDE method is often used for image smoothing and edge detail preservation, but it introduces blur in water flow videos.

[0004] Bilateral filter is a nonlinear filtering technique that can be improved according to the needs of different scenarios. The trilateral filter proposed by Roman Garnett et al. achieves universal noise reduction through a simple pulse detector. Liu et al. developed an improved trilateral filter that combines pulse detectors in the gradient domain and the intensity domain and adopts a switching mechanism to improve the removal of Gaussian and impulse noise. Xu Guangyu et al. introduced the EC-ROAD statistic into the trilateral filter to enhance the pulse detection performance. The trilateral filter proposed by Ai et al. uses temporal redundant information to further explore the similarities between video frames. These methods have alleviated the noise problem to a certain extent, but there is still a challenge to be solved on how to better utilize temporal and spatial information for image smoothing and denoising. Summary of the invention

[0005] In order to overcome the defects of the above-mentioned denoising technology in the prior art that it cannot effectively combine time and space information and is not conducive to retaining the time and space correlation of video frames, the present invention proposes a flow measurement video noise suppression method based on a spatiotemporal quadrilateral filter, which combines the time information of adjacent frames and the spatial structure characteristics of the target frame, fully mines and utilizes the spatiotemporal information of the water flow video, effectively smoothes the speckle and impulse noise, and retains the edge and texture details of the image, thereby significantly improving the noise reduction effect.

[0006] The present invention proposes a method for suppressing video noise based on a spatiotemporal quadrilateral filter. First, the acquired water flow video is decomposed frame by frame, an adaptive strategy is adopted to select the target frame to be processed, and the adjacent frame range of the target frame is calculated under a specified filtering window; the adjacent frames are highly similar to the target frame and are continuously distributed in time;

[0007] Calculate the temporal similarity weights between the target frame and the adjacent frames, and calculate the impulse weight for detecting impulse noise in combination with the ROAD value of the target frame under the filter window;

[0008] A quadrilateral filter is constructed by combining the temporal similarity weight, the pulse weight and the bilateral filter; the pixel fusion of the filter window area on the target frame is performed in combination with the adjacent frames corresponding to the filter window, and then the fused filter window area on the target frame is filtered using the quadrilateral filter to obtain a denoised frame of the filter window area;

[0009] The filter window is made to traverse the target frame to obtain the denoised frames of each filter window area, and then each filter window area on the target frame is replaced with the corresponding denoised frame to form a denoised image frame.

[0010] Preferably, let the adjacent frames T n With the target frame T i The temporal similarity weight is the temporal weight Gaussian kernel function w t (i,n) indicates that n is the adjacent frame T n The order after video decomposition, i is the order of the target frame after video decomposition;

[0011]

[0012] Where exp represents the exponential function, σ t is the set time similarity diffusion factor; I is the total number of frames after video decomposition.

[0013] Preferably, the calculation of the ROAD value is based on absolute difference ranking statistics of the target frame image.

[0014] Preferably, the ROAD value of the target frame under the specified filtering window is calculated as follows: determine the central pixel point of the filtering window area on the target frame and the absolute grayscale difference between the central pixel point and each neighboring pixel point, take the larger m grayscale absolute differences and sum them as the ROAD value, where m is a set value.

[0015] Preferably, the pulse weight w under the specified filter window Y (x 0 ) is calculated as follows:

[0016]

[0017] Among them, σ tis the set time similarity diffusion factor; ROAD m (x 0 ) is the calculated ROAD value, x 0 It is the center pixel of the specified filter window area on the target frame.

[0018] Preferably, the video frame image set after the water flow video is decomposed frame by frame is denoted as T; the adjacent frame range of the target frame is [Z 1 ,Z 2 ],Z 1 is the sequence number of the adjacent frame in set T that is before the target frame and farthest from the target frame, Z 2 is the sequence number of the adjacent frame in the set T that is after the target frame and farthest from the target frame; the kernel function w of the quadrilateral filter constructed by combining the temporal similarity weight, the pulse weight and the bilateral filter is:

[0019]

[0020] Among them, w d represents the geometric space weight function of the bilateral filter, w r represents the grayscale spatial weight function of the bilateral filter; x(n) represents the pixel point on the nth video frame in the set T, x 0 (n) represents the central pixel of the nth video frame in the set T under the specified filter window; Ω(n) represents x 0 (n); Y(x(n)) and Y(x 0 (n)) represent the pixel points x(n) and x 0 Gray value of (n); w t (i,n) represents the temporal similarity weight between the target frame and the nth video frame in the set T; J(x,x 0 ) represents the switching function of the filter control; σ J is the similarity diffusion factor of the set ROAD function; ROAD m (x 0 ,cen) is the median of the larger m values ​​of the grayscale absolute difference between the central pixel and each neighboring pixel in the filter window area on the target frame.

[0021] Preferably, the filter window area on the target frame is pixel-fused in combination with the adjacent frames corresponding to the filter window to obtain a fused target frame Y' under the filter window, which is expressed as:

[0022] Y′={Y′(x)|x∈Ω}

[0023]

[0024] Among them, x is the pixel point in the filter window on the target frame, x nis the pixel corresponding to x in the nth video frame in the set T; Ω is the filter window area; Y'(x) is the gray value of the pixel x on the fusion target frame Y'; Y(x n ) is the pixel x n The gray value of .

[0025] Preferably, the fused filter window area on the target frame is filtered using a four-sided filter to obtain a denoised frame of the filter window area. The formula is:

[0026]

[0027] Among them, w d (x,x 0 ) represents the combination of x and x 0 w is the boundary d Function value, w r (Y'(x),Y'(x 0 )) represents the relationship between Y'(x) and Y'(x 0 ) is the boundary w r Function value; w Y (x 0 ) is the pulse weight.

[0028] A system proposed in the present invention is characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and the processor is connected to the memory, and the processor is used to execute the computer program to implement the video noise suppression method based on the spatiotemporal quadrilateral filter.

[0029] The present invention provides a storage medium storing a computer program, wherein the computer program is used to implement the video noise suppression method based on a spatiotemporal quadrilateral filter when executed.

[0030] The advantages of the present invention are:

[0031] (1) The present invention fully exploits the temporal correlation between adjacent frames of the video image and the spatial structure information of the target frame by evaluating the similarity of the video frames and the ROAD statistics of the target frame images, thereby maintaining the edge and texture details of the image when removing speckle and impulse noise in the water flow video frame images.

[0032] (2) Before performing four-sided filtering, the present invention evaluates the similarity between the target frame and the adjacent frames by reasonably setting the threshold, selects the appropriate adjacent frame range, and determines the time weight of the adjacent frames to the target frame filtering window. This method makes full use of the similarity information between frames, and participates in the construction of the four-sided filter weight kernel through the time weight, which effectively overcomes the shortcomings of the traditional bilateral filter in removing image speckle noise. The present invention effectively smoothes the noise in the target frame image by combining the temporal and spatial information of the video data. In addition, the impulse weight is constructed using the ROAD statistic of the target frame image, and the four-sided filter weight kernel is further optimized, which solves the defects of the traditional bilateral filter in detecting and removing image impulse noise.

[0033] (3) The present invention integrates the spatiotemporal data of the target frame and adjacent frames, deeply analyzes the spatial similarity information of the ROAD value of the target frame image, and greatly improves the smoothing effect of speckle and impulse noise in the video image, while retaining the texture details of the image as much as possible. This method provides support for the credibility of water flow velocity measurement of video data. The denoised image can be used as the basis for video data in a variety of research and applications.

[0034] (4) The present invention introduces a four-sided filter for video image denoising, which can effectively reduce speckle and impulse noise. This filter adopts a four-dimensional weighted model, combining the temporal information of adjacent frames, the geometric and grayscale similarity of pixels within the frame, and the spatial structural similarity, so as to maintain image details while reducing noise. Through an adaptive frame selection strategy, the number of frames used for speckle noise filtering of the target frame is flexibly selected according to the similarity between adjacent frames and the target frame, so as to fully exploit the temporal correlation of the video data. In addition, by calculating the ROAD value of the image, the impulse weights related to the ROAD are generated to characterize the spatial structural information of the image, so as to retain the edge and texture of the image while eliminating noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic flow chart of a method for suppressing noise in a video stream based on a spatiotemporal quadrilateral filter according to the present invention;

[0036] Figure 2 It is a flow chart of a method for suppressing noise in a video stream based on a spatiotemporal quadrilateral filter provided by the present invention;

[0037] Figure 3 It is a four-edge (dimensional) weight kernel function composition diagram of the present invention;

[0038] Figure 4 It is a quantitative evaluation of the speckle noise removal effect of the four-side filtering method based on spatiotemporal information on video images at different thresholds T of the present invention;

[0039] Figure 5It is a comparison chart of the filtering results of the 149th frame image of the experimental water flow video by different algorithms; wherein a is the original image, b is the image of a after filtering by a filtering method based on a fourth-order partial differential equation (PDE), c is the image of a after filtering by an enhanced VBM3D filtering algorithm, d is the image of a after filtering by a trilateral filter (TF), and e is the image of a after denoising by the flow measurement video noise suppression method based on a spatiotemporal quadrilateral filter provided by the present invention; f, g, h, i, and j are local texture images of a, b, c, e, and d, respectively. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Reference Figure 1 , Figure 2 The method for suppressing noise in a video stream based on a spatiotemporal quadrilateral filter proposed in this embodiment includes the following steps:

[0042] S1. Obtain water flow video and decompose it into video frames. The video frame image set T = {T 1 ,T 2 ,…,T h ,…,T H}, T i Represents the hth video frame, 1≤h≤H; H represents the total number of video frames.

[0043] Combine the following steps to process the video frame T i To denoise, 1 <i<H;

[0044] S2, T i As the target frame, slide the filter window on the target frame, the filter window size is a×a, and determine the adjacent frame range of the target frame under each filter window [Z 1 ,Z 2 ]; in the video frame image set T, from the target frame T i Continuous forward Z 1 The frame is in the specified filter window and the target frame T i The gray value distances of the target frame T are all less than the set gray value threshold. i Continuous Z 2 The frame is in the specified filter window and the target frame T i The gray value distances are all less than the set gray threshold.

[0045] Specify the target frame T under the filter window iThe adjacent frame range [Z 1 ,Z 2 The steps to determine ] are as follows.

[0046] S21, Initialize Z 1 =i-1, Z 2 =i+1;

[0047] S22, calculate target frame T i The gray value under the specified filter window and the Zth gray value in the set T 1 The Euclidean distance of the grayscale value under the specified filter window of the video frame

[0048]

[0049] Among them, x i (j1,j2) represents the target frame T i The gray value of the pixel at row j1 and column j1 in the specified filter window. Indicates the Zth 1 The gray value of the pixel at the j1th row and j1th column in the specified filtering window of the video frame;

[0050] S23, if If it is less than the gray threshold, then Z 1 Update to Z 1 -1, then return to step S22; if If it is greater than or equal to the grayscale threshold, then Z 1 Update to Z 1 +1 and confirmed.

[0051] It is worth noting that

[0052] S24, calculate target frame T i The gray value under the specified filter window and the Zth gray value in the set T 2 The Euclidean distance of the grayscale value under the specified filter window of the video frame

[0053]

[0054] Among them, x i (j1,j2) represents the target frame T i The gray value of the pixel at row j1 and column j1 in the specified filter window. Indicates the Zth 2 The grayscale value of the pixel at the j1th row and j1th column in the specified filtering window of the video frame.

[0055] S25, if If it is less than the grayscale threshold, then Z 2 Update to Z 2+1, then return to step S22; if If it is greater than or equal to the grayscale threshold, then Z 2 Update to Z 2 -1 and ok.

[0056] It is worth noting that the grayscale threshold is the value that minimizes the mean square error (MSE) and mean absolute error (MAE) of the adjacent frame range. The grayscale thresholds corresponding to different target frames may be different or the same.

[0057] When step S2 is implemented, different grayscale thresholds can be set for the target frame and the adjacent frame range can be calculated, and then the adjacent frame range with the smallest mean square error (MSE) and mean absolute error (MAE) is obtained as the adjacent frame range finally determined by the target frame.

[0058] S3, let the target frame be the center pixel x under the specified filter window 0 The gray value is recorded as Y(x 0 ); Let the center pixel x 0 The neighborhood R(x 0 ) is x 0 Let the neighborhood R(x 0 )The grayscale value of the pixel x in is denoted as Y(x).

[0059] S4. Calculate the neighborhood R(x 0 ) is the absolute difference d(x,x 0 ), the neighborhood R(x 0 ) the largest m absolute differences d(x,x 0 ) is summed as the ROAD value corresponding to the specified filter window, denoted as ROAD m (x 0 ), neighborhood R(x 0 ) the largest m absolute differences d(x,x 0 ) is recorded as ROAD m (x 0 ,cen); combined with ROAD m (x 0 ) Calculate the pulse weight w under the filter window Y (x 0 );

[0060] d(x,x 0 )=|Y(x)-Y(x 0 )|

[0061]

[0062] Among them, σ tis the set time similarity diffusion factor; in the specific embodiment, let c=3, m=4.

[0063] S5, combined with bilateral filter and time-weighted Gaussian kernel function w t (i,n) and w Y (x 0 ) construct a quadrilateral filter kernel function;

[0064]

[0065] Among them, w d represents the geometric space weight function of the bilateral filter, w r represents the grayscale spatial weight function of the bilateral filter; x(n) represents the pixel point on the nth video frame in the set T, x 0 (n) represents the central pixel of the nth video frame in the set T under the specified filter window; Ω(n) represents the pixel x on the nth video frame in the set T. 0 (n) is the pixel area with a size of (2c+1)×(2c+1) at the center point; Y(x(n)) represents the grayscale value of the pixel point x(n), and Y(x 0 (n)) represents the pixel x 0 Gray value of (n); w d (x(n),x 0 (n)) represents the sum of x(n) and x 0 (n) is the boundary w d Function value, w r (Y(x(n)),Y(x 0 (n))) represents the sum of Y(x(n)) and Y(x 0 (n)) is the boundary w r Function value; J(x,x 0 ) represents the switching function of the filter control.

[0066]

[0067] Among them, σ J is the similar diffusion factor of the ROAD function, which is used to control J(x,x 0 ) function value;

[0068]

[0069] Where i is the subscript of the target frame, that is, the order of the target frame in the video frame image set T; n is the subscript of the adjacent frame, that is, the order of the adjacent frames in the video frame image set T; Z1≤n≤Z2; σ t is the set time similarity diffusion factor.

[0070] S6. Combine adjacent frames for fusion under the specified filter window to obtain a fusion target frame Y′={Y′(x)|x∈Ω} under the specified filter window;

[0071]

[0072] Where x is the pixel in the specified filter window on the target frame, Y'(x) is the grayscale value of the pixel x on the fused target frame Y'; x n is the pixel corresponding to x in the nth video frame in the set T, Y(x n ) is the pixel x n Gray value of; Ω is the specified filter window area.

[0073] S7, using a four-sided filter to denoise the fused target frame Y' under each filter window, to obtain a denoised frame Set the target frame T i Each filter window area above is replaced by the corresponding denoised frame To form a denoised image frame

[0074]

[0075] Among them, w d (x,x 0 ) represents the sum of x and x 0 w is the boundary d Function value, w r (Y'(x),Y'(x 0 )) represents the relationship between Y'(x) and Y'(x 0 ) is the boundary w r Function value; w Y (x 0 ) is the pulse weight.

[0076] The following is a verification of the above-mentioned method for suppressing noise in streaming video based on spatiotemporal quadrilateral filter in conjunction with a specific embodiment.

[0077] This embodiment uses real video data for verification. The video data contains 1190 frames with a frame rate of 30 Hz. The video frame rate is high and the lens has no obvious changes. For adjacent frames, there is no need to consider the image registration problem caused by the change of viewing angle. Therefore, it is suitable for verifying the proposed video noise suppression method based on spatiotemporal quadrilateral filter.

[0078] In this embodiment, the filter window size a is set to 3, and the time similarity diffusion factor σ t The diffusion factor σ is similar to the ROAD function J Both are set to 0.5.

[0079] This embodiment uses the equivalent view number ENL to quantitatively analyze the performance of the method proposed in the present invention.

[0080]

[0081] In formula (1), E(I) and Var(I) represent the mean and variance of a given image or region, respectively.

[0082] In order to illustrate the superiority of the method proposed in the present invention, several common video image filtering methods are selected for comparison, including a filtering method based on a fourth-order partial differential equation (PDE), an enhanced VBM3D filtering algorithm, and a trilateral filter (TF).

[0083] Table 1 Performance evaluation index (ENL) of video noise suppression method based on spatiotemporal quadrilateral filter

[0084]

[0085]

[0086] Combination Figure 5 And Table 1 for analysis, such as Figure 5 As shown, the first column of images are the overall effects of different filtering algorithms, and the second column of images selects representative texture detail areas from the first column of different filtering results for comparison.

[0087] The experimental conclusions of this embodiment are as follows:

[0088] (1) PDE only processes a single video image, and the TF algorithm is not very effective in processing strong speckle noise. The enhanced VBM3D filter is based on the VBM3D filter. It takes into account the temporal information in the video and specifically performs temporal averaging before VBM3D filtering to remove noise. The image obtained by filtering has a higher ENL value and better noise smoothing effect. They can achieve good noise reduction effects, but due to ignoring the temporal related information and the similarity of ROAD statistics, they cannot well preserve the details of the texture and edge areas in the image to a certain extent.

[0089] (2) Based on the traditional bilateral filtering theory, the method of the present invention further utilizes the time correlation and similarity information between adjacent frame images and the similarity information related to the ROAD statistic in the target frame image to perform four-sided filtering. As shown in the data in Table 1, the four-sided filter achieves a relatively ideal noise smoothing capability. The ENL value obtained by the denoised image is increased by 25% on average, and the speckle noise and impulse noise in the data can be well suppressed. In addition, by Figure 5 From the observations in (f)-(j), the method of the present invention achieves the best detail preservation effect in noise suppression of the target frame image.

[0090] (3) The running time of the quadrilateral filter proposed in the present invention is shorter than that of the enhanced VBM3D algorithm, but slightly longer than that of PDE and TF. The quadrilateral filter needs to extract the temporal similarity information between adjacent frame images and apply the pulse detector to the target frame image, so the required running time will increase slightly.

[0091] (4) In summary, the quadrilateral filter newly proposed in the present invention makes full use of the time information between adjacent frame images and applies a pulse detector related to the ROAD value, thereby achieving better noise suppression performance and relatively satisfactory operating efficiency.

[0092] Of course, it is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any reference numerals in the claims should not be regarded as limiting the claims involved.

[0093] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0094] The techniques, shapes, and structural parts not described in detail in the present invention are all well-known techniques.

Claims

1. A video noise suppression method based on spatiotemporal quadrilateral filter, characterized in that: Firstly, the acquired water flow video is decomposed frame by frame, and the target frame to be processed is selected by an adaptive strategy, and the adjacent frame range of the target frame is calculated under the specified filter window; the adjacent frames are highly similar to the target frame and are continuously distributed in time; Calculate the temporal similarity weights between the target frame and the adjacent frames, and calculate the impulse weight for detecting impulse noise in combination with the ROAD value of the target frame under the filter window; Combining temporal similarity weights, impulse weights, and bilateral filters to construct a quadrilateral filter; The pixel fusion of the filter window area on the target frame is performed in combination with the adjacent frames corresponding to the filter window, and then the fused filter window area on the target frame is filtered using a four-sided filter to obtain a denoised frame of the filter window area; The filter window is made to traverse the target frame to obtain the denoised frames of each filter window area, and then each filter window area on the target frame is replaced with the corresponding denoised frame to form a denoised image frame.

2. The video noise suppression method based on spatiotemporal quadrilateral filter according to claim 1, characterized in that: Let the adjacent frames T n With the target frame T i The temporal similarity weight is the temporal weight Gaussian kernel function w t (i,n) indicates that n is the adjacent frame T n The order after video decomposition, i is the order of the target frame after video decomposition; Where exp represents the exponential function, σ t is the set time similarity diffusion factor; I is the total number of frames after video decomposition.

3. The video noise suppression method based on spatiotemporal quadrilateral filter as claimed in claim 1, characterized in that: The calculation of the ROAD value is based on the absolute difference ranking statistics of the target frame image.

4. The video noise suppression method based on spatiotemporal quadrilateral filter as claimed in claim 3, characterized in that: The ROAD value of the target frame under the specified filter window is calculated as follows: determine the central pixel of the filter window area on the target frame and the absolute grayscale difference between the central pixel and each neighboring pixel, and take the larger m grayscale absolute differences as the ROAD value, where m is the set value.

5. The video noise suppression method based on spatiotemporal quadrilateral filter as claimed in claim 4, characterized in that: The pulse weight w under the specified filter window Y The calculation formula for (x0) is as follows: Among them, σ t is the set time similarity diffusion factor; ROAD m (x0) is the calculated ROAD value, and x0 is the center pixel of the specified filter window area on the target frame.

6. The video noise suppression method based on spatiotemporal quadrilateral filter as claimed in claim 4, characterized in that: Let the video frame image set after the water flow video is decomposed frame by frame be denoted as T; the adjacent frame range of the target frame is [Z1, Z2], Z1 is the adjacent frame number before the target frame and farthest from the target frame in the set T, and Z2 is the adjacent frame number after the target frame and farthest from the target frame in the set T; the kernel function w of the quadrilateral filter constructed by combining the temporal similarity weight, the impulse weight and the bilateral filter is: Among them, w d represents the geometric space weight function of the bilateral filter, w r represents the grayscale spatial weight function of the bilateral filter; x(n) represents the pixel point on the nth video frame in the set T, x0(n) represents the central pixel point of the nth video frame in the set T under the specified filter window; Ω(n) represents the neighborhood of x0(n); Y(x(n)) and Y(x0(n)) represent the grayscale values ​​of the pixels x(n) and x0(n), respectively; w t (i,n) represents the temporal similarity weight between the target frame and the nth video frame in the set T; J(x,x0) represents the switching function of the filter control; σ J is the similarity diffusion factor of the set ROAD function; ROAD m (x0,cen) is the median of the larger m values ​​of the central pixel of the filtering window area on the target frame and the absolute difference in grayscale between the central pixel and each neighboring pixel.

7. The video noise suppression method based on spatiotemporal quadrilateral filter as claimed in claim 6, characterized in that: Combine the adjacent frames corresponding to the filter window to perform pixel fusion on the filter window area on the target frame to obtain the fused target frame Y' under the filter window. The formula is expressed as: Y′={Y′(x)|x∈Ω} Among them, x is the pixel point in the filter window on the target frame, x n is the pixel corresponding to x in the nth video frame in the set T; Ω is the filter window area; Y'(x) is the gray value of the pixel x on the fusion target frame Y'; Y(x n ) is the pixel x n The gray value of .

8. The video noise suppression method based on spatiotemporal quadrilateral filter as claimed in claim 7, characterized in that: The fused filter window area on the target frame is filtered using a quadrilateral filter to obtain a denoised frame in the filter window area. The formula is: Among them, w d (x,x0) represents w bounded by x and x0 d Function value, w r (Y'(x),Y'(x0)) represents w bounded by Y'(x) and Y'(x0) r Function value; w Y (x0) is the pulse weight.

9. A system for implementing the video noise suppression method based on spatiotemporal quadrilateral filter as claimed in any one of claims 1 to 8, characterized in that: The invention comprises a memory and a processor, wherein a computer program is stored in the memory, the processor is connected to the memory, and the processor is used to execute the computer program to implement the video noise suppression method based on the spatiotemporal quadrilateral filter as described in any one of claims 1 to 8.

10. A storage medium, characterized in that: A computer program is stored, and when the computer program is executed, it is used to implement the video noise suppression method based on the spatiotemporal quadrilateral filter as described in any one of claims 1 to 8.

Images