Ultrahigh-definition video noise reduction method

Abstract

The invention relates to the technical field of video noise reduction, in particular to an ultra-high-definition video noise reduction method. The method comprises the following steps: acquiring video data and noise variance; obtaining a gray threshold and a gradient threshold based on the gray value and the gradient value of each frame of image pixel point of the video data, obtaining the region of each frame of image, and then determining a growth criterion through the gray difference and the gradient difference to block the image; determining a gray feature weight and a texture feature weight based on the difference between the gray value and the gray limit value and the difference between the gradient value and the gradient limit value; improving the original DCT coefficient based on the feature weight and the texture feature weight to construct an exclusive dictionary, and denoising, decoding and reconstructing the image based on the exclusive dictionary; and synthesizing each frame of reconstructed image into a complete video. According to the invention, optimal noise reduction of the video image frame is realized and detail information of the video is reserved to the greatest extent.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of video noise reduction, in particular to an ultra-high-definition video noise reduction method. BACKGROUND

[0002] The ultra-high-definition video imaging technology in a micro-light environment has important application value in the fields of public security monitoring, military night investigation, automatic driving night vision system and the like. Under low-illumination conditions, due to insufficient photons, the signal collected by the imaging device is weak, resulting in problems such as large noise, low contrast and fuzzy details in the video image, which seriously affects the practicability of the video and the accuracy of subsequent analysis. Especially in security monitoring, the noise reduction and enhancement of micro-light video are directly related to the extraction of key information. Therefore, it is of great significance to reduce the noise of ultra-high-definition video in a micro-light scene while retaining the weak texture and detail feature information in the dark area.

[0003] However, traditional noise reduction algorithms such as Gaussian filtering and non-local mean denoising are usually based on global smoothing strategies and are difficult to adapt to the characteristics of uneven noise distribution and low signal-to-noise ratio in a micro-light scene. While suppressing noise, such methods often excessively smooth the weak texture area, resulting in edge blurring and detail loss. In addition, the noise in the dark area of the micro-light video is more complex, and traditional algorithms are difficult to balance between noise reduction and retaining weak texture and detail information in the dark area. SUMMARY

[0004] In order to solve the technical problem of imbalance between noise reduction and detail information retention, the application provides an ultra-high-definition video noise reduction method, and the technical scheme adopted is as follows:

[0005] The application provides an ultra-high-definition video noise reduction method, which comprises the following steps:

[0006] Obtaining pre-processed video data and noise variance;

[0007] Obtaining each image of the video data, calculating the gradient value of each pixel point based on the gray value of the image, obtaining the brightness histogram of the image based on the gray value, obtaining the gray threshold value, obtaining the gradient threshold value based on the threshold segmentation method, dividing the pixel points into different regions based on the gray threshold value and the gradient threshold value, obtaining different combinations based on the different regions of the pixel points, determining the growth criterion based on the difference between the gray value of the pixel point and the gray value in its neighborhood, and the difference between the gradient value of the pixel point and the gradient value in its neighborhood, and dividing each image into blocks by the growth criterion of the seed point in each combination.

[0008] The original DCT coefficient of each pixel point is obtained based on a discrete cosine transform; a gray feature weight is determined based on the difference between the gray value of each pixel point and the gray limit value of the combination in which the pixel point is located; a texture feature weight is determined based on the difference between the gradient value of each pixel point and the gradient limit value of the combination in which the pixel point is located; the original DCT coefficient is modified based on the gray feature weight and the texture feature weight of the pixel point to obtain an improved DCT coefficient; the improved DCT coefficient of each combination pixel point is used to construct a dedicated dictionary, and an orthogonal matching pursuit algorithm is used to denoise each combination image block based on the dedicated dictionary and noise variance; and the denoised image block is decoded and aggregated to reconstruct an image.

[0009] Each reconstructed image frame is synthesized into a complete video.

[0010] In the above scheme, the scene region adaptive orthogonal matching pursuit algorithm is used to perform denoising processing on the ultra-high definition video in a low-light scene. The image is blocked according to the difference in gray and texture features of different regions of the image frame, differential processing coding is performed on different image blocks (such as high-noise areas, weak-texture areas, and dark areas), a scene region dedicated dictionary is constructed to represent different regions, and in-region denoising is performed on different regions, so that the optimal denoising of the video image frame is achieved while the detail information of the video is maximally preserved, and the denoising processing of the ultra-high definition video in the low-light scene is achieved.

[0011] In one embodiment, the gray threshold is the 25% and 75% quantile of the brightness histogram, and the gradient threshold is obtained by using an Otsu multi-threshold segmentation method on the gradient values of all pixel points.

[0012] In one embodiment, the method of dividing the pixel points into different regions based on the gray threshold and the gradient threshold is as follows:

[0013] If the pixel point belongs to a dark region; the pixel point belongs to a medium brightness region; the pixel point belongs to a bright region; wherein Gp is the gray value of the pixel point p, 25% quantile, 75% quantile.

[0014] If the pixel point belongs to a smooth region; the pixel point belongs to a weak texture region; the pixel point belongs to a high texture region; wherein is the smallest threshold value obtained by the Otsu multi-threshold segmentation method, is the largest threshold value obtained by the Otsu multi-threshold segmentation method.

[0015] In one embodiment, the combination refers to the intersection of the pixel points of different texture regions and the pixel points of different light-dark regions; the texture regions include smooth regions, weak texture regions and high texture regions; the light-dark regions include dark regions, medium brightness regions and bright regions.

[0016] In one embodiment, the function expression of the growth criterion is:

[0017] , represents the gray value of the seed point, represents the gray value of the qth pixel point in the eight-neighborhood of the seed point, represents the gradient value of the seed point, represents the gradient value of the qth pixel point in the eight-neighborhood of the seed point, represents the difference function, and respectively represent the preset threshold values of the gray value and the gradient value, represents the meaning of and, is a function expression.

[0018] In one embodiment, the gray feature weight is negatively correlated with the difference between the gray value of each pixel point and the gray limit value of the combination in which the pixel point is located.

[0019] In one embodiment, the texture feature weight is positively correlated with the difference between the gradient value of each pixel point and the gradient limit value of the combination in which the pixel point is located.

[0020] In one embodiment, the method for obtaining the gray limit value is:

[0021] When the combination corresponds to a dark region, the gray limit value is ; when the combination corresponds to a medium brightness region, the gray limit value is the mean value of and ; and when the combination corresponds to a bright region, the gray limit value is .

[0022] In one embodiment, the method for obtaining the gradient limit value is:

[0023] When the combination corresponds to a smooth texture region, the gradient limit value is ; when the combination corresponds to a weak texture region, the gradient limit value is the mean value of and ; and when the combination corresponds to a high texture region, the gradient limit value is .

[0024] In one embodiment, the improved DCT coefficient is positively correlated with the gray feature weight and the texture feature weight.

[0025] The beneficial effects of the present application are:

[0026] The present application carries out noise reduction processing on the ultra-high-definition video under the micro-light scene through a scene region adaptive orthogonal matching pursuit algorithm, blocks the image according to the differences in the gray scale and texture features of different regions of the image frame, realizes the differential processing and coding of different image blocks (such as high-noise areas, weak-texture areas, and dark areas), constructs a scene region exclusive dictionary representation, carries out in-region noise reduction on different regions, thereby realizing the optimal noise reduction of the video image frame and the maximum retention of the detail information of the video, and realizing the noise reduction processing of the ultra-high-definition video under the micro-light scene. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0028] Figure 1 A flowchart of an ultra-high-definition video noise reduction method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific embodiments, structures, features and effects of the ultra-high-definition video noise reduction method according to the present application are described in detail below in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0031] An embodiment of an ultra-high-definition video noise reduction method:

[0032] The specific scheme of the ultra-high-definition video noise reduction method provided by the present application is specifically described below in combination with the drawings.

[0033] Please refer to Figure 1 which shows a flowchart of an ultra-high-definition video noise reduction method provided by an embodiment of the present application. The method includes the following steps:

[0034] Step S001, acquiring video data and noise variance.

[0035] By shooting super high-definition video data in a low-light environment with a high-sensitivity camera, the original video is ensured to have a high resolution and dynamic range, so as to retain more scene details.

[0036] The frame buffer mechanism is used to store the continuous frames of the preprocessed video in the cache queue for subsequent analysis of each frame of the video. The frame buffer mechanism is a prior art, and its specific process will not be described again.

[0037] For each frame of image, the characteristics of the video noise under the low-light condition are counted using the Poisson-Gaussian mixed model-based method to establish a noise statistical model to obtain the noise variance, providing prior knowledge for the subsequent noise reduction algorithm. The method of establishing the noise statistical model is a prior art, and its specific process will not be described again. Then, the gamma correction is used to correct the dark area of the collected super high-definition video to improve the visibility of the dark area and avoid the failure of the algorithm to reduce the noise in the dark area due to the too weak signal in the dark area. The gamma correction is a prior art, and its specific process will not be described again. Finally, the average value method is used for gray-scale processing of each frame of image for subsequent noise reduction analysis. The average value method is a prior art, and its specific process will not be described again.

[0038] At this point, the preprocessed video data and the noise variance under the low-light scene are obtained.

[0039] In step S002, the gray threshold and gradient threshold are obtained based on the gray value and gradient value of each pixel point of each frame of image of the video data, and the region of each frame of image is obtained, and then the image is divided into blocks by determining the growth criterion through the gray difference and the gradient difference.

[0040] The video data under the low-light scene has a significant low signal-to-noise ratio characteristic, and the noise distribution is complex and non-uniform, including both Poisson noise related to the signal and Gaussian noise introduced by the camera sensor, resulting in serious degradation of image quality.

[0041] Among them, the detail preservation of weak texture area and dark area is the core challenge. Due to insufficient light, the signal-to-noise ratio of weak texture area (such as fine patterns on the surface of objects and edge transition zone) is extremely low, and traditional algorithms are prone to misjudgment of these low-frequency weak signals as noise and excessive smoothing, resulting in loss of key information. The dark area (such as shadows and low-reflective objects) not only has more serious noise interference, but also has details compressed due to the nonlinearity of the sensor response, resulting in the effective signal which is already weak being completely covered by noise, making it difficult for conventional global processing methods to distinguish noise from real details.

[0042] In addition, the content area of the image under the low-light scene is usually accompanied by a sudden change in brightness caused by non-uniform illumination, such as a local point light source causing an overexposed area in the image, appearing a strong light point, affecting the observation of the texture information of other objects in the surrounding scene, further increasing the difficulty of noise reduction.

[0043] Traditional noise reduction algorithms (such as Gaussian filtering, non-local mean denoising, etc.) are usually based on global smoothing strategy, which is difficult to adapt to the characteristics of uneven noise distribution and low signal-to-noise ratio in low-light scenes. While suppressing noise, such methods often over-smooth weak texture areas, resulting in blurred edges and loss of details, reducing the usability of the video.

[0044] To solve the above problems, the present application realizes the noise reduction processing of ultra-high-definition video in low-light scenes by using a scene region adaptive orthogonal matching pursuit algorithm. According to the differences in texture and brightness characteristics of different image blocks, a scene region adaptive dictionary is constructed by adaptive coding, and then the optimal dictionary atom is used for sparse coding denoising of the image block, improving the picture quality of the scene and objects in the video, and improving the practicality of the video in low-light scenes.

[0045] First, each frame of image in the pre-processed video data is blocked, and a scene region adaptive dictionary is constructed by adaptive coding according to the texture and brightness characteristics of different image blocks.

[0046] Considering that there are regions with complex gray scale and texture characteristics such as weak texture, dark area, overexposed point area, etc. in low-light scenes, directly coding the frame image will result in being affected by a large number of complex coding features in the subsequent search for the optimal sparse representation process, which makes the algorithm not good at sparse representation denoising of the image. The present application analyzes the image features (such as texture, gray value) of the frame image in low-light scenes, and then uses a region growing algorithm to segment different regions in the frame image to obtain image blocks containing unique features in the scene.

[0047] For each frame of image, the gradient value of each pixel point is calculated by the gray value of the pixel point. In this embodiment, the method for calculating the gradient value of the pixel point is Sobel operator, and the prior art for calculating the gradient value of the pixel point will not be described again.

[0048] The brightness histogram of each frame of image is calculated, and the 25% and 75% quantiles of the cumulative distribution function are taken as the gray threshold and to define the dark area, medium brightness area and bright area in low-light scenes. The calculation of the brightness histogram is a prior art, and the specific process will not be described again. Wherein for the gray value of the pixel point , the point belongs to the dark area; , the point belongs to the medium brightness area; , the point belongs to the bright area. Wherein is the 25% quantile, is the 75% quantile.

[0049] After the method based on the global statistics of the image, according to the texture intensity typical features of the global smooth region, weak texture region and high texture region of the image, the gradient threshold value and is obtained by sampling the typical region of the image under the low light scene, which can be set by the implementer according to the implementation scene, and no special limitation is made hereon. Wherein, for the gradient value of the pixel point , the point belongs to the smooth region; , the point belongs to the weak texture region; , the point belongs to the high texture region. In the embodiment, the gradient values of all the pixel points are input, and two gradient threshold values are obtained by the Otsu multi-threshold segmentation method. Wherein is the smallest threshold value obtained by the Otsu multi-threshold segmentation method, is the largest threshold value obtained by the Otsu multi-threshold segmentation method.

[0050] The region growing algorithm is used to merge the adjacent pixel points with similar brightness and texture features into continuous regions, so as to realize the blocking of the image frame. Firstly, the seed points are selected by the random selection method, so as to ensure that these seed points cover different combinations of gray value-texture features, which refers to the intersection of different texture regions and different light-dark regions. For example, all the pixel points corresponding to the intersection of the dark region and the smooth region must have a seed point.

[0051] The pixel points in the eight neighborhoods of the seed point are searched to see whether they meet the growth criterion or not. If they meet the growth criterion, the points are merged into the region of the seed point. This step is repeated until all the pixel points in the image have an attribution, and finally the blocking result of the image frame is output. The definition of the growth criterion of the region growing algorithm is the similarity of the adjacent pixel points and the features of the current seed point region. The function expression of the growth criterion is as follows:

[0052] , represents the gray value of the seed point, represents the gray value of the qth pixel point in the eight neighborhoods of the seed point, represents the gradient value of the seed point, represents the gradient value of the qth pixel point in the eight neighborhoods of the seed point, represents the difference function, and respectively represent the merging threshold values of the gray value and the gradient value, represents the meaning of and, is the function expression. Wherein and are set to 0.1 and 0.15.

[0053] The meaning of the function expression is that if the difference between the gray value of the seed point and the gray value of the neighboring pixel point is less than the merging threshold of the gray value and the difference between the gradient value of the seed point and the gradient value of the neighboring pixel point is less than the merging threshold of the gradient value, the function value is 1, otherwise the function value is 0.

[0054] Based on the above growth criterion, the region growing algorithm is used to segment each frame of image, and the specific process of the region growing algorithm is not described again. Finally, the segmentation result of the region growing algorithm is obtained different image blocks.

[0055] At this point, the segmentation of each frame of image is completed.

[0056] Step S003, based on the difference between the gray value and the gray limit value, the difference between the gradient value and the gradient limit value, the gray feature weight and the texture feature weight are determined; based on the feature weight and the texture feature weight, the original DCT coefficient is improved to construct the exclusive dictionary, and the image is reconstructed by denoising decoding based on the exclusive dictionary.

[0057] The image blocks of different regions obtained by segmenting each frame of image through the above steps are adaptively encoded according to the brightness and texture features of different image blocks, and the scene area exclusive dictionary is constructed, which avoids the problem that the sparse coding representation of different regional image blocks in the dim light scene is inaccurate caused by the dictionary coding mode of the traditional orthogonal matching pursuit algorithm.

[0058] For the image blocks of different regions segmented by the region growing algorithm, according to the threshold division of the pixel gray value and the texture intensity value, 9 kinds of combinations of different types of image blocks can be obtained. For example: weak texture dark area, weak texture bright area.

[0059] For all image blocks corresponding to each combination after segmentation, the discrete cosine transform (DCT) is optimized based on the gray and texture features to construct the regional dictionary, and the gray-texture is used to adjust the coefficients of DCT, so that it can adaptively encode different regional features. First, the original DCT coefficient of each pixel point is obtained based on the discrete cosine transform.

[0060] The original DCT coefficient is adjusted based on the gray feature weight and the texture feature weight, the gray feature weight is determined based on the difference between the gray value of each pixel point and the gray limit value of the combination where it is located, and the texture feature weight is determined based on the difference between the gradient value of each pixel point and the gradient limit value of the combination where it is located.

[0061] The gray scale feature weight is negatively correlated with the difference between the gray scale value of each pixel point and the gray scale limit value of the combination in which the pixel point is located, and the texture feature weight is positively correlated with the difference between the gradient value of each pixel point and the gradient limit value of the combination in which the pixel point is located.

[0062] It should be noted that the positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction, and when one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large; the specific relationship is determined by actual application, and the present application does not make special limitation.

[0063] It should be noted that the negative correlation means that when one variable increases, the other variable decreases, and the two variables change in opposite directions, and when one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small; the specific relationship is determined by actual application, and the present application does not make special limitation

[0064] Preferably, in the present embodiment, the expressions of the gray scale feature weight and the texture feature weight are as follows:

[0065] denotes the gray scale value of the pixel point corresponding to the coordinate in the lth combination, denotes the gray scale limit value of the lth combination, denotes the gradient value of the pixel point corresponding to the coordinate in the lth combination, denotes the gradient limit value of the lth combination, denotes the exponential function with a natural constant as the base, denotes the gray scale feature weight of the pixel point corresponding to the coordinate in the lth combination, denotes the texture feature weight of the pixel point corresponding to the coordinate in the lth combination, denotes the smoothing coefficient of the gray scale feature weight, denotes the smoothing coefficient of the texture feature weight. The smoothing coefficient of the gray scale feature weight and the smoothing coefficient of the texture feature weight are 0.1.

[0066] When the combination corresponds to a dark area, the gray scale limit value is ; when the combination corresponds to a medium brightness area, the gray scale limit value is and ; and when the combination corresponds to a bright area, the gray scale limit value is . When the combination corresponds to a smooth texture area, the gradient limit value is ; when the combination corresponds to a weak texture area, the gradient limit value is and ​​the mean value of the gradient of the corresponding region; when the combination corresponds to a high-texture region, the gradient limit value is .

[0067] The original DCT coefficient is modified based on the gray feature weight and the texture feature weight of the pixel point to obtain an improved DCT coefficient.

[0068] The improved DCT coefficient is in positive correlation with the gray feature weight and the texture feature weight.

[0069] Preferably, in the embodiment, the improved DCT coefficient is:

[0070] , denotes the coordinate of the corresponding pixel point denotes the coordinate of the corresponding pixel point denotes the coordinate of the corresponding pixel point denotes the coordinate of the corresponding pixel point

[0071] The DCT base function is optimized according to the gray and texture features specific to different image blocks, and the gray feature weight and the texture feature weight are used to match the features in the low-light scene to construct a region-specific dictionary, the optimal dictionary atom is searched in the region with the same features to update the sparse coding of the region, and adaptive noise reduction of the region is completed, which can avoid the problem of loss of detail information caused by the influence of other highlight and strong texture regions on the noise reduction of dark regions and weak texture regions in the traditional global noise reduction algorithm.

[0072] For each combination, a set of DCT coefficients of all pixel points is taken as a region-specific dictionary of the combination, and each value in the specific dictionary is taken as an atom.

[0073] Based on the orthogonal matching pursuit algorithm, the image blocks of each combination are denoised, specifically, the image blocks of each combination after coding are initialized to obtain an initial residual term , a support set is initialized, and then the inner product of the residual term and the atom in the specific dictionary is calculated to find the optimal dictionary atom to add to the support set , which is used to update the residual term obtained by initializing the image block , when the iteration period times is reached or the residual term of the first iteration is reached.L2 norm of the residual is less than the noise variance, the denoising of Sparse coding denoising of image blocks of this type. The orthogonal matching pursuit algorithm is the prior art, and its specific process will not be repeated here.

[0074] The denoising of all combined image blocks is completed through the above steps.

[0075] After all the sparse coding denoised image blocks are decoded, the image frame is reconstructed by the method of block aggregation, the denoising of each image frame is realized, and then the operation is repeated to complete the denoising of the image content of the entire video.

[0076] Step S004, the reconstructed each frame image is synthesized into a complete video.

[0077] After using the scene region adaptive based orthogonal matching pursuit algorithm to complete the frame-by-frame denoising of the video, the processed image frame needs to be post-processed and re-synthesized into a complete video.

[0078] Firstly, inter-frame consistency optimization is performed, time domain filtering technology is adopted to eliminate the inter-frame flicker or jump caused by denoising, and ensure the smoothness of video playback, secondly, sharpening filter is used to restore the texture information that may be lost in the denoising process, finally, the processed single frame image is re-encoded into a video stream according to the original time sequence, and the auxiliary information such as audio, timestamp, metadata and the like is synchronously fused, and the audio-visual synchronization is ensured.

[0079] The finally generated video reduces noise while retaining dark area details and weak texture features, meeting the high-definition application requirements in low-light scenes, and realizing the denoising of ultra-high-definition video in low-light scenes.

[0080] It should be noted that: the above-described embodiments are only 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 essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

[0081] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

Claims

1. A method for ultra-high-definition video noise reduction, characterized in that, The method comprises the following steps: Obtaining pre-processed video data and noise variance; Obtaining each image of the video data, calculating the gradient value of each pixel point based on the gray value of the image, obtaining the brightness histogram of the image based on the gray value, obtaining the gray threshold, obtaining the gradient threshold based on the threshold segmentation method, dividing the pixel points into different regions based on the gray threshold and the gradient threshold, obtaining different combinations based on the different regions of the pixel points, determining the growth criterion based on the difference between the gray value of the pixel point and the gray value in its neighborhood, the difference between the gradient value of the pixel point and the gradient value in its neighborhood, and dividing each image into blocks through the region growing of the seed point in each combination based on the growth criterion; Obtaining the original DCT coefficient of each pixel point based on the discrete cosine transform, determining the gray feature weight based on the difference between the gray value of each pixel point and the gray limit value of the combination it belongs to, determining the texture feature weight based on the difference between the gradient value of each pixel point and the gradient limit value of the combination it belongs to, modifying the original DCT coefficient based on the gray feature weight and the texture feature weight of the pixel point to obtain the improved DCT coefficient, constructing a dedicated dictionary for the improved DCT coefficient of each combination pixel point, and using the orthogonal matching pursuit algorithm based on the dedicated dictionary and the noise variance to denoise the image block of each combination; Synthesizing each reconstructed image into a complete video.

2. The ultra-high-definition video noise reduction method of claim 1, wherein, The gray threshold is the 25% and 75% quantile of the brightness histogram, and the gradient threshold is obtained by using the Otsu multi-threshold segmentation method based on the gradient value of all pixel points.

3. The ultra-high-definition video noise reduction method of claim 2, wherein, The method of dividing the pixel points into different regions based on the gray threshold and the gradient threshold is: If the pixel point belongs to a dark region; the pixel point belongs to a medium brightness region; the pixel point belongs to a bright region; wherein is a gray value of the pixel point p, is a 25% quantile, is a 75% quantile; If the pixel point belongs to a smooth region; the pixel point belongs to a weak texture region; the pixel point belongs to a high texture region; wherein is the minimum threshold value obtained by the Otsu multi-threshold segmentation method, is the maximum threshold value obtained by the Otsu multi-threshold segmentation method.

4. The ultra-high-definition video noise reduction method of claim 3, wherein, The combination refers to the intersection of pixel points in different texture regions and pixel points in different light and dark regions; the texture region includes a smooth region, a weak texture region and a high texture region; the light and dark region includes a dark region, a medium brightness region and a bright region.

5. The method of claim 1, wherein the method further comprises: The function expression of the growth criterion is: , denotes the gray value of the seed point, denotes the gray value of the qth pixel point in the eight-neighborhood of the seed point, denotes the gradient value of the seed point, denotes the gradient value of the qth pixel point in the eight-neighborhood of the seed point, denotes the difference function, and denote preset threshold values for the gray value and the gradient value, respectively, denotes the meaning of and, is a functional expression.

6. The ultra-high-definition video noise reduction method of claim 3, wherein, The gray feature weight is negatively correlated with the difference between the gray value of each pixel point and the gray limit value of the combination it belongs to.

7. The ultra-high-definition video noise reduction method as described in claim 3, characterized in that, The texture feature weight is positively correlated with the difference between the gradient value of each pixel point and the gradient limit value of the combination it belongs to.

8. The ultra-high-definition video noise reduction method of claim 6, wherein, The method of obtaining the gray limit value is: When the combination corresponds to a dark region, the gray level limit value is ; when the combination corresponds to a medium luminance region, the gray level limit value is and ; when the combination corresponds to a bright region, the gray level limit value is .

9. The ultra-high-definition video noise reduction method of claim 7, wherein, The method of obtaining the gradient limit value is: When combining corresponding smooth texture regions, the gradient limit value is ; when combining corresponding weak texture regions, the gradient limit value is and ; and when combining corresponding high texture regions, the gradient limit value is .

10. The method of claim 1, wherein, The improved DCT coefficient is positively correlated with the gray feature weight and the texture feature weight.

Images