[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN114125471A - Video coding pre-filtering method - Google Patents

Video coding pre-filtering method Download PDF

Info

Publication number
CN114125471A
CN114125471A CN202111426700.3A CN202111426700A CN114125471A CN 114125471 A CN114125471 A CN 114125471A CN 202111426700 A CN202111426700 A CN 202111426700A CN 114125471 A CN114125471 A CN 114125471A
Authority
CN
China
Prior art keywords
pixel
window
filtering
regularization parameter
gradient
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111426700.3A
Other languages
Chinese (zh)
Inventor
贾路恒
黄熙
王涵
任浩强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Technology
Original Assignee
Beijing University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Technology filed Critical Beijing University of Technology
Priority to CN202111426700.3A priority Critical patent/CN114125471A/en
Publication of CN114125471A publication Critical patent/CN114125471A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/85—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/80—Details of filtering operations specially adapted for video compression, e.g. for pixel interpolation

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

The invention discloses a video coding pre-filtering method, which is based on a guide filtering algorithm and comprises the following steps: for a video time domain, calculating the structural similarity of filtering blocks in the same region in a multi-frame video; for the picture content of a video space domain, the gradient size of a pixel region is calculated, a final regularization parameter is calculated according to the similarity size and the threshold range of the gradient size, and a guide filtering function is constructed for filtering the obtained regularization parameter value.

Description

Video coding pre-filtering method
Technical Field
The invention relates to the field of video compression coding and image processing, is mainly applied to a video preprocessing process in a video coding system, and particularly relates to a self-adaptive guided filtering algorithm.
Background
In recent years, the field of streaming media is rapidly developed, video media such as short videos and high-definition videos become a main entertainment mode in daily life of people, the increasing video demand of people also brings heavy pressure to storage of a streaming media server, noise addition in videos is difficult to avoid due to the existence of unstable factors such as video shooting environments and video shooting equipment, high-frequency information (mainly noise) which is difficult to be perceived by human eyes in videos has great influence on the coding efficiency of a coder in the video coding process, so that the problems of coding video volume increase, unsatisfactory coding effect and the like are caused. Therefore, a key technology of video coding is to remove some high-frequency redundant information of the video, thereby achieving the purpose of improving the video coding efficiency.
When high-frequency redundant information of a video is filtered, a traditional filtering method such as gaussian filtering, median filtering and other filtering algorithms often causes great damage to the picture content during filtering, and causes severe degradation of video PSNR (peak signal to noise ratio) quality index, thereby causing rapid degradation of video coding efficiency, and BM3D algorithm and anisotropic filtering algorithm with good filtering effect are often not suitable for a video coding system because of too high operation complexity.
A filter algorithm guiding filter algorithm which gives consideration to operation complexity and filter effect is commonly used in the field of image processing and obtains excellent effect, and a function expression of the filter algorithm guiding filter algorithm can be expressed as follows:
Figure BDA0003378904450000011
wherein:
Figure BDA0003378904450000012
Figure BDA0003378904450000013
Iifor input pixel gray value, qiTo output a pixel gray value, akAnd betakTo guide the filter coefficients, omegakIs IiA sliding window therein
Figure BDA0003378904450000014
Is the variance of the window in which the pixel is located, is the regularization parameter,
Figure BDA0003378904450000015
is the average of the window in which the pixel is located.
Figure BDA0003378904450000016
Figure BDA0003378904450000017
Phi is window omegakAfter the guided filter parameters are obtained, for the pixel IiAre also included in other pixel windows, so that a pixel can be regarded as a plurality of coefficients akAnd betakThe regularization parameter epsilon of all pixel points in the window is averaged to obtain the average regularization parameter value of the pixel currently processed, and the average regularization parameter value is substituted into formulas (2) and (3) to obtain
Figure BDA0003378904450000018
Figure BDA0003378904450000019
The above-mentioned calculation formula of the guided filtering function has a good effect on a picture or a static image, and when the picture is filtered, the regularization coefficient e is a set fixed value, but for video filtering, the fixed regularization parameter cannot perform sufficiently stable filtering processing on the picture of each frame of the video, such as motion smear and edge blur caused by unstable filtering effect.
Disclosure of Invention
The invention provides a self-adaptive filtering method based on a guide filter, which can analyze pixels around a pixel point to be processed according to an input video, and can calculate an optimal guide filtering function by combining the complexity of video content and the change condition between video frames, thereby achieving the purposes of self-adaptive filtering and improving the video coding efficiency.
The technical scheme adopted by the invention is a self-adaptive filtering method based on a guide filter, which comprises the following specific implementation steps:
for the first frame of the video sequence, because of the lack of reference, the forward frame is filtered by using a method of fixing the regularization parameter, and the following steps are described by taking a filtering processing method of the pixel i in the second frame as an example
(1) Calculating gradient value V of pixel i and sliding window W2 where pixel i is locatediAnd the forward reference frame window W1iStructural similarity of (2).
(2) And judging the calculation mode of the regularization parameter according to the gradient V and the structural similarity S, and calculating to obtain the regularization parameter.
(3) Sequentially obtaining a sliding window omega according to a guiding filter function formulaeAnd (3) repeating the operations of the steps 1) and 2) on the regularization parameters of other pixels in the image to obtain average regularization parameters and construct a calculation formula of the guide filter coefficient according to the average regularization parameters.
(4) Obtaining a pixel gray value q after filtering according to a guide filtering functioni。
(5) And performing the steps on each pixel in the video frame to obtain the self-adaptive filtering video.
In the step (1), when the gradient value V and the structural similarity S are calculated, a window WN with a pixel i as the center is seti(N is a frame number) and for structural similarity, its forward reference frame is set to (W (N-1))iFurthermore, for structural similarity S, reference may be made to the subsequent video frame window (W (N + k))iHowever, from video coding computational complexity considerations, the total reference frame does not exceed 3 frames and comprises a forward reference frame window (W (N-1))i。
In the step (2), the calculation mode of the regularization parameter is judged, and for the structural similarity S, when the structural similarity between the window and the adjacent frame window is greatly changed, that is, the value of S is smaller than the threshold SminWhen (S)min0.8), then a fixed regularization parameter (e 1-4) is set, andand jumping out of the current step, and calculating the regularization parameter value of the next pixel.
Judging the calculation mode of the regularization parameter in the step (2), and considering the current pixel neighborhood window W2 for the gradient value ViAnd if the gradient V is too large, the region is considered to be an edge region needing to be reserved, if the gradient V is too small, the region is considered to be a smooth region, if one of the gradient V and the edge region is established, a fixed regularization parameter is set, the current step is skipped, and the regularization parameter value of the next pixel is calculated.
Judging the calculation mode of the regularization parameter in the step (2), and when the gradient magnitude V and the structural similarity S are both in the set threshold range, adopting a calculation method of a regularization function, wherein the formula of the regularization function is as follows:
Figure BDA0003378904450000031
Figure BDA0003378904450000032
for the gradient and structure similarity versus normal coefficient of the normalization parameter, window W2 is in the neighborhood of pixel iiThe gradient in this range can be described by the gradient magnitude V in order to better retain the edge information; the change amplitude of the inter-frame picture information can be calculated through the structural similarity S of the front and back frame videos, more details are reserved for the video frame area with larger change, the filtering strength can be slightly increased for the similar pictures, and V and S do not contribute to the video coding efficiency differently from the viewpoint of video coding, so that different weights need to be distributed to the V and S.
In the step (3), the pair omegakWhen all the pixels are subjected to regularization parameter calculation, omegakThe sliding window set for guiding the filtering function, whose template is set to 3x3 in the present invention, needs to find ω when filtering pixel ikThe regularization parameters of all pixels in the image are substituted into the formulas (2) and (3) to obtain a guide filtering average coefficient
Figure BDA0003378904450000033
Constructing an adaptive guiding filtering formula, and obtaining the filtering value of the current pixel calculated in the step 4).
Compared with the prior art, the invention has the following advantages:
1. the area of the pixel is judged according to two dimensions of a picture time domain and a picture space domain, so that different areas of the video can be subjected to more accurate filtering operation, and due to the richness of picture contents and the uncertainty of a picture motion main body, the picture is subjected to regional filtering to obtain a higher-quality de-noised video.
2. Because the influence weight on video coding is considered when the regularization parameter is calculated, the denoising can be completed without influencing the subjective quality, and the video coding efficiency is obviously improved.
3. The algorithm can be used together with a box filter, and the filtering operation speed can be greatly improved on the premise of not influencing the filtering effect.
Drawings
Fig. 1 is a schematic diagram of a basic flow of the adaptive coding pre-filtering method according to the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and examples.
The method implements a reference-directed filter function, the formula is as follows:
Figure BDA0003378904450000034
wherein:
Figure BDA0003378904450000035
Figure BDA0003378904450000036
Iifor input pixel gray value, qiIs output pixel grayValue of akAnd betakTo guide the filter coefficients, omegakIs IiA sliding window therein
Figure BDA0003378904450000041
Is the variance of the sliding window in which the pixel is located, is the regularization parameter,
Figure BDA0003378904450000042
is the average of the sliding window in which the pixel is located.
Figure BDA0003378904450000043
Figure BDA0003378904450000044
Phi is window omegakAfter the guided filter parameters are obtained, for the pixel IiAre also included in other pixel windows, so that a pixel can be regarded as a plurality of coefficients akAnd betakThe regularization parameter values epsilon corresponding to all the pixel points in the window are averaged to obtain the average regularization parameter of the pixel currently processed, and the average regularization parameter is substituted into formulas (2) and (3) to obtain
Figure BDA0003378904450000045
Calculating a guided filter coefficient
Figure BDA0003378904450000046
The method comprises the steps of obtaining regularization parameter values through two dimensionalities of a video space domain and a video time domain in a self-adaptive mode respectively so as to adjust the intensity level of denoising filtering, wherein the key parameter belongs to the regularization parameter epsilon, and the size of the filtering degree is determined, and the flow of the steps is shown in figure 1.
1. When calculating the structural similarity S, a window W with a pixel i as the center is setiThe structural similarity is calculated by adopting an SSIM structural similarity algorithm, such asIf the structural similarity of the current window is smaller than the set maximum threshold (0.8) of S, skipping the calculation of the gradient V of the current pixel window, and directly obtaining the fixed regularization parameter value of the pixel, wherein the effect is optimal when the fixed value of the regularization parameter is set within the range of 1-4. When the size of the structural similarity S satisfies the threshold range (S)>0.8), proceed with the pixel window WiThe gradient value of (2) is calculated.
2. When calculating the window gradient of the pixel region, the Sobel operator is used for calculating the window W of the pixel regioniThe gradient magnitude in the x and y directions is respectively calculated, the gradient magnitude V of the current pixel i is obtained according to the gradient directionality, and when the picture is a relatively smooth area (V)<10) Adopting larger regularization parameters to enhance the filtering effect, and when the calculated gradient value is larger and represents that the edge is more obvious (V)>40) Then a smaller regularization parameter is set, reducing the degree of filtering. The relationship for the gradient magnitude V and the regularization parameter can be represented by equation (6):
∈(V)＝100V-2 (6)
thus, the fixed regularization parameter value of the pixel can be obtained directly by the formula for the threshold values in both cases of V <10 and V > 40.
3. When the gradient magnitude V and the structural similarity S are both within a given threshold range (10< V <40, S >0.8), the regularization parameter needs to be solved by a regularization function e (V, S), and the function expression is as follows:
Figure BDA0003378904450000047
wherein
Figure BDA0003378904450000048
ω is the weight coefficient of V and S, respectively, which represents the weight of the gradient component and the similarity component to the encoder,
Figure BDA0003378904450000049
ω is defined as 0.4 and 0.6, resulting in a regularization parameter.
4. The preceding step being a single imageCalculation procedure of regularization parameter, ωkThe sliding window set for the guided filter function is set as a fixed template of 3x3 in the present invention, and the video coding effect is best under the template. In finding ωkAfter the regularization coefficients of all the pixel values are obtained, the average value is calculated to obtain the average guide filter coefficient
Figure BDA0003378904450000051
And taking the value as the coefficient value of the guide filter function to construct the guide filter function of the current processed pixel.

Claims (7)

1. A method for pre-filtering in video coding, the method comprising the steps of: for the first frame of the video sequence, filtering it by using a method of fixing the regularization parameter, the filtering processing method of the pixel i in the second frame is described as follows,
step 1) calculating a sliding window W2 where a pixel i is locatediThe gradient value V of, the sliding window W2 where the pixel i is locatediAnd the forward reference frame window W1iStructural similarity of (a);
step 2) judging the calculation mode of the regularization parameter in the guide filter according to the gradient V and the structural similarity S, and calculating to obtain the regularization parameter
Step 3) sequentially obtaining a sliding window omega according to a guiding filter function formulakThe regularization parameters of other pixels are repeated, and the operations of the steps 1) and 2) are repeated to obtain a window omegakThe average regularization parameter and a guiding filter coefficient calculation formula is constructed according to the average regularization parameter;
step 4) obtaining the pixel gray value q after filtering according to the guiding filtering functioni；
And 5) executing the steps 1) to 4) on each pixel in the video frame to obtain the self-adaptive filtering video.
2. The method according to claim 1, wherein the step (1) of calculating the gradient magnitude V and the structural similarity S of the pixel i sets a window WN centered on the pixel iiAnd N is a frame number.
3. The method according to claim 1, wherein the formula of the guided filtering algorithm used in step 2) is defined as:
Figure FDA0003378904440000011
Iifor input pixel gray value, qiTo output a pixel gray value, akAnd betakTo guide the filter coefficients, omegakIs IiThe sliding window is located; guided filter coefficient akAnd betakThe definition is as follows:
Figure FDA0003378904440000012
Figure FDA0003378904440000013
wherein
Figure FDA0003378904440000014
Is the variance of the window in which the pixel is located, is the regularization parameter,
Figure FDA0003378904440000015
is the average of the window in which the pixel is located.
4. The method of claim 1, wherein in step 2), when window W2 is selected, the method further comprisesiAnd the forward reference frame window W1iWhen the structural similarity between the S and the S is greatly changed, namely the numerical value of the S is smaller than the threshold value SminSetting a fixed regularization parameter for the pixel i, skipping the current step, and calculating the regularization parameter value of the next pixel; to expandDomain information, for structural similarity S reference backward video frame window (W (2+ n))iThe filtering processing is facilitated; from video coding computational complexity considerations, the reference frame does not exceed three frames and must contain a forward reference frame.
5. The method according to claim 1, wherein the regularization parameter in step 2) is calculated by determining a window W2 in which the current pixel i is located according to a gradient ViIf the gradient v is too large, the edge area is reserved, if the gradient is too small, the edge area is a smooth area, if one of the two is true, a fixed regularization parameter is set, the current step is skipped, and the regularization parameter value of the next pixel is calculated.
6. The method according to claim 1, wherein the regularization parameter is calculated in step (2), and when the gradient magnitude V and the structural similarity S are both within a set threshold range, a regularization function is used for calculation, and the regularization function formula is as follows:
Figure FDA0003378904440000021
Figure FDA0003378904440000022
for the coefficients of the regularization parameter, the gradient and the structural similarity do not contribute to the video coding efficiency uniformly from the viewpoint of video coding, and therefore they are assigned non-uniform weights.
7. The method of claim 1, wherein the calculation formula according to which the step 3) is based is as follows:
Figure FDA0003378904440000023
Figure FDA0003378904440000024
ωkis IiThe sliding window is located, phi is window omegakAfter the guided filter parameters are obtained, other pixel windows also contain the pixel IiThe filtering of one pixel is regarded as a plurality of coefficients akAnd betakThe regularization parameters epsilon corresponding to all the pixel point windows in the window are averaged, and the average value is substituted into the formula (2) and the formula (3) to obtain the parameters
Figure FDA0003378904440000025
Finally, the guiding filter function formula (6) is substituted to obtain the output filter pixel value.
CN202111426700.3A 2021-11-27 2021-11-27 Video coding pre-filtering method Pending CN114125471A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111426700.3A CN114125471A (en) 2021-11-27 2021-11-27 Video coding pre-filtering method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111426700.3A CN114125471A (en) 2021-11-27 2021-11-27 Video coding pre-filtering method

Publications (1)

Publication Number Publication Date
CN114125471A true CN114125471A (en) 2022-03-01

Family

ID=80370729

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111426700.3A Pending CN114125471A (en) 2021-11-27 2021-11-27 Video coding pre-filtering method

Country Status (1)

Country Link
CN (1) CN114125471A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116824526A (en) * 2023-08-29 2023-09-29 广州视安智能科技有限公司 Digital intelligent road monitoring system based on image processing

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8731282B1 (en) * 2013-02-28 2014-05-20 Sonic Ip, Inc. Systems and methods for region of interest background smoothing and compression enhancement
CN109375125A (en) * 2018-10-25 2019-02-22 哈尔滨理工大学 A kind of compressed sensing magnetic resonance imaging method for reconstructing for correcting regularization parameter
CN111314711A (en) * 2020-03-31 2020-06-19 电子科技大学 Loop filtering method based on self-adaptive self-guided filtering
CN111626962A (en) * 2020-05-27 2020-09-04 重庆邮电大学 CMOS endoscope image enhancement method
CN111652821A (en) * 2020-06-05 2020-09-11 北京环境特性研究所 Low-light-level video image noise reduction processing method, device and equipment based on gradient information
CN111667433A (en) * 2020-06-09 2020-09-15 中国电子科技集团公司第五十四研究所 Unmanned aerial vehicle image defogging method based on simple linear iterative clustering optimization
CN112053298A (en) * 2020-08-27 2020-12-08 上海电机学院 Image defogging method
CN113344973A (en) * 2021-06-09 2021-09-03 南京信息工程大学 Target tracking method based on space-time regularization and feature reliability evaluation
CN113610734A (en) * 2021-08-23 2021-11-05 眸芯科技（上海）有限公司 Image edge processing method based on oriented filtering and application

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8731282B1 (en) * 2013-02-28 2014-05-20 Sonic Ip, Inc. Systems and methods for region of interest background smoothing and compression enhancement
CN109375125A (en) * 2018-10-25 2019-02-22 哈尔滨理工大学 A kind of compressed sensing magnetic resonance imaging method for reconstructing for correcting regularization parameter
CN111314711A (en) * 2020-03-31 2020-06-19 电子科技大学 Loop filtering method based on self-adaptive self-guided filtering
CN111626962A (en) * 2020-05-27 2020-09-04 重庆邮电大学 CMOS endoscope image enhancement method
CN111652821A (en) * 2020-06-05 2020-09-11 北京环境特性研究所 Low-light-level video image noise reduction processing method, device and equipment based on gradient information
CN111667433A (en) * 2020-06-09 2020-09-15 中国电子科技集团公司第五十四研究所 Unmanned aerial vehicle image defogging method based on simple linear iterative clustering optimization
CN112053298A (en) * 2020-08-27 2020-12-08 上海电机学院 Image defogging method
CN113344973A (en) * 2021-06-09 2021-09-03 南京信息工程大学 Target tracking method based on space-time regularization and feature reliability evaluation
CN113610734A (en) * 2021-08-23 2021-11-05 眸芯科技（上海）有限公司 Image edge processing method based on oriented filtering and application

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
何涛、谢有庆、李一航、徐昊: "一种改进的有雾图像去雾算法", 《科学技术与工程》, vol. 20, no. 2, pages 2 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116824526A (en) * 2023-08-29 2023-09-29 广州视安智能科技有限公司 Digital intelligent road monitoring system based on image processing
CN116824526B (en) * 2023-08-29 2023-10-27 广州视安智能科技有限公司 Digital intelligent road monitoring system based on image processing

Similar Documents

Publication Publication Date Title
CN108921800B (en) Non-local mean denoising method based on shape self-adaptive search window
US7076113B2 (en) Apparatus and method for adaptive spatial segmentation-based noise reducing for encoded image signal
Kim et al. A pseudo-blind convolutional neural network for the reduction of compression artifacts
US20040227851A1 (en) Frame interpolating method and apparatus thereof at frame rate conversion
US20080152247A1 (en) Methods and Systems for Image Noise Processing
US20080279279A1 (en) Content adaptive motion compensated temporal filter for video pre-processing
CN113034401B (en) Video denoising method and device, storage medium and electronic equipment
CN106210448B (en) Video image jitter elimination processing method
US20070140587A1 (en) Auto-regressive method and filter for denoising images and videos
WO2021012965A1 (en) Image processing method and apparatus, mobile terminal video processing method and apparatus, device and medium
CN110796615A (en) Image denoising method and device and storage medium
CN110610467A (en) Multi-frame video compression noise removing method based on deep learning
JP2012249038A (en) Image signal processing apparatus and image signal processing method
CN104992419A (en) Super pixel Gaussian filtering pre-processing method based on JND factor
CN110378860A (en) Method, apparatus, computer equipment and the storage medium of restored video
CN112381744A (en) Adaptive preprocessing method for AV1 synthetic film grains
CN114125471A (en) Video coding pre-filtering method
Agarwal et al. Compressing video calls using synthetic talking heads
CN114119383A (en) Underwater image restoration method based on multi-feature fusion
US20110097010A1 (en) Method and system for reducing noise in images in video coding
CN115829873A (en) Image restoration method and processing system
Munezawa et al. Noise removal method for moving images using 3-D and time-domain total variation regularization decomposition
De-yong et al. Enhancement algorithms for images in coal mines base on dark light denoising networks
Kim et al. Image coding based on selective super-resolution network
Wada et al. Extended joint bilateral filter for the reduction of color bleeding in compressed image and video

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20220301

RJ01 Rejection of invention patent application after publication