CN105306957B - Adaptive loop filter method and apparatus - Google Patents
Adaptive loop filter method and apparatus Download PDFInfo
- Publication number
- CN105306957B CN105306957B CN201510698486.5A CN201510698486A CN105306957B CN 105306957 B CN105306957 B CN 105306957B CN 201510698486 A CN201510698486 A CN 201510698486A CN 105306957 B CN105306957 B CN 105306957B
- Authority
- CN
- China
- Prior art keywords
- filter shape
- filter
- shape
- sizes
- pixel
- 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.)
- Active
Links
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
Embodiment of the present invention discloses the adaptive loop filter method and apparatus for video image encoding and decoding, and the method and apparatus for handling using adaptive loop filter the video image of coding.For adaptive loop filter, the final filter shape that building cross line style filter shape is superimposed at center with square filter shape, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than sample window number in the vertical direction.The adaptive loop filter of embodiment according to the present invention, while improving picture quality, hardware consumption is minimum.
Description
Technical field
The present invention relates to the technical field of the video image encoding and decoding in MultiMedia Field more particularly to a kind of video images
Adaptive loop filter method and apparatus used in encoding and decoding and the video that coding is handled using adaptive loop filter
The method and apparatus of image.
Background technique
Modern video encoding and decoding technique is all converted using the block of block-based prediction and quantization.Generally current encoded frame is drawn
It is divided into the coding unit not being overlapped mutually, this coding unit can be AVC/H.264 (Motion Picture Experts Group's advanced video coding
MPEG-4AVC) or the macro block (MB, macro block) in AVS1 (internal video coding standard the 1st edition), it is also possible to HEVC/
H.265(High Efficiency Video Coding;High efficiency coding and decoding video) or AVS2 (internal video coding standard
2 editions) in coding unit (CU, coding unit).Coding unit further can also be divided into coded sub-units as son.Benefit
With block-based prediction, the block of pixels of reconstruction can be used to predict other block of pixels.The block transformation of block-based prediction and quantization
Use can lead to the losses of the original image informations such as noise inside the blocking artifact of block border and block.
In HEVC/H.265 and AVS2 standard, adaptive loop filter (ALF) is used further to eliminate in block
The noise in portion improves the quality of reconstruction image, improves forecasting efficiency.ALF technology passes through encoded filter shape and filter factor,
Reconstruction image is filtered at decoder, reduces the compression artefacts in decoding image, and provide height for subsequent encoding and decoding image
The prediction reference image of quality, to further increase compression efficiency.The building of filter factor is so that reconstruction image and original graph
Mean square deviation (MSE) as between reaches minimum.But ALF be scramble time longest in these coding/decoding systems, calculation amount most
Greatly, the highest part of complexity.For this purpose, existing ALF algorithm be usually reconstructed image quality and ALF algorithm complexity,
Calculation amount, compression efficiency etc. carry out balanced.ALF in AVS2 standard has been carried out relative to the ALF in HEVC/H.265
Certain simplification.
In addition, video encoder and decoder are usually by hard-wired, still, existing various ALF algorithms are not
It is designed and develops towards hard-wired direction is conducive to.
Summary of the invention
The purpose of the present invention is intended to provide a kind of improved adaptive loop filter method, is improving the same of picture quality
When, in hardware realization, cost is smaller.
According to the first aspect of the invention, a kind of adaptive loop filter method for video image encoding and decoding is provided,
Super clear image suitable for 4k.This method comprises: reconstruction image is divided into multiple blocks;And in the multiple block
Each block, be filtered according to certain filter shape and filter factor, the filter shape is by multiple sample window structures
At each sample window is used to select the pixel of corresponding position, and is provided with a filter factor, and the filter factor is set
It counts into and makes mean square deviation between the reconstruction image and original image minimum, for each block in multiple blocks, according to
It includes: current pixel in current block that certain filter shape and filter factor, which is filtered, is selected according to filter shape
It selects current pixel and is filtered used surrounding pixel, surrounding pixel and corresponding filter factor the sum of are multiplied, and it is new to obtain
Pixel replace current pixel.For a block in the multiple block, cross line style filter shape and pros are constructed
The final filter shape that shape filter shape is superimposed at center, final filter shape are more to choose water relative to vertical direction
The unsymmetric structure of square upward pixel.The sample window number of the cross line style filter shape in the horizontal direction is greater than
Sample window number in the vertical direction, the filter factor of the sample window corresponding to final filter shape are about final filter
Waveform shape central symmetry or quadrant symmetrical, wherein the cross line style filter shape for constituting the final filter shape is 9 ×
7 sizes, and the square filter shape for constituting the final filter shape is 3 × 3 sizes;Alternatively, constituting the final filter
The cross line style filter shape of waveform shape is 7 × 5 sizes, and constitutes the square filter shape point of the final filter shape
It Wei not 3 × 3 sizes.
According to the second aspect of the invention, a kind of adaptive loop filter equipment for video image encoding and decoding is provided,
Super clear image suitable for 4k.The equipment includes: division module, for reconstruction image to be divided into multiple blocks;And filtering
Module, for being filtered according to certain filter shape and filter factor, institute for each block in the multiple block
It states filter shape to be made of multiple sample windows, each sample window is used to select the pixel of corresponding position, and is provided with one
A filter factor, the filter factor are designed to make the mean square deviation between the reconstruction image and original image minimum;Filter
Wave module is specifically used for selecting current pixel to be filtered for the current pixel in current block according to filter shape and being used
Surrounding pixel, surrounding pixel and corresponding filter factor the sum of are multiplied obtained new pixel to replace current pixel.It is right
A block in the multiple block, building cross line style filter shape are superimposed at center with square filter shape
Final filter shape, final filter shape are that the unsymmetrical knot of the pixel in horizontal direction is more chosen relative to vertical direction
Structure.The sample window number of the cross line style filter shape in the horizontal direction is greater than sample window number in the vertical direction
Mesh, the filter factor of the sample window corresponding to final filter shape are about final filter shape central symmetry or quadrant pair
Claim, wherein the cross line style filter shape for constituting final filter shape is 9 × 7 sizes, and constitutes final filter shape
Square filter shape is 3 × 3 sizes;Or constituting the cross line style filter shape of final filter shape is 7 × 5 sizes or 11
× 9 sizes, and the square filter shape for constituting final filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
According to the third aspect of the invention we, a kind of video image being handled coding using adaptive loop filter is provided
Method, the super clear image suitable for 4k.This method comprises: receiving the reconstruction image number for corresponding to the video image data of coding
According to;By adaptive loop filter be applied to the reconstruction image data to generate the video image data through filtering, adaptively
Loop filtering includes selecting the current picture according to filter shape for the current pixel in current block in reconstructed image data
Element is filtered used surrounding pixel, and obtained new pixel that surrounding pixel and corresponding filter factor the sum of are multiplied is come
Instead of current pixel;And provide the video data through filtering.The final filter shape that the adaptive loop filter uses is logical
It crosses cross line style filter shape to be superimposed at center with square filter shape to construct, final filter shape is relative to vertical
More choose the unsymmetric structure of the pixel in horizontal direction in direction.The cross line style filter shape is in the horizontal direction
Sample window number is greater than sample window number in the vertical direction, the filtering of the sample window corresponding to final filter shape
Coefficient is about final filter shape central symmetry or quadrant symmetrical, wherein constitutes the cross line style of final filter shape
Filter shape is 9 × 7 sizes, and the square filter shape for constituting final filter shape is 3 × 3 sizes;Or it constitutes most
The cross line style filter shape of whole filter shape is 7 × 5 sizes or 11 × 9 sizes, and constitutes the pros of final filter shape
Shape filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
According to the fourth aspect of the invention, a kind of video image being handled coding using adaptive loop filter is provided
Equipment, the super clear image suitable for 4k.The equipment includes: receiving module, for receiving the video image data for corresponding to coding
Reconstructed image data;Filter module, for adaptive loop filter to be applied to the image data of the reconstruction to generate warp
The video image data of filtering, specifically for the current pixel being directed in reconstructed image data in current block, according to the filter
Waveform shape selection current pixel is filtered used surrounding pixel, and surrounding pixel is multiplied with the corresponding filter factor
The sum of obtained new pixel replace the current pixel;And module is provided, for providing the video data through filtering.Institute
It states the final filter shape that adaptive loop filter uses and passes through cross line style filter shape and square filter shape at center
For place's superposition to construct, final filter shape is that the unsymmetrical knot of the pixel in horizontal direction is more chosen relative to vertical direction
Structure.The sample window number of the cross line style filter shape in the horizontal direction is greater than sample window number in the vertical direction
Mesh, the filter factor of the sample window corresponding to final filter shape are about final filter shape central symmetry or quadrant pair
Claim, wherein the cross line style filter shape for constituting final filter shape is 9 × 7 sizes, and constitutes final filter shape
Square filter shape is 3 × 3 sizes;Or constituting the cross line style filter shape of final filter shape is 7 × 5 sizes or 11
× 9 sizes, and the square filter shape for constituting final filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
The cross line style filter shape and square filter shape that various embodiments are proposed according to the present invention are at center
Locate the final filter shape of superposition, which is the pixel more chosen in horizontal direction relative to vertical direction
Unsymmetric structure design.To not only improve ALF filter effect, but also realize and spend at least in hardware realization.
Detailed description of the invention
It refers to the following detailed description in conjunction with the accompanying drawings, the feature, advantage and other aspects of each embodiment of the present invention will become
It obtains more obvious.In the accompanying drawings:
Fig. 1 shows the schematic block diagram of the encoder according to HEVC/H.265 standard;
Fig. 2 shows the schematic block diagrams according to the decoder of HEVC/H.265 standard;
Fig. 3 shows the two-dimentional diamond of 5 × 5,7 × 7 and 9 × 7 three kinds of sizes used by the ALF of HEVC/H.265 standard
Type filter shape;
Fig. 4 shows the two-dimensional symmetric filter shape that size used by the ALF of AVS2 standard is 7 × 7;
Fig. 5 shows the basic block unit and label sequential schematic that the reconstruction image of AVS standard divides;
Fig. 6 shows the asymmetric filter shape of two dimension used by ALF according to embodiment of the present invention;And
Fig. 7 shows the asymmetric filter shape of two dimension used by ALF according to another implementation of the invention.
Specific embodiment
The principle and spirit of the invention are described with reference to several illustrative embodiments below in conjunction with attached drawing.It should manage
Solution, provide these embodiments just for the sake of make those skilled in the art can better understand that in turn realize the present invention, and
Not limit the scope of the invention in any way.
In HEVC/H.265 standard, image is divided into basic coding unit when being encoded, i.e., maximum coding is single
First (LCU), such as 64 × 64 pixels.According to the difference of coding process, and LCU is divided for three parts, respectively coding unit
(CU), predicting unit (PU) and converter unit (TU).
HEVC/H.265 standard additionally uses intra prediction and inter-frame prediction techniques, wherein using vision signal space and
Redundancy of time achievees the purpose that Information Compression: when compressing current coding unit or coded sub-units, using around it
Encoded Pixel Information predicts the pixel value in current coding unit or coded sub-units.Prediction process is usually to make
With certain several prediction algorithm predetermined on the basis of neighborhood pixels, one, current coded unit/subelement is generated in advance
Survey block of pixels.According to the assessment to distortion performance, the highest prediction mode of encoder efficiency of selection.Encoder related with this
Concrete operations include: 1) encode used in prediction mode index;2) current coded unit/subelement and prediction pixel block
It is poor to make, and obtains residual block;3) residual block converted, quantified and entropy coding.The respective operations of decoder include: 1) decoding prediction
The index of mode obtains prediction mode according to index, and calculates corresponding intra-frame prediction block;2) entropy decoding transformation coefficient, inverse amount
Change, inverse transformation obtains residual block;3) block of pixels that addition prediction block and residual block are rebuild.
Fig. 1 shows the schematic block diagram of the encoder according to HEVC/H.265 standard, and Fig. 2 shows bases
The schematic block diagram of the decoder of HEVC/H.265 standard.As shown in Figure 1, given current prediction unit PU, is indicated with x, then may be used
Prediction PU is obtained first to predict by intra prediction (or interframe), is indicated with x '.Prediction PU, x ' it then can be by from working as
It is subtracted in preceding PU, x, generates PU residual error, indicated with e.And being grouped to PU residual error associated with coding unit CU, e
The CU residual error of generation may then pass through transformation and be converted, and per next converter unit TU, generate the PU residual error in transform domain,
It is indicated with E.The transformation can be converted for example using the block of square or non-square.
PU residual error, E may then pass through quantizer module 118 and quantified, so that high-precision transformation coefficient is converted into
A limited number of probable value.As it should be appreciated, quantization be damage operation, and quantify lose can not usually restore.
Quantified coefficient may then pass through entropy encoding/decoding module 120 and carry out entropy encoding/decoding, generate last compression ratio
It is special.It should be noted that depend on performed encoding and decoding standard, it is described above prediction, transform and quantization can be directed to any view
Frequency data block executes, for example, being directed to the PU or TU of CU, or is directed to LCU.
In order to promote time and spatial prediction, quantified transformation coefficient E can be also taken, and utilize inverse quantization module
122 pairs its carry out inverse quantization, to generate inverse quantized transformation coefficient E '.Then inverse quantized transformation coefficient passes through inversion
It changes the mold block 124 and carries out inverse transformation, generate reconstructed PU residual error, indicated with e '.Reconstructed PU residual error, e ' are then in time
Or spatially it is added into corresponding prediction PU, x ', to form reconstructed PU, indicated with x ".
In loop filter 126, de-blocking filter (" DFB ") can be executed on reconstructed predicting unit PU, x "
Operation, to reduce blocking artifact first.It can be after completing for the de-blocking filter processing of decoded image conditionally
Adaptive sampling point compensation (" SAO ") process is executed, this compensates the biasing of the pixel value between reconstructed pixel and original pixels.?
After DBF operation and SAO process, auto-adaptive loop filter ALF function, adaptive loop filter ALF function can also be performed
It can be conditionally executed on reconstructed PU.Adaptive loop filter ALF minimize output and input image it
Between encoding and decoding distortion.In some embodiments, loop filter module 126 works during inter picture prediction loop.If
Reconstructed image is reference picture, then they can be stored in reference buffer 128 predicts for time in the future.
As shown in Fig. 2, it illustrates the schematic block diagrams according to the decoder of HEVC/H.265 standard.Decoder 145
Entropy decoder module 146 can decode code stream after coding.Entropy decoder module 146 can be in conjunction with entropy encoding/decoding module 120 to execute
The inverse process of the process of description.Then coefficient can be provided to inverse quantization module 147 by entropy decoder module 146, to coefficient square
Battle array carries out inverse quantization, generates E '.Inverse quantization module 147 can provide inverse quantized coefficient to inverse transform module 149.Inversion
Transform operations can be executed to coefficient by changing the mold block 149, generate e '.Then, loop can be applied in a manner of combining Fig. 1 description
Filtering and spatial prediction.
For HEVC/H.265 standard, as previously mentioned, a complete loop filtering process includes 3 links: deblocking
It filters (Deblocking Filter, DBF), adaptive sampling point compensation (Sample Adaptive Offset, SAO), adaptive
Loop filtering (Adaptive Loop Filter, ALF).Reconstruction image is after passing through deblocking filtering and adaptive equalization, area
Noise is still had inside block.In HEVC/H.265 standard, adaptive loop filter (ALF) is after SAO or deblocking filtering
It carries out, for restoration and reconstruction image to reach the mean square deviation between reconstruction image and original image (MSE) minimum, so as to further
Reduce the distortion between reconstruction image and source images.The picture quality for thus further improving reconstructed frame, to provide fortune
The accuracy of dynamic compensation and estimation.ALF uses two dimensional wiener filtering device, and filter shape uses 5 × 5,7 × 7 and 9 × 7 three
The two-dimentional brilliant filter construction of kind size, as shown in Figure 3.Filter shape is defined and is needed when being filtered to a pixel
Consider the position of surrounding pixel.The filter factor of sample window corresponding to filter shape carries out certainly according to local characteristics
It adapts to calculate.At encoder, ALF calculates one group of filter coefficient according to Wiener filtering principle and is transmitted to this group of coefficient
Decoder.The filter factor of these filter constructions is calculated by adaptive wiener filter, can be by adaptively instructing
It gets.
For every width reconstruction image, ALF can be used for a whole frame, can be used for Partial Block therein.According to filter
The difference of wave control mode, ALF can be divided into three classes: the ALF based on frame, the ALF based on block, the ALF based on quaternary tree.Base
It is suggested earliest in the ALF of frame, whole frame application ALF can be directed to, only one mark is used to indicate to answer when present frame
It is filtered with ALF.After being possible to will lead to using the ALF based on frame and finishing filtering, the image in some regions in the frame image
Quality can improve, but the pixel error in other regions increases instead, image quality decrease, thus develop based on block
ALF.ALF based on quaternary tree uses finer block division methods, and smaller filter is used at image border and details
Wave block, and bigger block is used for image flat region.The size of block is automatic according to the property of the different zones of image
It adjusts, to improve code efficiency.
AVS2 standard equally uses deblocking filtering equally using block-based prediction and quantization scheme wherein
DBF, SAO and adaptive loop filter ALF.Fig. 4 show filter shape used by AVS2 standard and filter factor (C0,
C1 ... C8) layout.For AVS2 compared with HEVC/H.265 standard, the filter shape of ALF filter uses simpler pair
Claim structure, and filter factor is about central point central symmetry, to greatly simplifie the calculation amount and complexity of ALF filtering
Degree.
In AVS2 standard, present image is divided into 16 fundamental regions, as shown in figure 5, the label of fundamental region is suitable
Sequence can be sequence as figure 5 illustrates.It can be using 7 × 7 cross line styles and 3 × 3 as shown in Figure 4 just for each fundamental region
One filter of the symmetric shape of rectangular superposition.Filter factor (C0, C1 ... C8) is center symmetrical structure, can be according to
The smallest principle of mean square error trains the filter factor of the filter of each fundamental region, and wherein the smallest principle of mean square error is
It is that reconstructed pixel value and original pixel value are distorted the smallest principle.
It, can be using filter shape used by ALF function and filter factor as fundamental region at encoder
ALF parameter coding is transmitted to decoder into code stream.At decoder, decoder can parse the ALF ginseng transmitted from encoder
Number obtains filter shape and filter factor.When carrying out ALF filtering in encoding-decoding process, for every in a fundamental region
A pixel selects to be filtered used surrounding pixel to the pixel according to the filter shape in the region, corresponding position
Pixel and corresponding filter factor the sum of be multiplied obtained new pixel to substitute current pixel value, wherein filter shape
Center window corresponds to the pixel.
Hereinafter, improved ALF filter shape according to various embodiments of the present invention is specifically described.
Due in video image coding and decoding process be by raster scan order (raster scan) Lai Jinhang, that
When taking vertical orientation data, full line data must be cached in hardware realization, horizontal direction is because rigid encoding and decoding come out, in hardware
Backup has corresponding pixel value, therefore, if the neighbor pixel on the vertical direction for taking a pixel in the realization of ALF filtering
When, a few row data and following a few row data above are accessed then will use, then this memory space resource for needing several rows.This is right
Be in the consumption of hardware resource it is very big, especially face 4K super clear image when, the data volume to be stored is very big.Cause
This, it is possible to reduce the pixel on vertical direction is chosen, and selects the neighbor pixel in horizontal direction more to carry out ALF filtering.
For this purpose, according to various embodiments of the present invention, building cross line style filter shape and square filter shape exist
The final filter shape being superimposed at center, or be Filtering Template, wherein the cross hairs type filter shape is in the horizontal direction
Sample window number be greater than sample window number in the vertical direction (corresponding to a square in Fig. 6).In this way, passing through
Horizontal direction increases the selection of filtered samples window (for selecting the pixel of corresponding position), not only increases the quality of image,
And the cost that hardware realization is spent is seldom, also can be good at the quality and filter effect that improve image.
Fig. 6 shows the Filtering Template of the two-dimentional unsymmetric structure of ALF function according to an embodiment of the present invention.It should
Filtering Template be superimposed by cross line style filter shape and the square filter shape of 3 × 3 sizes of 9 × 7 sizes at center and
At.Filter factor (C0, C1 ... C9) use centrosymmetric structure.Alternatively, filter factor (C0, C1 ... C9)
Asymmetrical structure can be used.
Fig. 7 shows the Filtering Template of the two-dimentional unsymmetric structure of the ALF function of another embodiment according to the present invention.
The Filtering Template is superimposed with the square filter shape of 3 × 3 sizes at center by the cross line style filter shape of 7 × 5 sizes
It forms.
The Filtering Template of another embodiment according to the present invention, the two-dimentional unsymmetric structure of ALF function can be by 11
The cross line style filter shape of × 9 sizes is formed by stacking at center with the square filter shape of 3 × 3 sizes.
Another embodiment according to the present invention, the Filtering Template of the two-dimentional unsymmetric structure of ALF function can be by can
To be formed by stacking at center by the cross line style filter shape of 11 × 9 sizes and the square filter shape of 4 × 4 sizes.
Embodiment of the present invention between reconstructed frame and primitive frame by using improved final asymmetric filter shape
(or Filtering Template), such as 9 × 7 asymmetric template as shown in FIG. 6, Lai Youhua loop filter.It should be appreciated that the present invention is real
The Filtering Template that the mode of applying is proposed can in conjunction with HEVC/H.265 standard, AVS2 standard and above-mentioned standard evolution version and
It uses, the corresponding portion in them can be replaced.Filter factor corresponding to the ALF Filtering Template more proposed can be according to such as
The smallest principle of mean square error in HEVC/H.265 standard, AVS2 standard is trained.Additionally, embodiment of the present invention institute
The ALF filter shape of proposition can also be applied in other coding and decoding video schemes.
Although in the above description, ALF is used in after DBF and SAO, the present invention is not intended to limit itself and DBF
With the relative positional relationship of SAO.For example, ALF can be filtered between DBF and SAO, or independently of DBF and SAO
Wave.
It should be appreciated that embodiment according to the present invention, the division of the fundamental region of present image as shown in Figure 5 and base
The label sequence of one's respective area is only a kind of example, and the ALF Filtering Template of embodiment of the present invention can also be with other base regions
Domain division mode and label sequence are used in combination.
According to various embodiments of the present invention, a kind of adaptive loop filter side for video image encoding and decoding is proposed
Method, comprising:
Reconstruction image is divided into multiple blocks;And
For each block in the multiple block, it is filtered according to certain filter shape and filter factor, institute
It states filter shape to be made of multiple sample windows, each sample window is used to select the pixel of corresponding position, and is provided with one
A filter factor, the filter factor are designed to make the mean square deviation between the reconstruction image and original image minimum.
For a block in the multiple block, cross line style filter shape and square filter shape are constructed in
The final filter shape being superimposed at the heart, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than
Sample window number in the vertical direction.
For example, the block size that reconstruction image is divided into can be 64 × 64,32 × 32,16 × 16 and 8 × 8.Weight
The block for building image is also based on quaternary tree and is further divided into sub-block, applies such sub-block according to this hair
Bright various embodiment ALF.
According to various embodiments of the present invention, it is also proposed that a kind of adaptive loop filter for video image encoding and decoding
Equipment, comprising:
Division module, for reconstruction image to be divided into multiple blocks;And
Filter module, for being according to certain filter shape and filtering for each block in the multiple block
Number is filtered, and the filter shape is made of multiple sample windows, and each sample window is used to select the pixel of corresponding position,
And it is provided with a filter factor, the filter factor is designed to make equal between the reconstruction image and original image
Variance is minimum.
For a block in the multiple block, cross line style filter shape and square filter shape are constructed in
The final filter shape being superimposed at the heart, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than
Sample window number in the vertical direction.
According to various embodiments of the present invention, it is also proposed that a kind of view that coding is handled using adaptive loop filter
The method of frequency image, comprising:
Receive the reconstructed image data for corresponding to the video image data of coding;
Adaptive loop filter is applied to the image data of the reconstruction to generate the video image data through filtering;
And
Video data through filtering is provided.
The final filter shape that the adaptive loop filter uses is filtered by cross line style filter shape and square
Shape is superimposed at center to construct, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than
Sample window number in the vertical direction.
According to various embodiments of the present invention, it is also proposed that a kind of view that coding is handled using adaptive loop filter
The equipment of frequency image, comprising:
Receiving module, for receiving the reconstructed image data for corresponding to the video image data of coding;
Filter module is filtered for adaptive loop filter to be applied to the image data of the reconstruction with generating
Video image data;And
Module is provided, for providing the video data through filtering.
The final filter shape that the adaptive loop filter uses is filtered by cross line style filter shape and square
Shape is superimposed at center to construct, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than
Sample window number in the vertical direction.
According to various embodiments of the present invention, compared with the filter shape used in AVS2 standard, level side is changed
To the selection number with the filtered samples window (for selecting the pixel of corresponding position) on vertical direction, so that in the horizontal direction
On sample window number be greater than sample window number in the vertical direction, and improve the quality (selection of filtering image
More neighbor pixels), and make the cost incrementss very little in hardware realization.With traditional two-dimentional brilliant template phase
Than the window on focal selection current central sample window periphery is reduced in filtering computational complexity.
It is soft by being referred in AVS2 by taking the 9 × 7 of embodiment of the present invention as shown in FIG. 6 asymmetric Filtering Templates as an example
The ALF filtering (as shown in Figure 47 × 7 templates) of AVS2 standard is carried out on part RD11.0 and using the 9 × 7 non-of the embodiment
The performance comparison test of the ALF filtering of symmetrical Filtering Template, test result is as follows shown in table.
In the test of upper table, have chosen 14 cycle tests altogether, including UHD resolution ratio 2 image sequences,
4 image sequences of 4 image sequences of 1080p resolution ratio, 4 image sequences of WVGA resolution ratio and 720p resolution ratio.It is right
In this 14 test image sequences, quantization parameter quantization step (QP) is had chosen respectively and is followed successively by 27,32,38 and 45 building tests
Use-case is tested.
As seen from the above table, ALF filtering are carried out using as shown in FIG. 69 × 7 asymmetric Filtering Templates, with use such as Fig. 4 institute
The filter shape shown carries out ALF filtering and compares, and averagely has 0.06% promotion for the luminance component Y, RA of image, LD averagely has
0.2% promotion, AI averagely have 0.02% promotion;For the chromatic component U and V of image, RA averagely has 0.07% He respectively
0.28% promotion, LD averagely have 0.12% and 0.02% promotion respectively, and AI averagely has 0.08% and 0.08% to mention respectively
It rises.
Inventor is also to as shown in FIG. 69 × 7 asymmetric Filtering Templates and including 7 × 5 asymmetric filters as shown in Figure 7
Wave template and above-mentioned 11 × 9 two kinds of asymmetric Filtering Templates have carried out contrast test, and test result shows shown in fig. 6
Filter shape is that effect is optimal in terms of the equilibrium of image quality improvement amount and computation complexity.
It should be appreciated that although the ALF scheme that embodiment of the present invention is proposed is particularly suitable for hardware realization, the party
Case also may be implemented as the combination of software or hardware and software.Hardware can be by being designed to perform the conversion collection of above-mentioned function
At circuit (ASIC), digital signal processor (DSP), programmable logic device (PLD), field programmable gate array (FPGA),
Processor, controller, microprocessor, other electronic units or combinations thereof are realized.Software can pass through the mould of the above-mentioned function of execution
Block is realized.Software can be stored in storage unit and be run by processor.As storage unit or processor, this can be used
Various units well known to the technical staff of field.
Give specification of the invention for the purpose for showing and describing, but its be not intended to exhaustion or
Person is limited to the invention of disclosed form.It may occur to persons skilled in the art that many modification and variation.The present invention is not being departed from
Under the premise of spirit, all modifications and replacement made are fallen in the scope of the present invention that appended claims define.
Claims (4)
1. a kind of adaptive loop filter method for video image encoding and decoding, the super clear image suitable for 4k, comprising:
Reconstruction image is divided into multiple blocks;And for each block in the multiple block, according to certain filtering
Shape and filter factor are filtered, and the filter shape is made of multiple sample windows, and each sample window is for selection pair
The pixel of position is answered, and is provided with a filter factor, the filter factor is designed to make the reconstruction image and original
Mean square deviation between beginning image is minimum,
Each block in the multiple block, is filtered packet according to certain filter shape and filter factor
It includes: for the current pixel in current block, selecting the current pixel to be filtered according to the filter shape used
Surrounding pixel the sum of is multiplied the surrounding pixel and the corresponding filter factor obtained new pixel to replace described work as
Preceding pixel;
It is characterized in that, building cross line style filter shape and square filter for a block in the multiple block
The final filter shape that shape is superimposed at center, the final filter shape are more to choose level relative to vertical direction
The unsymmetric structure of pixel on direction, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is big
In sample window number in the vertical direction, the filter factor of the sample window corresponding to the final filter shape be about
The final filter shape central symmetry or quadrant symmetrical, wherein
The cross line style filter shape for constituting the final filter shape is 9 × 7 sizes, and constitutes the final filter shape
Square filter shape be 3 × 3 sizes;Or
The cross line style filter shape for constituting the final filter shape is 7 × 5 sizes or 11 × 9 sizes, and described in composition
The square filter shape of final filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
2. a kind of adaptive loop filter equipment for video image encoding and decoding, the super clear image suitable for 4k, comprising:
Division module, for reconstruction image to be divided into multiple blocks;And
Filter module, for for each block in the multiple block, according to certain filter shape and filter factor into
Row filtering, the filter shape are made of multiple sample windows, and each sample window is used to select the pixel of corresponding position, and
It is provided with a filter factor, the filter factor is designed to make the mean square deviation between the reconstruction image and original image
Minimum, filter module are specifically used for for the current pixel in current block, select the current picture according to the filter shape
Element is filtered used surrounding pixel, and the surrounding pixel and the corresponding filter factor the sum of are multiplied, and it is new to obtain
Pixel replace the current pixel;
It is characterized in that, building cross line style filter shape and square filter for a block in the multiple block
The final filter shape that shape is superimposed at center, the final filter shape are more to choose level relative to vertical direction
The unsymmetric structure of pixel on direction, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is big
In sample window number in the vertical direction, the filter factor of the sample window corresponding to the final filter shape be about
The final filter shape central symmetry or quadrant symmetrical, wherein
The cross line style filter shape for constituting the final filter shape is 9 × 7 sizes, and constitutes the final filter shape
Square filter shape be 3 × 3 sizes;Or
The cross line style filter shape for constituting the final filter shape is 7 × 5 sizes or 11 × 9 sizes, and described in composition
The square filter shape of final filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
3. a kind of method for the video image for handling coding using adaptive loop filter, suitable for the super clear image of 4k, packet
It includes:
Receive the reconstructed image data for corresponding to the video image data of coding;
Adaptive loop filter is applied to the image data of the reconstruction to generate the video image data through filtering, it is described from
Adapting to loop filtering includes selecting institute according to filter shape for the current pixel in current block in the reconstructed image data
It states current pixel and is filtered used surrounding pixel, the surrounding pixel and corresponding filter factor the sum of are multiplied and obtained
New pixel replace the current pixel;And
Video data through filtering is provided,
It is characterized in that, the final filter shape that the adaptive loop filter uses passes through cross line style filter shape and pros
Shape filter shape is superimposed to construct at center, and the final filter shape is more to choose level side relative to vertical direction
The unsymmetric structure of upward pixel, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than
Sample window number in the vertical direction, the filter factor of the sample window corresponding to the final filter shape are about institute
State final filter shape central symmetry or quadrant symmetrical, wherein
The cross line style filter shape for constituting the final filter shape is 9 × 7 sizes, and constitutes the final filter shape
Square filter shape be 3 × 3 sizes;Or
The cross line style filter shape for constituting the final filter shape is 7 × 5 sizes or 11 × 9 sizes, and described in composition
The square filter shape of final filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
4. a kind of equipment for the video image for handling coding using adaptive loop filter, suitable for the super clear image of 4k, packet
It includes:
Receiving module, for receiving the reconstructed image data for corresponding to the video image data of coding;
Filter module, for adaptive loop filter to be applied to the image data of the reconstruction to generate the video figure through filtering
As data, specifically for selecting institute according to filter shape for the current pixel in current block in the reconstructed image data
It states current pixel and is filtered used surrounding pixel, the surrounding pixel and corresponding filter factor the sum of are multiplied and obtained
New pixel replace the current pixel;And
Module is provided, for providing the video data through filtering,
It is characterized in that, the final filter shape that the adaptive loop filter uses passes through cross line style filter shape and pros
Shape filter shape is superimposed to construct at center, and the final filter shape is more to choose level side relative to vertical direction
The unsymmetric structure of upward pixel, wherein the sample window number of the cross hairs type filter shape in the horizontal direction is greater than
Sample window number in the vertical direction, the filter factor of the sample window corresponding to the final filter shape are about institute
State final filter shape central symmetry or quadrant symmetrical, wherein
The cross line style filter shape for constituting the final filter shape is 9 × 7 sizes, and constitutes the final filter shape
Square filter shape be 3 × 3 sizes;Or
The cross line style filter shape for constituting the final filter shape is 7 × 5 sizes or 11 × 9 sizes, and described in composition
The square filter shape of final filter shape is respectively 3 × 3 sizes or 4 × 4 sizes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510698486.5A CN105306957B (en) | 2015-10-23 | 2015-10-23 | Adaptive loop filter method and apparatus |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510698486.5A CN105306957B (en) | 2015-10-23 | 2015-10-23 | Adaptive loop filter method and apparatus |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105306957A CN105306957A (en) | 2016-02-03 |
CN105306957B true CN105306957B (en) | 2019-04-26 |
Family
ID=55203632
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510698486.5A Active CN105306957B (en) | 2015-10-23 | 2015-10-23 | Adaptive loop filter method and apparatus |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105306957B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017133660A1 (en) * | 2016-02-04 | 2017-08-10 | Mediatek Inc. | Method and apparatus of non-local adaptive in-loop filters in video coding |
US20180041778A1 (en) * | 2016-08-02 | 2018-02-08 | Qualcomm Incorporated | Geometry transformation-based adaptive loop filtering |
WO2019170259A1 (en) * | 2018-03-09 | 2019-09-12 | Huawei Technologies Co., Ltd. | Method and apparatus for image filtering with adaptive multiplier coefficients |
CN109862374A (en) * | 2019-01-07 | 2019-06-07 | 北京大学 | A kind of adaptive loop filter method and device |
WO2021128265A1 (en) * | 2019-12-27 | 2021-07-01 | 富士通株式会社 | Filtering method and device |
CN113824956B (en) * | 2020-08-24 | 2022-12-23 | 杭州海康威视数字技术股份有限公司 | Filtering method, device, equipment and machine-readable storage medium |
CN113852831B (en) * | 2021-03-05 | 2023-03-28 | 杭州海康威视数字技术股份有限公司 | Filtering method, device, equipment and machine-readable storage medium |
WO2022218281A1 (en) * | 2021-04-12 | 2022-10-20 | Beijing Bytedance Network Technology Co., Ltd. | Guided filter in video coding |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101572542A (en) * | 2008-04-28 | 2009-11-04 | 汤姆森特许公司 | Cycle slip detection for timing recovery |
CN103051892A (en) * | 2011-10-14 | 2013-04-17 | 联发科技股份有限公司 | Method and apparatus for in-loop filtering |
CN103503456A (en) * | 2011-05-10 | 2014-01-08 | 联发科技股份有限公司 | Method and apparatus for reduction of in-loop filter buffer |
CN103891292A (en) * | 2011-10-24 | 2014-06-25 | 联发科技股份有限公司 | Method and apparatus for non-cross-tile loop filtering |
CN104702963A (en) * | 2015-02-13 | 2015-06-10 | 北京大学 | Boundary treatment method and device for adaptive loop filter |
CN104735450A (en) * | 2015-02-26 | 2015-06-24 | 北京大学 | Method and device for carrying out adaptive loop filtering during video coding and decoding |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP4095839B2 (en) * | 2002-06-24 | 2008-06-04 | 株式会社日立グローバルストレージテクノロジーズ | Two-stage actuator positioning controller |
-
2015
- 2015-10-23 CN CN201510698486.5A patent/CN105306957B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101572542A (en) * | 2008-04-28 | 2009-11-04 | 汤姆森特许公司 | Cycle slip detection for timing recovery |
CN103503456A (en) * | 2011-05-10 | 2014-01-08 | 联发科技股份有限公司 | Method and apparatus for reduction of in-loop filter buffer |
CN103051892A (en) * | 2011-10-14 | 2013-04-17 | 联发科技股份有限公司 | Method and apparatus for in-loop filtering |
CN103891292A (en) * | 2011-10-24 | 2014-06-25 | 联发科技股份有限公司 | Method and apparatus for non-cross-tile loop filtering |
CN104702963A (en) * | 2015-02-13 | 2015-06-10 | 北京大学 | Boundary treatment method and device for adaptive loop filter |
CN104735450A (en) * | 2015-02-26 | 2015-06-24 | 北京大学 | Method and device for carrying out adaptive loop filtering during video coding and decoding |
Also Published As
Publication number | Publication date |
---|---|
CN105306957A (en) | 2016-02-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105306957B (en) | Adaptive loop filter method and apparatus | |
KR102401749B1 (en) | Apparatus and method for image coding and decoding | |
CN103503456B (en) | In-loop treatment method for reestablishing video and apparatus thereof | |
CN113287316B (en) | Method and apparatus for reducing loop filter buffer | |
CN103931180B (en) | Image decoding apparatus | |
CN104935941B (en) | The method being decoded to intra prediction mode | |
CN103460699B (en) | Loop circuit filtering method and its equipment | |
TWI729378B (en) | Method and apparatus of neural network for video coding | |
CN103209324B (en) | A kind of method for applying edge compensation | |
TWI647951B (en) | Image coding equipment | |
CN103947208B (en) | Reduce the method and device of deblocking filter | |
MX2013002070A (en) | Method and apparatus of adaptive loop filtering. | |
JP2014195269A (en) | Multilevel effectiveness mapping scanning | |
KR20130045154A (en) | Image decoding apparatus | |
MX2014012311A (en) | Moving image encoding device, moving image decoding device, moving image encoding method and moving image decoding method. | |
JP2019106742A (en) | Image encoding device, image decoding device, and programs for the same | |
CN105357538B (en) | Adaptive loop filtering method and apparatus | |
US10979712B2 (en) | Block-based video decoding method using pre-scan and apparatus thereof | |
CN116114246B (en) | Intra-frame prediction smoothing filter system and method | |
CN116438567A (en) | In-loop filtering based on neural network for video coding | |
CN105049846B (en) | The method and apparatus of image and coding and decoding video |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
TA01 | Transfer of patent application right | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20190327 Address after: 411100 room 1006, innovation building, 9 Xiao Tang Road, hi tech Zone, Xiangtan, Hunan Applicant after: Xiangtan Zhongxing Electronics Co., Ltd. Address before: 100083 Beijing Haidian District 35 Xueyuan Road 16 building Applicant before: Beijing Vimicro Corporation |
|
GR01 | Patent grant | ||
GR01 | Patent grant |