CN106507116B - A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis - Google Patents
A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis Download PDFInfo
- Publication number
- CN106507116B CN106507116B CN201610889330.XA CN201610889330A CN106507116B CN 106507116 B CN106507116 B CN 106507116B CN 201610889330 A CN201610889330 A CN 201610889330A CN 106507116 B CN106507116 B CN 106507116B
- Authority
- CN
- China
- Prior art keywords
- conspicuousness
- block
- depth
- information
- present encoding
- 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
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
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/597—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding specially adapted for multi-view video sequence encoding
-
- 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
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/103—Selection of coding mode or of prediction mode
- H04N19/109—Selection of coding mode or of prediction mode among a plurality of temporal predictive coding modes
-
- 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
- H04N19/86—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression involving reduction of coding artifacts, e.g. of blockiness
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention relates to a kind of 3D-HEVC coding methods predicted based on 3D conspicuousness information and View Synthesis, comprise the concrete steps that: 1) establishing 3D conspicuousness model, obtain conspicuousness information；2) the assigned area of present encoding block is determined；3) coding strategy in different demarcation region is determined；4) original View Synthesis prediction algorithm is improved；5) multi-view video sequences are encoded.This method utilizes human-eye visual characteristic, effectively reduces blocking artifact, while guaranteeing to encode objective quality, can save code rate, improve code efficiency.
Description
Technical field
The present invention relates to a kind of three-dimensional high-performance video coding (Three-dimensional high efficiency
Video coding, 3D-HEVC) method, it is especially a kind of that (View is predicted based on 3D conspicuousness information and View Synthesis
Synthesis prediction, VSP) 3D-HEVC coding method.
Background technique
Currently, video compression technology development is swift and violent, while also having driven the development and research of 3D video coding technique.HEVC
It is current state-of-the-art coding standard, and the proposition of 3D-HEVC is intended to improve coding techniques on the basis of HEVC, it is more with Efficient Compression
Viewpoint video and corresponding depth data.
In order to further increase 3D video coding performance, VSP has been added in 3D-HEVC coding framework, that is, utilizes
Depth information carries out better predictive coding to texture.VSP is a kind of interview prediction coding method, and algorithm idea is to refer to
In viewpoint, a prediction block is obtained using the present encoding block that synthesizes of depth information.VSP points are predicted for forward direction View Synthesis
(Forward view synthesis prediction, FVSP) and backward View Synthesis predict (Backward view
Synthesis prediction, BVSP), wherein FVSP is suitable for the mode of coding depth after first encoding texture, and what is utilized is
Pixel all in reference picture is synthesized in virtual image by the depth map of reference view.It is deep that BVSP is suitable for first coding
The mode of encoding texture after degree, what is utilized is the depth map of current view point, finds pair in present encoding block in a reference image
Pixel is answered, the prediction block of current block is obtained.Since FVSP decoding complex degree is very high, 3D-HEVC is using follow-on
BVSP.The texture of present encoding block reference is encoded image between different points of view, and the depth of reference is that encoded image is corresponding
Depth map, it is therefore an objective to find a prediction block reduction view for predicting unit (Prediction units, the PU) block of present encoding
Redundancy between point.Using the View Synthesis prediction algorithm based on maximum depth value in the platform of current HTM-15.0, that is, compare
Relatively entirely encode the upper left of PU block and the depth value of bottom right, upper right and lower-left.If upper left is greater than bottom right and upper right is greater than lower-left,
Then it is divided into 28 × 4 pieces for all 8 × 8 pieces, then the depth value at more each 8 × 4 pieces of 4 angles, maximum value therein is taken to make
8 × 4 pieces of depth value thus；Conversely, being then divided into 24 × 8 pieces for all 8 × 8 pieces.Then each 8 × 4 or 4 × 8 are small
The parallax of maximum depth value conversion in block is stored.Since 3D-HEVC does not do deblocking filtering to depth block, so generate
Depth map has obvious blocking artifact, and the maximum value in four pull angle value of sub- PU block is taken in VSP primal algorithm, if the son
Block is just on the boundary of blocking artifact, then obtaining maximum depth value may be inaccurate, to influence coding result.Therefore it needs
The case where innovatory algorithm is wanted to avoid sub-block from being located at blocking artifact boundary generation.
In recent years, people were more and more interested in the research of human-eye visual characteristic, and perception Video coding is increasingly by weight
Depending on.Since human visual system (Human visual system, HVS) has selectivity to the perception of video scene and image,
There is different sensing capabilities to different zones or different objects.In general, object or texture-rich of the human eye to motion intense
Region have higher perceptual sensitivity, these regions are known as salient region.Traditional video encoding standard focus on from
Redundancy is reduced to improve distortion performance, and ignores influence of the HVS perception diversity to Video coding.Therefore, the present invention mentions
Coding method out based on 3D conspicuousness information.
Summary of the invention
It is a kind of based on 3D conspicuousness information and view the purpose of the present invention is proposing on the basis of current video coding technique
The 3D-HEVC coding method of point synthesis prediction.This method is based on 3D conspicuousness information and improved View Synthesis prediction algorithm,
Code efficiency can be effectively improved, coding quality is improved.
In order to achieve the above objectives, insight of the invention is that
3D conspicuousness model is established according to the visual characteristic of human eye first, obtains the 3D conspicuousness information of coded sequence, and
Conspicuousness information is imported in original encoding mechanism；Encoding block is divided into significantly further according to the conspicuousness information of coded sequence
Region, non-significant region and the type of intermediate region three, different quantization parameters is distributed according to different types of region
(Quantization parameter, QP) and coding mode；Then in order to avoid in VSP coded sub-blocks appear in blocking artifact side
Situation in boundary occurs, and the position at each four angles of sub-block is translated a pixel unit respectively inwards, takes four after inside contracting
The maximum value at angle is as the depth value to be stored；Conspicuousness information and improved VSP algorithm is finally combined to carry out Video coding,
Improve code efficiency.
According to above-mentioned design, the technical scheme is that
A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis, specific steps include:
1) 3D conspicuousness model is established, conspicuousness information is obtained: on the basis of existing 2D conspicuousness model, passing through increase
The depth information of video image establishes the 3D conspicuousness model with depth perception information, obtains the 3D conspicuousness of encoded content
Information；
2) the assigned area of present encoding block is determined: according to the 3D conspicuousness information of encoded video, by present encoding
Block division of teaching contents is salient region, non-limiting region or intermediate region；
3) coding strategy in different demarcation region is determined: aobvious for difference after carrying out region division to present encoding block
Work degree distributes different Q P value, and sets different coding modes；
4) original View Synthesis prediction algorithm is improved: for the boundary effect for avoiding depth block, by original VSP neutron PU block
Four angles inside contract a pixel respectively, i.e., the X at sub- four angles of PU block, Y-axis translate a pixel respectively inwards, then take newly-generated
The maximum depth value at four angles stored；
5) multi-view video sequences are encoded: utilizing 3D conspicuousness information and improved VSP algorithm, carry out multi-view point video sequence
The coding of column.
In above-mentioned steps 1) in, first it is normalized the depth map D of original series to obtain depth weights omega, due to
Depth pixel value is 0-255, numerical value is bigger represent it is closer with a distance from video camera, thereforeFollowed by existing at present
2D conspicuousness model obtain texture video notable figure SC, texture notable figure and depth multiplied by weight are obtained into 3D notable figure
SDC, i.e. SDC=ω * SC；SDCDepth information is increased on the basis of 2D texture notable figure, meets human eye vision attention machine
System, has achieved the purpose that its significance of more closer intermediate object is higher.
The step 2) includes the following steps:
A) the 3D notable figure S for first obtaining step 1)DCIt carries out binary conversion treatment and obtains Sb, i.e.,
Wherein, T is adaptive threshold, is also changed correspondingly according to the value of the different T of video content, after such binaryzation
Notable figure SbPossessing and is more clear clearly demarcated profile, the highlighted place that pixel value is 1 in image is the area-of-interest of human eye, as
Element is background area for 0；
B) the significant mean value S of present encoding block is calculatedAve=ST/ (H*W), wherein STIndicate binaryzation in present encoding block
The pixel value summation of notable figure, H, W respectively represent the height and width of present encoding block, SAveAs present encoding block is averaged
Significance carries out region division to current block by calculating the average significance of current block:
For intermediate region, and it is divided into two kinds of situations:
In this way according to average significance, present encoding block four kinds of different types of regions have been divided into just.
In above-mentioned steps 3) in, coding dependency using present encoding block and contiguous block time and spatially, to current
Block is terminated skipped in advance, i.e. SKIP mode treatment:
A) if present encoding block is in non-limiting region, if in five adjacent blocks there are three or three SKIP selected above
As optimal mode, then it is assumed that the optimum prediction mode of current block be SKIP, otherwise select SKIP and Inter 2N × 2N as
Interframe candidate pattern；
B) when present encoding block is in salient region, if five adjacent blocks all select SKIP mode, then it is assumed that current block
It selects SKIP as optimum prediction mode, all inter-frame modes is otherwise traversed to current block；
C) when present encoding block is in intermediate region, if there are four or four or more adjacent blocks select SKIP, then it is assumed that it is current
Block optimal mode is SKIP, otherwise traverses all inter-frame forecast modes；
D) different quantization parameter QP is distributed to improve code efficiency to different zones block:
In cataloged procedure, coding unit (Coding Units, CU) usual very little of texture complex region, simple flat area
Therefore suitable QP is selected for CU, will have a direct impact on coding quality and bit number with biggish CU domain (such as background)；People
These regions that eye is paid special attention to are exactly salient region, therefore, the sense of coding are improved by adjusting the QP of marking area CU
Know quality；Based on significant information, marking area gives lesser QP, and biggish QP, QP here are distributed in non-significant region0Refer to flat
Initial quantization parameter in platform.
In above-mentioned steps 4) in:
A) upper left of PU block and the depth value of bottom right, upper right and lower-left are relatively entirely encoded: in original coding platform, being write from memory
Recognizing with flag bit vspSize indicates division size, when upper left depth is greater than bottom right and upper right depth is greater than lower-left,
VspSize is 1, otherwise is 0；If vspSize=1, it is divided into two identical 8 × 4 pieces for all 8 × 8 pieces；If
VspSize=0 is then divided into two identical 4 × 8 pieces for all 8 × 8 pieces；
B) in order to avoid the blocking artifact on boundary, comparing each 8 × 4 or 4 × 8 pieces of four angles: upper left, the right side in original platform
Under, upper right, the depth value size of lower-left be changed to original four Angle Positions inside contracting a pixel respectively, i.e., horizontal, ordinate difference
A pixel unit is translated inwards.
In above-mentioned steps 5) in, using 3D-HEVC encoded test platform, text is configured using 3view+3depth in experiment
Part, input is left-in-textures of right three viewpoints and corresponding depth map carry out encoded test；8 videos are tested in experiment in total
Sequence, different sequences have different coding parameters.
Compare with the existing technology of the present invention has following obvious prominent substantive distinguishing features and remarkable advantage:
(1) gained maximum depth value inaccuracy when group PU block is on blocking artifact boundary in View Synthesis prediction is avoided
The case where, it is correspondingly processed to maximum depth value comparison position is chosen, the influence of blocking artifact can be effectively reduced, selection is most
Information is more acurrate for the disparity vector (Disparity vector, DV) of big depth value conversion, to preferably predict current block.
(2) human-eye visual characteristic is fully taken into account, by encoding using information guiding significant in HVS, not by region distribution
Code rate can be saved while guaranteeing coding objective quality with QP value, improves code efficiency.
(3) depth map information is incorporated in conspicuousness model, preferably embodies the characteristic of 3D Video coding.
Detailed description of the invention
Fig. 1 is coding method flow chart of the invention.
Fig. 2 is 3D conspicuousness method for establishing model block diagram.
Fig. 3 is the image that different sequences pass through that 3D conspicuousness model extraction goes out.
Fig. 4 contiguous block schematic diagram for the present encoding block time and spatially.
Fig. 5 is improved View Synthesis predictive coding block schematic diagram.
Specific embodiment
Details are as follows for one embodiment of the present of invention son combination attached drawing table:
Used in the embodiment of the present invention is 3D-HEVC encoded test platform HTM-15.0 version, uses 3view+ in experiment
3depth configuration file, input is left-in-textures of right three viewpoints and corresponding depth map carry out encoded test.In experiment in total
8 video sequences are tested, different sequences have different coding parameters.
Referring to Fig. 1, based on the 3D-HEVC coding method that 3D conspicuousness information and View Synthesis are predicted, first according to people
The visual characteristic of eye establishes 3D conspicuousness model, obtains the 3D conspicuousness information of coded sequence, and conspicuousness information is imported original
In the encoding mechanism of beginning；Further according to coded sequence conspicuousness information by encoding block be divided into marking area, non-significant region and
The type of intermediate region three, different QP and coding mode are distributed according to different types of region；Then in order to avoid being compiled in VSP
Numeral block appears in that blocking artifact is borderline to be happened, and the position at each four angles of sub-block is translated a pixel respectively inwards
Unit takes the maximum value at four after inside contracting angles as the depth value to be stored；Finally combine conspicuousness information and improved
VSP algorithm carries out Video coding, improves code efficiency.Its concrete operation step is:
1) 3D conspicuousness model is established, conspicuousness information is obtained: on the basis of existing 2D conspicuousness model, passing through increase
The depth information of video image establishes the 3D conspicuousness model with depth perception information, obtains the 3D conspicuousness of encoded content
Information；
2) the assigned area of present encoding block is determined: according to the 3D conspicuousness information of encoded video, by present encoding
Block division of teaching contents is salient region, non-limiting region or intermediate region；
3) coding strategy in different demarcation region is determined: aobvious for difference after carrying out region division to present encoding block
Work degree distributes different Q P value, and sets different coding modes；
4) original View Synthesis prediction algorithm is improved: for the boundary effect for avoiding depth block, by original VSP neutron PU block
Four angles inside contract a pixel respectively, i.e., the X at sub- four angles of PU block, Y-axis translate a pixel respectively inwards, then take newly-generated
The maximum depth value at four angles stored；
5) multi-view video sequences are encoded: utilizing 3D conspicuousness information and improved VSP algorithm, carry out multi-view point video sequence
The coding of column.
Referring to fig. 2, Fig. 3, above-mentioned steps 1) in establish 3D conspicuousness model obtain conspicuousness information, refer to original sequence first
The depth map D of column is normalized to obtain depth weights omega, since depth pixel value is 0-255, numerical value is bigger represent from
Video camera distance is closer.ThereforeThe significant of texture video is obtained followed by existing 2D conspicuousness model at present
Scheme SC, texture notable figure and depth multiplied by weight are obtained into 3D notable figure SDC, i.e. SDC=ω * SC。SDCIn 2D texture notable figure
On the basis of increase depth information, meet human eye vision attention mechanism, reached more closer intermediate its significance of object more
High purpose.
Above-mentioned steps 2) in determine that the method for the assigned area of present encoding block is:
A) the 3D notable figure S for first obtaining step 1)DCIt carries out binary conversion treatment and obtains Sb, i.e.,
Wherein, T is adaptive threshold, is also changed correspondingly according to the value of the different T of video content, after such binaryzation
Notable figure SbPossessing and is more clear clearly demarcated profile, the highlighted place that pixel value is 1 in image is the area-of-interest of human eye, as
Element is background area for 0；
B) the significant mean value S of present encoding block is calculatedAve=ST/ (H*W), wherein STIndicate binaryzation in present encoding block
The pixel value summation of notable figure, H, W respectively represent the height and width of present encoding block, SAveAs present encoding block is averaged
Significance carries out region division to current block by calculating the average significance of current block:
For intermediate region, and it is divided into two kinds of situations:
In this way according to average significance, present encoding block four kinds of different types of regions have been divided into just.
Referring to fig. 4, in the step 3) determine different demarcation regional code strategy method be: using present encoding block with
Contiguous block time and coding dependency spatially, terminate current block in advance and are skipped, is i.e. SKIP mode treatment:
A) if present encoding block is in non-limiting region, if in five adjacent blocks there are three or three SKIP selected above
As optimal mode, then it is assumed that the optimum prediction mode of current block be SKIP, otherwise select SKIP and Inter 2N × 2N as
Interframe candidate pattern；
B) when present encoding block is in salient region, if five adjacent blocks all select SKIP mode, then it is assumed that current block
It selects SKIP as optimum prediction mode, all inter-frame modes is otherwise traversed to current block；
C) when present encoding block is in intermediate region, if there are four or four or more adjacent blocks select SKIP, then it is assumed that it is current
Block optimal mode is SKIP, otherwise traverses all inter-frame forecast modes；
D) different quantization parameter QP is distributed to improve code efficiency to different zones block:
In cataloged procedure, the coding unit of texture complex region, i.e. Coding Units, the usual very little of CU, simple flat
Biggish CU is generally used in region, therefore, is selected suitable QP for CU, be will have a direct impact on coding quality and bit number；Human eye is special
These regions of concern are exactly salient region, therefore, the perceived quality of coding are improved by adjusting the QP of marking area CU；
Based on significant information, marking area gives lesser QP, and biggish QP, QP here are distributed in non-significant region0Refer in platform most
First quantization parameter.
Referring to Fig. 5, above-mentioned steps 4) in improve the method for original View Synthesis prediction algorithm and be:
A) upper left of PU block and the depth value of bottom right, upper right and lower-left are relatively entirely encoded: in original coding platform, being write from memory
Recognizing with flag bit vspSize indicates division size, when upper left depth is greater than bottom right and upper right depth is greater than lower-left,
VspSize is 1, otherwise is 0；If vspSize=1, it is divided into two identical 8 × 4 pieces for all 8 × 8 pieces；If
VspSize=0 is then divided into two identical 4 × 8 pieces for all 8 × 8 pieces；
B) in order to avoid the blocking artifact on boundary, comparing each 8 × 4 or 4 × 8 pieces of four angles: upper left, the right side in original platform
Under, upper right, the depth value size of lower-left be changed to original four Angle Positions inside contracting a pixel respectively, i.e., horizontal, ordinate difference
A pixel unit is translated inwards.
The coding parameter of the present embodiment experiment cycle tests is as shown in table 1, is configured in experiment using 3view+3depth
File, input is left-in-textures of right three viewpoints and corresponding depth map carry out encoded test.
Table 1
| Sequence names | Resolution ratio | Encode frame number | Coded views sequence | QP0Texture | QP0Depth |
| Kendo | 1024×768 | 9 | 3-1-5 | 25、30、35、40 | 30、35、40、45 |
| Balloons | 1024×768 | 9 | 3-1-5 | 25、30、35、40 | 30、35、40、45 |
| Newspaper | 1024×768 | 9 | 4-2-6 | 25、30、35、40 | 30、35、40、45 |
| Lovebird | 1024×768 | 9 | 6-4-8 | 25、30、35、40 | 30、35、40、45 |
| UndoDancer | 1920×1088 | 9 | 5-1-9 | 25、30、35、40 | 30、35、40、45 |
| PoznanStreet | 1920×1088 | 9 | 4-5-3 | 25、30、35、40 | 30、35、40、45 |
| Shark | 1920×1088 | 9 | 5-1-9 | 25、30、35、40 | 30、35、40、45 |
| GTFly | 1920×1088 | 9 | 5-1-9 | 25、30、35、40 | 30、35、40、45 |
The present embodiment finally obtained notable figure of cycle tests in establishing 3D conspicuousness model process is as shown in Figure 3.It compiles
The comparison of the experimental data of innovatory algorithm and primary platform is as shown in table 2 during code.
Table 2
| Sequence names | Video0 | Video1 | Video2 |
| Kendo | 0.00% | - 0.8% | - 0.6% |
| Balloons | 0.00% | - 0.8% | - 1.3% |
| Newspaper | 0.00% | - 1.2% | - 0.7% |
| Lovebird | 0.00% | - 0.6% | - 1.8% |
| UndoDancer | 0.00% | - 1.7% | - 1.6% |
| PoznanStreet | 0.00% | - 1.8% | - 1.7% |
| Shark | 0.00% | - 3.8% | - 3.6% |
| GTFly | 0.00% | - 2.2% | - 2.3% |
As can be seen that the present invention in an encoding process can reasonable contemplation human-eye visual characteristic factor, guarantee objective matter
Amount saves code rate and improves code efficiency.
In summary as it can be seen that the present invention considers human-eye visual characteristic comprehensively, the difference of the encoding block according to human eye concern is aobvious
The quantization parameter and coding mode of work degree adjustment coding, while boundary blocking artifact when View Synthesis is predicted is avoided, guarantee coding
Quality improves code efficiency.
Claims (4)
1. a kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis, specific steps include:
1) 3D conspicuousness model is established, conspicuousness information is obtained: on the basis of 2D conspicuousness model, by increasing video image
Depth information, establish have depth perception information 3D conspicuousness model, obtain the 3D conspicuousness information of encoded content；
2) the assigned area of present encoding block is determined:, will be in present encoding block according to the 3D conspicuousness information of encoded video
Appearance is divided into salient region, non-limiting region or intermediate region；
3) coding strategy in different demarcation region is determined: after carrying out region division to present encoding block, for different significances
Different Q P value is distributed, and sets different coding modes；
4) it improves original View Synthesis and predicts VSP algorithm: for the blocking artifact for avoiding boundary, by four of original VSP neutron PU block
Angle inside contracts a pixel respectively, i.e., the X at sub- four angles of PU block, Y-axis translate a pixel respectively inwards, then take newly-generated four
The maximum depth value at a angle is stored；
5) multi-view video sequences are encoded: utilizing 3D conspicuousness information and improved VSP algorithm, carries out multi-view video sequences
Coding；
In above-mentioned steps 1) in, it first is normalized the depth map D of original series to obtain depth weights omega, due to depth
Pixel value is 0-255, bigger according to nearlyr its conspicuousness of adjustment of object distance, thereforeFollowed by 2D conspicuousness
The notable figure S of model acquisition texture videoC, texture notable figure and depth multiplied by weight are obtained into 3D notable figure SDC, i.e. SDC=ω *
SC；SDCDepth information is increased on the basis of 2D texture notable figure, meets human eye vision attention mechanism, has reached closer
The higher purpose of its significance of object；
In above-mentioned steps 3) in, coding dependency using present encoding block and contiguous block time and spatially, to current block into
Row terminates skip in advance, i.e. SKIP mode treatment:
A) if present encoding block is in non-limiting region, if in five adjacent blocks there are three or three SKIP conducts selected above
Optimal mode, then it is assumed that the optimum prediction mode of current block is SKIP, and SKIP and Inter2N × 2N is otherwise selected to wait as interframe
Lectotype；
B) when present encoding block is in salient region, if five adjacent blocks all select SKIP mode, then it is assumed that current block selection
Otherwise SKIP traverses all inter-frame modes as optimum prediction mode to current block；
C) when present encoding block is in intermediate region, if there are four or four or more adjacent blocks select SKIP, then it is assumed that current block is most
Good mode is SKIP, otherwise traverses all inter-frame forecast modes；
D) different quantization parameter QP is distributed to improve code efficiency to different zones block:
Based on significant information, marking area gives lesser QP, and biggish QP, QP here are distributed in non-significant region0Refer in platform
Initial quantization parameter.
2. the 3D-HEVC coding method according to claim 1 predicted based on 3D conspicuousness information and View Synthesis, special
Sign is that the step 2) includes the following steps:
A) the 3D notable figure S for first obtaining step 1)DCIt carries out binary conversion treatment and obtains Sb, i.e.,
Wherein, T is adaptive threshold, is also changed correspondingly according to the value of the different T of video content, significant after such binaryzation
Scheme SbPossess and be more clear clearly demarcated profile, the highlighted place that pixel value is 1 in image is the area-of-interest of human eye, and pixel is
0 is background area；
B) the significant mean value S of present encoding block is calculatedAve=ST/ (H*W), wherein STIndicate that binaryzation is significant in present encoding block
The pixel value summation of figure, H, W respectively represent the height and width of present encoding block, SAveAs present encoding block is average significant
Degree carries out region division to current block by calculating the average significance of current block:
For intermediate region, and it is divided into two kinds of situations:
In this way according to average significance, present encoding block four kinds of different types of regions have been divided into just.
3. the 3D-HEVC coding method according to claim 1 predicted based on 3D conspicuousness information and View Synthesis, special
Sign is, in above-mentioned steps 4) in:
A) relatively entirely encode the upper left of PU block and the depth value of bottom right, upper right and lower-left: in original coding platform, default is used
Flag bit vspSize indicates division size, and when upper left depth is greater than bottom right and upper right depth is greater than lower-left, vspSize is
1, on the contrary it is 0；If vspSize=1, it is divided into two identical 8 × 4 pieces for all 8 × 8 pieces；If vspSize=0,
It is divided into two identical 4 × 8 pieces for all 8 × 8 pieces；
B) in order to avoid the blocking artifact on boundary, corresponding to each 8 × 4 or 4 × 8 pieces of four angles in original platform: upper left, bottom right,
Upper right, the depth value size of lower-left are changed to original four Angle Positions inside contracting a pixel respectively, i.e., horizontal, ordinate difference is inwards
Translate a pixel unit.
4. the 3D-HEVC coding method according to claim 1 predicted based on 3D conspicuousness information and View Synthesis, special
Sign is, in above-mentioned steps 5) in, using 3D-HEVC encoded test platform, 3view+3depth configuration file is used in experiment,
Input is left-in-textures of right three viewpoints and corresponding depth map carry out encoded test；It is carried out in experiment using 8 video sequences
Test, different sequences have different coding parameters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610889330.XA CN106507116B (en) | 2016-10-12 | 2016-10-12 | A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610889330.XA CN106507116B (en) | 2016-10-12 | 2016-10-12 | A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106507116A CN106507116A (en) | 2017-03-15 |
| CN106507116B true CN106507116B (en) | 2019-08-06 |
Family
ID=58294802
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610889330.XA Active CN106507116B (en) | 2016-10-12 | 2016-10-12 | A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106507116B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109151479A (en) * | 2018-08-29 | 2019-01-04 | 南京邮电大学 | Significance extracting method based on H.264 compression domain model with feature when sky |
| CN110213566B (en) * | 2019-05-20 | 2021-06-01 | 歌尔光学科技有限公司 | Image matching method, device, equipment and computer readable storage medium |
| CN110460832B (en) * | 2019-07-31 | 2021-09-07 | 南方医科大学南方医院 | Processing method, system and storage medium of double-viewpoint video |
| WO2021248349A1 (en) * | 2020-06-10 | 2021-12-16 | Plantronics, Inc. | Combining high-quality foreground with enhanced low-quality background |
| CN113194312B (en) * | 2021-04-27 | 2021-12-07 | 中国科学院国家空间科学中心 | Planetary science exploration image adaptive quantization coding system combined with visual saliency |
| CN113709464A (en) * | 2021-09-01 | 2021-11-26 | 展讯通信(天津)有限公司 | Video coding method and related device |
| CN114979652A (en) * | 2022-05-20 | 2022-08-30 | 北京字节跳动网络技术有限公司 | Video processing method and device, electronic equipment and storage medium |
| CN115019151B (en) * | 2022-08-05 | 2022-10-21 | 成都图影视讯科技有限公司 | Non-salient feature region accelerated neural network architecture, method and apparatus |
| CN115866251B (en) * | 2023-02-22 | 2023-06-02 | 浙江鼎立实业有限公司 | Image information rapid transmission method based on semantic segmentation |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103873876A (en) * | 2014-03-17 | 2014-06-18 | 天津大学 | Conspicuousness-based multi-viewpoint color plus depth video coding method |
| CN105230022A (en) * | 2013-05-31 | 2016-01-06 | 高通股份有限公司 | Use based on the disparity vector of neighbor derive for 3D video coding and derivation disparity vector of passing through |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013173282A1 (en) * | 2012-05-17 | 2013-11-21 | The Regents Of The University Of Califorina | Video disparity estimate space-time refinement method and codec |
| WO2015062002A1 (en) * | 2013-10-31 | 2015-05-07 | Mediatek Singapore Pte. Ltd. | Methods for sub-pu level prediction |
-
2016
- 2016-10-12 CN CN201610889330.XA patent/CN106507116B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105230022A (en) * | 2013-05-31 | 2016-01-06 | 高通股份有限公司 | Use based on the disparity vector of neighbor derive for 3D video coding and derivation disparity vector of passing through |
| CN103873876A (en) * | 2014-03-17 | 2014-06-18 | 天津大学 | Conspicuousness-based multi-viewpoint color plus depth video coding method |
Non-Patent Citations (3)
| Title |
|---|
| A Novel Saliency Model for Stereoscopic Images;Hao Cheng et.al;《2015 International Conference on Digital Image Computing》;20151125;全文 |
| Backward view synthesis prediction using virtual depth map for multiview video plus depth map coding;Shinya Shimizu et.al;《2013 Visual Communications and Image Processing》;20131120;全文 |
| 基于边缘的面向虚拟视绘制的深度编码算法;刘超等;《上海大学学报》;20140831;全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106507116A (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106507116B (en) | A kind of 3D-HEVC coding method predicted based on 3D conspicuousness information and View Synthesis | |
| CN106105191B (en) | Method and apparatus for handling multiview video signal | |
| US9521413B2 (en) | Optimized filter selection for reference picture processing | |
| CN101710993B (en) | Block-based self-adaptive super-resolution video processing method and system | |
| CN102648631B (en) | For the method and apparatus of coding/decoding high-definition picture | |
| CN104378643B (en) | A kind of 3D video depths image method for choosing frame inner forecast mode and system | |
| CN105430415A (en) | Fast intraframe coding method of 3D-HEVC depth videos | |
| CN108781284A (en) | The method and device of coding and decoding video with affine motion compensation | |
| CN105120290B (en) | A kind of deep video fast encoding method | |
| CN110933426B (en) | Decoding and encoding method and device thereof | |
| CN104798375A (en) | Method and apparatus of constrained disparity vector derivation in 3d video coding | |
| CN102801996B (en) | Rapid depth map coding mode selection method based on JNDD (Just Noticeable Depth Difference) model | |
| CN105812797B (en) | A kind of coding unit selection method and device | |
| CN105208387A (en) | HEVC intra-frame prediction mode fast selection method | |
| CN103024381B (en) | A kind of macro block mode fast selecting method based on proper discernable distortion | |
| CN103338370A (en) | Multi-view depth video fast coding method | |
| CN109587503A (en) | A kind of 3D-HEVC depth map intra-frame encoding mode high-speed decision method based on edge detection | |
| CN104539970A (en) | 3D-HEVC interframe coding merge mode fast decision making method | |
| CN109756719A (en) | The 3D-HEVC interframe fast method of Bayesian decision is divided based on CU | |
| CN103391439B (en) | A kind of H.264/AVC bit rate control method hidden based on active macro block | |
| CN101557519B (en) | Multi-view video coding method | |
| CN101783956A (en) | Back-prediction forecast method based on spatio-temporal neighbor information | |
| CN106878754A (en) | A kind of 3D video depths image method for choosing frame inner forecast mode | |
| CN104618725A (en) | Multi-view video coding algorithm combining quick search and mode optimization | |
| CN105704497B (en) | Coding unit size fast selection algorithm towards 3D-HEVC |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |