CN102868888B - Dynamic slice control method oriented to parallel video encoding - Google Patents
Dynamic slice control method oriented to parallel video encoding Download PDFInfo
- Publication number
- CN102868888B CN102868888B CN201210351083.XA CN201210351083A CN102868888B CN 102868888 B CN102868888 B CN 102868888B CN 201210351083 A CN201210351083 A CN 201210351083A CN 102868888 B CN102868888 B CN 102868888B
- Authority
- CN
- China
- Prior art keywords
- slice
- frame
- state
- time
- present frame
- 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.)
- Expired - Fee Related
Links
Abstract
The invention provides a dynamic slice control method oriented to parallel video encoding. According to the method, during an encoding process, a Markov chain model is constructed according to a slice concurrent state of an encoded frame to predict a slice concurrent state of the current to-be-encoded frame, so as to determine the number of slices of the current to-be-encoded frame; and then, the computation complexity of the current frame is predicted according to weighted encoding time of the encoded frame in a same scene, allocation units with corresponding number are allocated to each slice so as to enable the encoding time of each slice to be approximately equal, and the computation complexity among the slices is balanced. The experiment result shows that the method provided by the invention improves the encoding speed by 30% in comparison with the conventional H.264-based fixed slice parallel encoding method, and has approximate video compression efficiency.
Description
Technical field
The present invention relates to computer video coding field, be specifically related to a kind of dynamic slice control method towards parallel video coding.
Background technology
Polycaryon processor is prevailing computer market, makes the personal computer that has 4 core CPU become very general.This also make as video conference to be applied in real time parallelization on personal computer more feasible.Parallel video coding method roughly can be divided into 3 classes: the parallel encoding method based on frame, the parallel encoding method based on sheet (slice) and parallel encoding method (the reference paper 1:C.MeenDerinck based on macro block, A.Azevedo, M.Alvarez, et.al., " Parallel Scalability of H.264; " Proc.of 1th Workshop on Programmability Issues for Multi-Core Computers, 2008).Parallel encoding method based on frame is usually used in off-line type Video coding, and is not suitable for the real-time video such as Online Video meeting coding because in this scene, when present frame is encoded its subsequent frame toward contact obtain less than.Parallel encoding method based on macro block is utilized the concurrency of macro block in same frame, can be used for real-time video coding, but the parallel granularity of parallel method based on macro block is very little, therefore larger for synchronous expense.Parallel encoding method based on slice is the compromise of above two kinds, and synchronization overhead is little, and it also can be applicable to real-time video application simultaneously, is the method generally adopting in current encoder device.
In traditional parallel video coding method based on slice, the slice number of every frame is (often the equaling CPU check figure) who fixes, and the macroblock number that each slice comprises also equates.The parallel method of this fixing slice has two shortcomings.First, in actual coding process, because some cpu resource is taken by other application programs, make some slice of current encoded frame not be assigned to cpu resource, the multiple slice in same frame may not can start coding simultaneously.To cause these slice that are not assigned to cpu resource can not get started coding like this, may need other slice by the time could start after having encoded, thereby increase the stand-by period of other slice of coding side.Secondly, the texture of the diverse location in same two field picture and motion complexity are different.Such as, the video sequence of video conference often has larger motion at image mid portion, and has relatively little motion at top and bottom.Thereby by macro block uniform distribution to slice, can make the scramble time of slice inhomogeneous, this also can increase the mutual stand-by period of slice cross-thread.At present, also nobody proposes the effective solution for first problem, and for Second Problem, existing scholar has proposed some and separated annual reporting law.The people such as F.H.Seitner propose a kind of method, when a thread completes after the coding of its slice, to can not wait for other threads, a but independent slice(reference paper 2:F.H.Seitner of the next frame that starts to encode, M.Bleyer, R.M.Schreier and M.Gelautz, " H.264 Decoding of Evaluation of Date-parallel Splitting Approaches for; " Proc.of the 6th International Conference on Advances in Mobile Computing and Multimedia, 2008).In fact the method has introduced parallel method based on frame and has improved the concurrency of encoder, but and is not suitable for real-time video coding.The people such as N.Zhang (reference paper 3:K.Ba, X.Jin and S.Goto, " H.264/AVC Parallel Encoder of A Dynamic Slice-resize Algorithm for Fast, " Proc.of International Symposium on Intelligent Signal Processing and Communication Systems, pp.1-4, 2010) and people (the reference paper 4:K.Ba such as K.Ba, X.Jin and S.Goto, " H.264/AVC Parallel Encoder of A Dynamic Slice-resize Algorithm for Fast, " Proc.of International Symposium on Intelligent Signal Processing and Communication Systems, pp.1-4, 2010) respectively the encode method of computation complexity of thread of the scramble time balance that proposes a kind of former frame of use.But for the video sequence that has frequent scene to switch, the method is owing to there will be the prediction of mistake, effect is also bad.The people such as B.Jung propose a kind of two stage adaptive slice size selection scheme (reference paper 5:B.Jung and B.Jeon, " Adaptive Slice-level Parallelism for is Encoding Using Pre Macroblock Mode Selection H.264/AVC, " Journal of Visual Communication and Image Representation, pp.558-572, 2008), the method is at pretreatment stage, first estimate the computation complexity of each macro block, then just enter real coding stage, for each slice encodes.This method often can this frame of Accurate Prediction computation complexity, but the pretreatment stage of the method has also increased extra coding expense.
Summary of the invention
The object of the invention is two shortcomings of the parallel encoding method in order to solve above-mentioned existing fixing slice.The present invention proposes a kind of dynamically slice control method, is intended to improve the performance of the parallel video coding method based on slice.
A kind of dynamic slice control method towards parallel video coding that the present invention proposes, specifically comprises the steps:
Step 1: utilize by the Markov-chain model that the concurrent state of slice of coded frame builds and predict the concurrent state of slice of current frame to be encoded, thereby determine the number of slice in frame to be encoded.
If the state of encoder has two kinds: state A represents that all coding threads have all been assigned to CPU simultaneously, state B represents to have at least a coding thread unallocated to CPU in the time that coding starts; The state of encoder has formed Markov Chain, and encoder is at next time point t
i+1state be only subject to current time t
iimpact.The number S of the slice of present frame determines according to following formula:
M represents the maximum available check figure of current C PU; E (A) and E (B) represent respectively the present frame expectation parallel encoding time in the time that former frame is state A and state B respectively, determine according to following formula:
E(X)＝P
ZA·T(X|A)+P
ZB·T(X|b)
P
xYrepresentative is the transition probability to state Y by state X, and the initial condition of establishing before coding the first frame is state A, initial transition probability P
aBand P
aAbe 0.5; T (X|Y) represents that the virtual condition of present frame is Y, and the parallel encoding time of this frame while supposing that its state is X, X, Y ∈ { A, B}.
Step 2: S the slice obtaining for step 1 distributes allocation units.Specifically: whether have scene switch, if having, quantity and the weighting sad value SAD of coded frame are set if first detecting present frame
wbe 0, and allocation units be averagely allocated to each slice of present frame; If no, calculate the balance forecast scramble time t of the each slice of present frame
p(s), then distribute the allocation units of respective number to each slice according to time complexity, make the scramble time of each slice be approximately equal to t
p(s).
Described balance forecast scramble time t
p(s) determine according to following formula:
K represents that present frame is K frame, and R represents the number of allocation units, t
p(f
k[j]) represent predictive coding time of j allocation units in present frame, t
p(f
k[j])=t
w(j), j ∈ [1, R], t
w(j) be present frame K-1 before the weighted average of the scramble time of j allocation units of coded frame, if present frame has scene to switch or present frame is the 1st frame, t
w(j) value is set to 0.
Step 3: coding present frame, then upgrades transition probability P
xYwith weighted average t
w(j), finish the coding of present frame, utilize the relevant parameter upgrading to start to carry out the coding of next frame.
Advantage and the good effect of the inventive method are: the present invention dynamically determines the number of slice in cataloged procedure, and balance the computation complexity between slice, improved code efficiency.When machine is busy, compared with the method for the inventive method and fixing slice, coding rate can on average promote 30%, and simultaneity factor distortion performance does not almost change.
Brief description of the drawings
Fig. 1 is the overall flow chart of steps of the dynamic slice control method of the present invention;
Fig. 2 is coder state transition diagram of the present invention;
Fig. 3 is the scramble time figure of each allocation units of 8 successive frames in Same Scene in the present invention;
Fig. 4 is the SAD(Sum of Absolute Difference of the allocation units of 8 successive frames in Same Scene in the present invention, absolute residuals and) value figure;
Fig. 5 is speed-up ratio and CPU check figure relation curve comparison diagram in the embodiment of the present invention;
Fig. 6 is the rate distortion curve comparison figure in the embodiment of the present invention.
Embodiment
Below in conjunction with drawings and Examples, technical scheme of the present invention is described in further detail.
The present invention proposes a kind of dynamic slice control method towards parallel video coding, in the lower encoder of two kinds of different conditions (idle/busy) desired value of required scramble time, determines the slice quantity of present frame by computing equipment.The method is based on Markov-chain model.The present invention is also by predicting that the scramble time of each allocation units in frame to be encoded is for the allocation units of each slice distribution respective number, thus the encoder complexity of each slice in the same frame of balance.Experimental result shows, the inventive method is compared with the traditional fixing slice parallel encoding method based on H.264, and coding rate has improved 30%, and has close video compression efficiency.
A kind of dynamic slice control method towards parallel video coding that the present invention proposes, overall step as shown in Figure 1, is specifically described each step below.
Step 1: utilize by the Markov-chain model that the concurrent state of slice of coded frame builds and predict the concurrent state of slice of current frame to be encoded, thereby determine the number of slice in frame to be encoded.
If the state of encoder has two kinds: state A represents that all coding threads have all been assigned to CPU simultaneously, state B represents to have at least a coding thread unallocated to CPU in the time that coding starts.
The beginning scramble time of each frame in setting video sequence is a discrete time point (t
0, t
1..., t
n-1), wherein N represents the frame number of having encoded.At t
iin the moment, the state of encoder will be the one in state A and state B.Meanwhile, suppose that encoder is at next time point t
i+1state be only subject to current time t
iimpact, and be independent of the residing state of time point before any.The state of encoder has formed Markov Chain, and its state transitions process as shown in Figure 2.Wherein P
xYrepresentative is the transition probability to state Y by state X, X, Y ∈ { A, B}.
Because the encoding state of former frame was known before coding present frame, as long as thereby know state transition probability this moment, just present frame is positioned at each shape probability of state and can knows so.The inventive method will be calculated the frequency of every kind of state transitions generation, and using them as actual transition probability, such as P
xYbe set to encoder in cataloged procedure and transfer to the frequency of state Y from state X.Because the occurrence frequency of a state transitions can change in cataloged procedure, therefore P
xYvalue also can along with coding carrying out and dynamically upgrade, after the complete frame of every coding, the state of adding up all coded frame upgrades P
xY.
Make T (X|Y) for the virtual condition of present frame be Y, and parallel encoding time of this frame while supposing that its state is X.If the maximum available check figure of current C PU is M, present frame can be divided at most M slice, and the non-parallel time of coding present frame is T
f, each slice of present frame needed total time of having encoded is successively T
f.In the methods of the invention, suppose to only have at most a slice coding thread to distribute less than CPU at every turn.Ideally, the scramble time of each slice is identical, due to the inventive method balance the computation complexity between slice, so this point can obtain general guarantee.T (X|Y), X, Y ∈ A, B can be calculated by following formula:
T(A|A)＝T
f/M （1）
T(A|B)＝2·T
f/M （2）
T(B|B)＝T(B|A)＝T
f/(M-1) （3）
The coding thread of the corresponding slice of formula (2) is not assigned to CPU, thereby the coding thread of any one other slice by the time of having to could start after finishing.
The state of the former frame of supposing is Z, and { A, B} make E (X) for the expectation parallel encoding time of frame to be encoded in the time of state X to Z ∈.E (X) can be calculated by following formula:
E(X)＝P
ZA·T(X|A)+P
ZB·T(X|B) （4）
The desired value of scramble time will can be obtained in above-mentioned formula (1), (2), (3) substitution (4).The number S of slice can be decided by following formula:
As shown in Figure 1, in step 1, first determine frame to be encoded expectation parallel encoding time E (A) in the time of state A and state B and E (B) respectively according to formula above, then determine the number S of slice according to formula (5).The initial condition of encoding in the inventive method before the first frame is A, initial condition transition probability P
aBand P
aAbe 0.5.
Step 2: according to the number of slice definite in step 1, predict the computation complexity of present frame according to the weighted coding time of the coded frame in Same Scene, the allocation units of respective number are distributed to each slice, make the scramble time approximately equal of each slice.
After having determined the number of slice, start to consider the encoder complexity of balance slice, the step 2 in this part corresponding diagram 1.For convenience of description, the present invention introduces the concept of allocation units, and allocation units are minimum code unit of distributing to each slice, and its size can be a macro block or a macro-block line or plural multiple macro-block line.In the present invention, allocation units are made as to a macro-block line.Fig. 3 has shown the scramble time of each allocation units of 8 successive frames in Same Scene, and wherein, abscissa represents allocation units number, and ordinate is the corresponding scramble time.Can see, the scramble time of the allocation units that in Same Scene, time domain is adjacent is similar.The present invention utilizes this point just, predicts the scramble time of each allocation units in frame to be encoded.
Suppose the individual coded frame f of K-1 in Same Scene
1, f
2..., f
k-1, and a frame f to be encoded
k, every frame has R allocation units.Make t (f
i[j]) be the actual coding time of j allocation units of i frame, wherein i ∈ [1, K-1], j ∈ [1, R].For the complexity between balance slice, whether occurrence scene switches first to survey present frame, and then determines the border of each slice.Scene switches that to refer to that texture information that present frame comprises and former frame have very large different.
Switch for the scene how detecting efficiently in video sequence, have at present scholar to carry out extensive work.The inventive method detects scene by a kind of simple method based on SAD and switches.Show methodical research and show, the sad value of the two continuous frames in Same Scene is close with the sad value of other any two continuous frames.Fig. 4 has shown the sad value of all allocation units, and wherein, abscissa represents allocation units number, and ordinate is the sad value between the allocation units of adjacent two frames.Sad value SAD (i) between i frame and i-1 frame is defined by formula (6):
SAD(i)＝∑|p
i(r)-p
i-1(r)| （6）
Wherein, p
i(r) pixel value of the pixel that expression i frame raster order is r.
Then, a weighting sad value SAD of definition
w, this value is initially 0, in the step 3 of having encoded after a frame, it is upgraded.
If SAD (i) is greater than α SAD
w, think that it is different scenes that present frame is compared previous frame, detects scene switching.Here α is a predefined coefficient, and it is best that experimental result shows that this value is made as 1.3 o'clock effects, and therefore in the embodiment of the present invention, α is set to 1.3.If scene switching detected at present frame, remove the information of the frame that has completed coding, and encoded frame number (K-1) and weighting sad value SAD are set
wbe 0, and be the slice mean allocation allocation units of present frame.If scene switching do not detected at present frame, come to distribute allocation units for the slice of this frame with following methods.
Make t
p(f
k[j]) be the predictive coding time of j allocation units in current K frame.T
p(f
k[j]) value obtained by following formula:
t
p(f
K[j])＝t
w(j)，j∈[1，R] （7）
Wherein t
w(j) be present frame K-1 before the weighted average of the scramble time of j allocation units of coded frame.If present frame comprises a new scene or present frame is the 1st frame, the quantity of coded frame is 0, now t
w(j) value is set to 0.T
w(j) in the step 3 of value after the current K frame of having encoded, upgrade, be shown below:
t
w(j)＝0.5*t
w(j)+0.5*t(f
K[j]) （8）
The balance forecast scramble time t of the each slice of present frame
p(s) calculated by following formula:
Then, distribute the allocation units of respective number to each slice according to time complexity, make the scramble time of each slice be approximately equal to t
p(s).
Step 3: coding present frame, the relevant parameter using in step of updating one and step 2, the coding of the present frame that is so far through with, can utilize the relevant parameter of renewal to start to carry out the coding of next frame.Repeated execution of steps one to three, can complete the coding of whole video.
Table 1 is shown the comparative result of several method coding rate, and speed was represented by the average scramble time of every frame.Except the method (Proposed) of the present invention's proposition and the scramble time of fixing slice method (F.S.), the non-parallel encoding time (N.P.) also provides in table.Can see, the method that the present invention proposes on average can promote coding rate 30% with respect to fixing slice method.
The comparison of table 1 scramble time
Fig. 5 has shown the speed-up ratio of method of the present invention and reference method, and wherein, abscissa is CPU check figure, and ordinate is speed-up ratio (Speed-ip ratio).Dotted line Ideal in figure represents desirable speed-up ratio, and the number of the slice of each frame equals the check figure of CPU.F.S. ' solid line representing is the speed-up ratio obtaining in the time standby F.S. method of test machine free time.The solid line that the solid line that Proposed represents and F.S. represent is respectively method and the speed-up ratio of F.S. method in the situation that test machine is busy that the present invention proposes.As can be seen from the figure,, when machine is busy, method of the present invention is compared F.S. method higher speed-up ratio.
As shown in Figure 6, also test the distortion performance of the inventive method, wherein, abscissa presentation code code check (Biterate, the kbps of unit), ordinate represents brightness peak signal to noise ratio (Peak Signal-to-Noise Ratio, PSNR), has provided the rate distortion curve obtaining of encoding to two video sequence Vidyo1 and ShuttleStart in Fig. 6.Solid line is illustrated in test and encodes when the machine free time, and dotted line is illustrated in test machine and encodes busy time.Test result as shown in Figure 6, has proved that method of the present invention compares F.S. method and only have very small distortion performance difference.
Claims (6)
1. towards a dynamic slice control method for parallel video coding, it is characterized in that, the method specifically comprises the steps:
Step 1: utilize by the Markov chain model that the concurrent state of slice of coded frame builds, predict the concurrent state of slice of present frame, thereby determine the slice number S of present frame; Slice represents sheet;
If the state of encoder has two kinds: state A represents that all coding threads have all been assigned to CPU simultaneously, state B represents to have at least a coding thread unallocated to CPU in the time that coding starts; The state of encoder has formed Markov Chain, and encoder is at next time point t
i+1state be only subject to current time t
iimpact; The number S of the slice of present frame determines according to following formula:
M represents the maximum available check figure of current C PU; E (A) and E (B) represent respectively the present frame expectation parallel encoding time in the time that former frame is state A and state B respectively, determine according to following formula:
E(X)＝P
ZA·T(X|A)+P
ZB·T(X|B)
If the state of former frame is Z, Z ∈ { A, B}; P
xYrepresentative is the transition probability to state Y by state X, and the initial condition of establishing before coding the first frame is state A, initial transition probability P
aBand P
aAbe 0.5; T (X|Y) represents that the virtual condition of present frame is Y, and the parallel encoding time of this frame while supposing that its state is X, X, Y ∈ { A, B};
Step 2: S the slice obtaining for step 1 distributes allocation units, specifically: whether have scene switch, if having, quantity and the weighting sad value SAD of coded frame are set if first detecting present frame
wbe 0, and allocation units be averagely allocated to each slice of present frame; If no, calculate the balance forecast scramble time t of the each slice of present frame
p(s), then distribute the allocation units of respective number to each slice according to time complexity, make the scramble time of each slice be approximately equal to t
p(s); Described balance forecast scramble time t
p(s) determine according to following formula:
K represents that present frame is K frame, and R represents the number of allocation units, t
p(f
k[j]) represent j allocation units f in present frame
kthe predictive coding time of [j], t
p(f
k[j])=t
w(j), j ∈ [1, R], t
w(j) be present frame K-1 before the weighted average of the scramble time of j allocation units of coded frame, if present frame has scene to switch or present frame is the 1st frame, t
w(j) value is set to 0;
Step 3: coding present frame, then upgrades transition probability P
xYwith weighted average t
w(j), finish the coding of present frame.
2. a kind of dynamic slice control method towards parallel video coding according to claim 1, is characterized in that, the T described in step 1 (X|Y) determines according to following formula:
T(A|A)＝T
f/M
T(A|B)＝2·T
f/M
T(B|B)＝T(B|A)＝T
f/(M-1)
T
fthe non-parallel time of presentation code present frame; The coding thread of the corresponding slice of T (A|B) is not assigned to the situation of CPU.
3. a kind of dynamic slice control method towards parallel video coding according to claim 1, it is characterized in that, allocation units described in step 2 are minimum code unit of distributing to each slice, and size is a macro block or a macro-block line or plural macro-block line.
4. a kind of dynamic slice control method towards parallel video coding according to claim 1, it is characterized in that, whether the detection present frame described in step 2 has scene to switch, to adopt the method based on SAD to detect, specifically: determine sad value SAD (the i)=∑ between i frame and i-1 frame | p
i(r)-p
i-1(r) |, p
i(r) value of the pixel that the raster order of expression i frame is r, p
i-1(r) value of the pixel that the raster order of expression i-1 frame is r; SAD represent absolute residuals and; Judge whether SAD (i) is greater than α SAD
w, if so, there is scene to switch, otherwise, do not have scene to switch, SAD
wrepresent weighting sad value, SAD
winitial value be that 0, α is the coefficient setting in advance.
5. a kind of dynamic slice control method towards parallel video coding according to claim 4, is characterized in that, described factor alpha is set to 1.3.
6. a kind of dynamic slice control method towards parallel video coding according to claim 1, is characterized in that the weighted average t in described step 3
w(j) upgrade according to following formula:
t
w(j)＝0.5*t
w(j)+0.5*t(f
K[j])
Wherein, t (f
k[j]) be the actual coding time of j allocation units of present frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210351083.XA CN102868888B (en) | 2012-04-27 | 2012-09-19 | Dynamic slice control method oriented to parallel video encoding |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210128721.1 | 2012-04-27 | ||
| CN201210128721 | 2012-04-27 | ||
| CN201210351083.XA CN102868888B (en) | 2012-04-27 | 2012-09-19 | Dynamic slice control method oriented to parallel video encoding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102868888A CN102868888A (en) | 2013-01-09 |
| CN102868888B true CN102868888B (en) | 2014-11-26 |
Family
ID=47447456
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210351083.XA Expired - Fee Related CN102868888B (en) | 2012-04-27 | 2012-09-19 | Dynamic slice control method oriented to parallel video encoding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102868888B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106954066A (en) * | 2016-01-07 | 2017-07-14 | 鸿富锦精密工业（深圳）有限公司 | Video encoding/decoding method |
| CN107071424B (en) * | 2017-03-17 | 2018-09-25 | 山东科技大学 | A kind of load-balancing method based on scramble time prediction model |
| CN108600756B (en) * | 2018-05-04 | 2019-10-29 | 上海大学 | Coding parameter self-adapting regulation method based on Monte Carlo method Markov chain |
| CN110933457B (en) * | 2019-12-02 | 2022-01-11 | 杭州当虹科技股份有限公司 | Multi-node low-delay parallel coding method for 8K ultra-high definition |
| CN113747154A (en) * | 2021-08-27 | 2021-12-03 | 杭州当虹科技股份有限公司 | Code rate control method and encoder |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101969560A (en) * | 2010-11-01 | 2011-02-09 | 北京中科大洋科技发展股份有限公司 | Slice code rate allocation method of Mpeg2 high-definition coder under multi-core platform |
-
2012
- 2012-09-19 CN CN201210351083.XA patent/CN102868888B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101969560A (en) * | 2010-11-01 | 2011-02-09 | 北京中科大洋科技发展股份有限公司 | Slice code rate allocation method of Mpeg2 high-definition coder under multi-core platform |
Non-Patent Citations (4)
| Title |
|---|
| A Dynamic Slice-resize Algorithm for Fast H.264/AVC Parallel Encoder;K.Ba 等;《Proc.of International Symposium on Intelligent Signal Processing and Communication Systems》;20101231;1-4 * |
| K.Ba 等.A Dynamic Slice-resize Algorithm for Fast H.264/AVC Parallel Encoder.《Proc.of International Symposium on Intelligent Signal Processing and Communication Systems》.2010,1-4. * |
| 基于Slice的H.264并行视频编码算法;宁华 等;《计算机工程》;20050228;第31卷(第4期);181-182,211 * |
| 宁华 等.基于Slice的H.264并行视频编码算法.《计算机工程》.2005,第31卷(第4期),181-182,211. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102868888A (en) | 2013-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102868888B (en) | Dynamic slice control method oriented to parallel video encoding | |
| CN101461248B (en) | Method and apparatus for adaptively determining a bit budget for encoding video pictures | |
| CN101107863B (en) | Method and apparatus for real time parallel encoding | |
| KR101644208B1 (en) | Video encoding using previously calculated motion information | |
| CN105491377B (en) | A kind of video decoded macroblock grade Method of Scheduling Parallel of computation complexity perception | |
| CN102835111B (en) | The motion vector of previous block is used as the motion vector of current block, image to be carried out to the method and apparatus of coding/decoding | |
| Su et al. | Complexity-constrained H. 264 video encoding | |
| US8934544B1 (en) | Efficient motion estimation in hierarchical structure | |
| US20170085892A1 (en) | Visual perception characteristics-combining hierarchical video coding method | |
| US20110200107A1 (en) | Apparatus and method for motion estimation and image processing apparatus | |
| CN103517069A (en) | HEVC intra-frame prediction quick mode selection method based on texture analysis | |
| CN104067619A (en) | Video decoder, video encoder, video decoding method, and video encoding method | |
| CN102625108B (en) | Multi-core-processor-based H.264 decoding method | |
| CN102857746B (en) | Loop filtering decoding method and device | |
| CN103731668A (en) | Image encoding apparatus and image encoding method | |
| CN104885455A (en) | Content adaptive bitrate and quality control by using frame hierarchy sensitive quantization for high efficiency next generation video coding | |
| Zhao et al. | A dynamic slice control scheme for slice-parallel video encoding | |
| Ahn et al. | Complexity model based load-balancing algorithm for parallel tools of HEVC | |
| US8358693B2 (en) | Encoding visual data with computation scheduling and allocation | |
| CN105025298B (en) | The method and apparatus that image is encoded/decoded | |
| CN107277519A (en) | The method and electronic equipment of a kind of frame type for judging frame of video | |
| CN104980764A (en) | Parallel coding/decoding method, device and system based on complexity balance | |
| CN104104966A (en) | Video block dividing method and device | |
| CN106878754B (en) | A kind of 3D video depth image method for choosing frame inner forecast mode | |
| CN102946533A (en) | Video coding |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20141126 Termination date: 20160919 |