阅读说明：本技术 用于视频编解码的子预测单元时间运动向量预测 (Sub- predicting unit temporal motion vector prediction for coding and decoding video ) 是由 陈俊嘉 徐志玮 陈庆晔 黄毓文 于 2018-03-30 设计创作，主要内容包括：本发明之一方面提供一种视频编解码方法,其用子预测单元时间运动向量预测模式处理当前预测单元。该方法可以包括：接收包括多个子预测单元的当前预测单元；确定为当前预测单元的空间相邻块的运动向量的初始运动向量；基于初始运动向量,执行搜索流程以在当前预测单元的参考图像序列中搜索主同位图像；以及获得用于当前预测单元的多个子预测单元的主同位图像中的同位运动信息。搜索流程可以包括：开启运动向量缩放操作,以用于搜索参考图像序列的子集,以及关闭运动向量缩放操作,以用于在参考图像序列中搜索其他参考图像。(On the one hand the present invention provides a kind of video coding-decoding method, with sub- predicting unit temporal motion vector prediction mode treatment current prediction unit.This method may include: to receive the current prediction unit including multiple sub- predicting units；It is determined as the initial motion vector of the motion vector of the spatial neighboring blocks of current prediction unit；Based on initial motion vector, search routine is executed to search for main co-sited image in the reference image sequence of current prediction unit；And acquisition is for the same position motion information in the main co-sited image of multiple sub- predicting units of current prediction unit.Search routine may include: opening movement vector zoom operations, with the subset and closing movement vector zoom operations for searching for reference image sequence, for searching for other reference pictures in reference image sequence.)

1. a kind of video coding-decoding method is used for sub- predicting unit temporal motion vector prediction mode treatment current predictive list Member, comprising:

Receive the current prediction unit including multiple sub- predicting units；

It is determined as the initial motion vector of the motion vector of the spatial neighboring blocks of the current prediction unit；

Based on the initial motion vector, it is main same to search in the reference image sequence of predicting unit in this prior to execute search routine Bit image, wherein the search routine include:

Opening movement vector zoom operations, with the subset for searching for the reference image sequence, and

The motion vector zoom operations are closed, for searching for other reference pictures in the reference image sequence；And

Obtain the same position motion information in the main co-sited image for multiple sub- predicting unit of the current prediction unit.

2. video coding-decoding method according to claim 1, which is characterized in that the subset of the reference image sequence include with One in lower:

In this prior in the first reference picture list of predicting unit and the second reference picture list of the current prediction unit The top n in the reference image sequence at least one；Or

Preceding N0 reference picture in first reference picture list of predicting unit in this prior, and predicting unit in this prior Second reference picture list in preceding N1 reference picture；

Wherein first reference picture list and second reference picture list be with reference key descending order, with And

N, N0 and N1 is greater than or equal to 0 integer value.

3. video coding-decoding method according to claim 1, which is characterized in that open the motion vector zoom operations packet It includes:

Motion vector scaling is executed, to scale the initial motion vector to obtain in the subset for detecting the reference image sequence One scaling motion vector.

4. video coding-decoding method according to claim 3, which is characterized in that when motion vector scaling is based on first Between distance and the second time gap, the first time distance be include the current prediction unit present image be detected The reference image sequence subset in an image between time gap, second time gap be include that this is current pre- Survey the time gap between the present image reference picture relevant to the initial motion vector of unit.

5. video coding-decoding method according to claim 1, wherein opening the motion vector zoom operations includes:

When an image in the subset of the reference image sequence is not reference picture relevant to the initial motion vector, hold Row motion vector scaling, obtains one in the subset for detecting the reference image sequence to scale the initial motion vector The scaling motion vector of image.

6. video coding-decoding method according to claim 1, wherein closing the motion vector zoom operations includes:

Using the initial motion vector, for detecting other reference pictures in the reference image sequence.

7. video coding-decoding method according to claim 1, wherein further include:

Determine the search order for being used for the search routine；And

According to the determining search order, the search routine is executed.

8. video coding-decoding method according to claim 1, wherein further include:

The search for transmitting or parsing for searching for the main co-sited image in reference image sequence of predicting unit in this prior is suitable Sequence.

9. video coding-decoding method according to claim 1, wherein further include:

According to search order, the search routine is executed, wherein in the search order, reference relevant to the initial motion vector Image is first searched, and after multiple reference pictures in two reference picture lists of the current prediction unit in an alternating manner and It is searched with reference key incremental order.

10. video coding-decoding method according to claim 1, wherein further include:

Transmit or parse the main co-sited image.

11. a kind of video encoding/decoding apparatus is used for sub- predicting unit temporal motion vector prediction mode treatment current predictive list Member, the device include circuit, are used for:

Receive the current prediction unit including multiple sub- predicting units；

It is determined as the initial motion vector of the motion vector of the spatial neighboring blocks of the current prediction unit；

Based on the initial motion vector, it is main same to search in the reference image sequence of predicting unit in this prior to execute search routine Bit image, wherein the search routine include:

Opening movement vector zoom operations, with the subset for searching for the reference image sequence, and

The motion vector zoom operations are closed, for searching for other reference pictures in the reference image sequence；And

Obtain the same position motion information in the main co-sited image for multiple sub- predicting unit of the current prediction unit.

12. video encoding/decoding apparatus according to claim 11, wherein the subset of the reference image sequence includes in following One:

In this prior in the first reference picture list of predicting unit and the second reference picture list of the current prediction unit The top n in the reference image sequence at least one；Or

Preceding N0 reference picture in first reference picture list of predicting unit in this prior, and predicting unit in this prior Second reference picture list in preceding N1 reference picture；

Wherein first reference picture list and second reference picture list be with reference key descending order, with And

N, N0 and N1 is greater than or equal to 0 integer value.

13. video encoding/decoding apparatus according to claim 11, wherein the circuit is also used to:

Motion vector scaling is executed, to scale the initial motion vector to obtain in the subset for detecting the reference image sequence One scaling motion vector.

14. video encoding/decoding apparatus according to claim 13, wherein the motion vector scaling be based at the first time away from From with the second time gap, the first time distance be include the current prediction unit present image be detected this The time gap between an image in the subset of reference image sequence, the second time gap be include the current prediction unit Present image reference picture relevant to the initial motion vector between time gap.

15. video encoding/decoding apparatus according to claim 11, wherein the circuit is also used to:

When an image in the subset of the reference image sequence is not reference picture relevant to the initial motion vector, hold Row motion vector scaling, obtains one in the subset for detecting the reference image sequence to scale the initial motion vector The scaling motion vector of image.

16. video encoding/decoding apparatus according to claim 11, wherein the circuit is also used to:

Using the initial motion vector, for detecting other reference pictures in the reference image sequence.

17. video encoding/decoding apparatus according to claim 11, wherein the circuit is also used to:

Determine the search order for being used for the search routine；And

According to the determining search order, the search routine is executed.

18. video encoding/decoding apparatus according to claim 11, wherein the circuit is also used to:

The search for transmitting or parsing for searching for the main co-sited image in reference image sequence of predicting unit in this prior is suitable Sequence.

19. video encoding/decoding apparatus according to claim 11, wherein the circuit is also used to:

According to search order, the search routine is executed, wherein in the search order, ginseng relevant to the initial motion vector It examines image to be first searched, and after multiple reference pictures in two reference picture lists of the current prediction unit in an alternating manner And it is searched with reference key incremental order.

20. a kind of non-transitory computer-readable medium, stores multiple instruction, make the processor when executed by the processor A kind of video coding-decoding method is executed, for sub- predicting unit temporal motion vector prediction mode treatment current prediction unit, This method comprises:

Receive the current prediction unit including multiple sub- predicting units；

It is determined as the initial motion vector of the motion vector of the spatial neighboring blocks of the current prediction unit；

Based on the initial motion vector, it is main same to search in the reference image sequence of predicting unit in this prior to execute search routine Bit image, wherein the search routine include:

Opening movement vector zoom operations, with the subset for searching for the reference image sequence, and

The motion vector zoom operations are closed, for searching for other reference pictures in the reference image sequence；And

Obtain the same position motion information in the main co-sited image for multiple sub- predicting unit of the current prediction unit.

Technical field

The present invention relates to video coding and decoding technologies.

Background technique

Prior art description provided herein is typically used as illustrating the purpose of context of the invention.Signature inventor at present Action, both included the content of work described in this background section, be not considered when being also contained in application It is the various aspects of the specification of prior art, these are both indefinite or are not recognized as prior art of the invention with implying that.

In image and coding and decoding video, using the scheme based on tree construction, image and its respective sample array can be by Divide blocking.Each piece can be handled with one of a variety of tupes.Merging patterns are one of these tupes, wherein Spatial neighboring blocks and temporally adjacent piece can share identical motion parameter set.Therefore, motion vector transport overhead can be dropped It is low.

Summary of the invention

Aspect of the invention provides a kind of video coding-decoding method, with sub- predicting unit temporal motion vector prediction mode Handle current prediction unit.This method may include: to receive the current prediction unit including multiple sub- predicting units；It is determined as working as The initial motion vector of the motion vector of the spatial neighboring blocks of preceding predicting unit；Based on initial motion vector, search routine is executed To search for main co-sited image in the reference image sequence of current prediction unit；And it obtains for the multiple of current prediction unit Same position motion information in the main co-sited image of sub- predicting unit.Search routine may include: opening movement vector zoom operations, With the subset and closing movement vector zoom operations for searching for reference image sequence, in reference image sequence Search for other reference pictures.

In one embodiment, the subset of reference image sequence include current prediction unit the first reference picture list and The top n in reference image sequence at least one of second reference picture list of current prediction unit, or including Preceding N0 reference picture in the first reference picture list of current prediction unit, and the second reference in current prediction unit Preceding N1 reference picture in image list.First reference picture list and the second reference picture list are passed with reference key Increase tactic.In addition, N, N0 and N1 are greater than or equal to 0 integer value.

In one embodiment, opening movement vector zoom operations include: to execute motion vector scaling, to scale initial motion Vector is to obtain one scaling motion vector in the subset for detecting reference image sequence.In one example, it moves Vector scaling is in the subset based on the present image for including current prediction unit and the reference image sequence being studied First time distance between one and the present image including current prediction unit and reference relevant with initial motion vector The second time gap between image.

In one embodiment, opening movement vector zoom operations include: when a figure in the subset of reference image sequence When as being not reference picture relevant to initial motion vector, motion vector scaling is executed, to scale initial motion vector to obtain The scaling motion vector of an image in the subset of reference image sequence must be used to detect.In one embodiment, fortune is closed Moving vector zoom operations include: using initial motion vector, for detecting other reference pictures in reference image sequence.

The embodiment of this method further include: determine the search order for being used for search routine；And it is suitable according to determining search Sequence executes search routine.The embodiment of this method further includes that will be used in the reference image sequence of current prediction unit search for The search order of main co-sited image is transmitted from video encoder to Video Decoder.

In one embodiment, this method further include: according to search order, search routine is executed, wherein in search order, Reference picture relevant to initial motion vector is first searched, and multiple in two reference picture classifications of current prediction unit It is in an alternating manner and searched with reference key incremental order after reference picture.In one embodiment, this method further include: will lead Co-sited image is transmitted from video encoder to Video Decoder.

Aspect of the invention provides a kind of video encoding/decoding apparatus, with sub- predicting unit temporal motion vector prediction mode Handle current prediction unit.The present apparatus may include circuit, be used for: receive the current predictive list including multiple sub- predicting units Member；It is determined as the initial motion vector of the motion vector of the spatial neighboring blocks of current prediction unit；Based on initial motion vector, hold Row search routine in the reference image sequence of current prediction unit to search for main co-sited image；And it obtains and is used for current predictive Same position motion information in the main co-sited image of multiple sub- predicting units of unit.Search routine may include: opening movement to Zoom operations are measured, with the subset for searching for reference image sequence, wherein N is that predefined thresholds and closing movement vector contract Operation is put, for searching for other reference pictures in reference image sequence.

Aspect of the invention provides a kind of non-transitory computer-readable medium.Medium stores instruction, in fact current son The method of predicting unit temporal motion vector prediction mode treatment current prediction unit.

Detailed description of the invention

It will be described in detail in conjunction with following schema to exemplary various embodiments of the present invention are provided as, wherein Identical symbol indicates identical component, and wherein:

Fig. 1 is example video encoder according to an embodiment of the invention；

Fig. 2 is exemplary video decoder according to an embodiment of the invention；

Fig. 3 is according to some embodiments of the invention for advanced motion vector forecasting (advanced motion Vector prediction, advanced motion vector forecasting) mode derives motion vector predictor (motion vector Predictor, MVP) spatial candidate position and time candidate bit candidate or for deriving merging candidate with merging patterns The example set；

Fig. 4 is the example of motion vector zoom operations according to some embodiments of the invention；

Fig. 5 is according to some embodiments of the invention with sub- predicting unit temporal motion vector prediction (sub- Prediction unit temporal motion vector prediction, Sub-PU TMVP) mode treatment is currently pre- Survey the example flow of unit；

Fig. 6 is according to some embodiments of the invention with sub- predicting unit temporal motion vector prediction mode treatment current block Example flow；

Fig. 7 is the first of reduction calculating cost relevant to motion vector zoom operations according to some embodiments of the invention Exemplary method；

Fig. 8 is according to an embodiment of the invention currently pre- with sub- predicting unit temporal motion vector prediction mode treatment The main exemplary search sequence with position reference picture is searched for during surveying the process of unit；And

Fig. 9 is the second of reduction calculating cost relevant to motion vector zoom operations according to some embodiments of the invention Exemplary method.

Specific embodiment

Fig. 1 shows example video encoder 100 according to an embodiment of the present invention.Encoder 100 may include pre- in frame Survey module 110, Inter prediction module 120, first adder 131, residual coder 132, entropy coder 141, residual error decoding device 133, second adder 134 and decoded picture buffer device 151.Inter prediction module 120 can also include motion compensating module 121 and motion estimation module 122.As shown in Figure 1, these components can be coupled together.

Encoder 100 receives inputting video data 101, and executes video compress process to generate bit stream 102 as defeated Out.Inputting video data 101 may include image sequence.Each image may include one or more color components, such as bright Spend component or chromatic component.Bit stream 102 can have the format for meeting video encoding and decoding standard, such as advanced video compiles solution Code (Advanced Video Coding, AVC) standard, efficient video encoding and decoding (High Efficiency Video Coding, HEVC) standard etc..

Encoder 100 can be blocking by the image segmentation in inputting video data 101, for example, using based on tree construction Splitting scheme.In one example, encoder 100 can divide the image into coding unit (coding in a recursive manner Unit, CU).For example, image can be divided into coding tree unit (coding tree unit, CTU).Each coding tree unit Four more lower Item units can be recursively divided into, until default size is implemented.It is obtained from the recursive subdivision process Coding unit can be square block, but have different sizes.

Then, the block obtained can be handled with different tupes, for example, intra prediction mode or interframe it is pre- Survey mode.In some instances, the coding unit obtained can be further processed into predicting unit (prediction unit, PU it), and with prediction mode is handled.In some instances, the coding unit obtained can be further partitioned into multiple pre- Survey unit.In some instances, predicting unit may include the block of luma samples and/or the block of one or two chroma sample. Therefore, herein, predicting unit and prediction block (prediction block, PB) can be used interchangeably, to use for referring to The block of luma samples or chroma sample that prediction encoding/decoding mode is handled.In general, the segmentation of image can be with adaptivity In the local content of image.Therefore, the block (coding unit or predicting unit) obtained can have at the different location of image There are various sizes or shape.

In Fig. 1, intra-framed prediction module 110 may be configured to execute intra prediction with true during video compress process The prediction of fixed block (referred to as current block) currently being handled.Intra-frame image prediction can be based on image identical with current block The adjacent pixel of interior current block.For example, specified in 35 kinds of intra prediction modes such as HEVC standard.

Inter prediction module 120 may be configured to execute inter-prediction to determine current block during video compress process Prediction.For example, motion compensating module 121 can receive motion information (the movement number of current block with ego-motion estimation module 122 According to).In one example, motion information may include horizontal motion vector shift value and vertical motion vectors shift value, one or The mark of two reference picture index and reference picture list relevant to each index.Based on motion information and it is stored in One or more reference pictures in decoded picture buffer device 151, motion compensating module 121 can determine the prediction of current block. For example, as specified in HEVC standard, two reference picture lists, i.e. list 0 and list 1 can be configured to for compiling solution Code B type fragment, and each list may include the mark (identification, ID) of reference image sequence.List Each member can be related to reference key.Therefore, reference key and corresponding reference picture list can be believed together in movement It is used in breath, to identify the reference picture in this reference picture list.

Motion estimation module 122 may be configured to determine the motion information of current block, and motion information is supplied to fortune Dynamic compensating module 122.For example, using inter-frame mode module 123 or merging patterns module 124, motion estimation module 122 can be with It is handled with a pair of current block of multiple inter-frame forecast modes.For example, inter-frame forecast mode may include advanced motion vector Predict (advanced motion vector prediction, AMVP) mode, merging patterns, skip mode, sub- predicting unit Temporal motion vector prediction mode etc..

When current block is handled by inter-frame mode module 123, inter-frame mode module 123 may be configured to execute fortune Dynamic estimation process, to search for reference block similar with current block in one or more reference pictures.Reference block may be used as working as Preceding piece of prediction.In one example, one or more motion vectors and corresponding reference picture, which can be determined that, is based on being made The result of the estimation process of one-way or bi-directional prediction technique.For example, obtained reference picture can be by reference picture Index indicate, and it is bi-directional predicted by use in the case where, by corresponding reference picture list identify indicate.Since movement is estimated Process is counted, motion vector and coherent reference index can be determined for single directional prediction or two motion vectors and two Respective coherent reference index can be determined for bi-directional predicted.In addition, corresponding to each related ginseng for bi-directional predicted The reference picture list (list 0 or list 1) for examining index can also be identified.These motion informations (including it is identified One or two motion vector, coherent reference index and respective reference picture list) it is provided to motion compensating module 121. In addition, these motion informations can be included in the motion information 103 for being sent to entropy coder 141.

In one example, advanced motion vector prediction mode is used for the predictability encoded motion at inter-frame mode module 123 Vector.For example, motion vector predictor (motion vector predictor, MVP) candidate list can be constructed.Movement Vector predictor candidate list may include obtaining from current block space neighboring prediction block group or temporally adjacent prediction block group Motion vector predictor sequence.For example, the fortune of the space neighboring prediction block or temporally adjacent prediction block that are located at certain positions Moving vector is selected and is scaled, to obtain motion vector predictor sequence.Optimal motion vectors predict that sub- candidate can be from fortune The sub- candidate list of motion vector prediction (it is properly termed as motion vector prediction competition) selection, with previously determined for predictive coding Motion vector.Therefore, difference motion vector (motion vector difference, MVD) can be obtained.For example, having The motion vector predictor candidate of optimal motion vectors encoding-decoding efficiency can be selected.In this way, working as advanced motion vector forecasting When advanced motion vector prediction mode is applied to current block, the motion vector prediction of selected motion vector predictor candidate Subindex (referred to as motion vector predictor index) and respective motion vector difference can be included in motion information 103, And it is provided to entropy coder 141, to replace respective motion vector.

When current block is handled by merging patterns module 124, merging patterns module 124 may be configured to execute conjunction And the operation of mode, to determine the exercise data collection for the current block for being supplied to motion compensating module 121.For example, the son of candidate blocks Set can be from the spatial neighboring blocks and temporally adjacent piece of Resource selection for the current block being located at default candidate position.For example, Temporally adjacent piece can be located at Advanced Video, for example, the reference of current block (or present image including current block) At first reference picture of image list, i.e. list 0 or list 1.Then, when merging candidate list can be based on selected Between candidate blocks collection or spatial candidate block collection be constructed.Merging candidate list may include multiple entries.Each entry can wrap Include the motion information of candidate blocks.For time candidate blocks, before being formulated into merging candidate list, respective motion information (motion vector) can be scaled.In addition, the motion information corresponded in the merging candidate list of vision candidate blocks can have It is set as 0 reference key (meaning that the first image in list 0 or list 1 is used as reference picture).

Then, merge the best motion information for merging candidate and being selected and be confirmed as current block in candidate list (prediction competition).For example, it is assumed that respective entry is used as the motion information of current block, each entry can be evaluated.With most The merging candidate of high distortion performance can be determined to be shared by current block.Then, motion information to be shared can be provided To motion compensating module 121.In addition, the index for merging the selected entry in candidate list including exercise data to be shared can For indicating and transmitting this selection.This index is known as merging index.Motion information 103 can be included in by merging index In, and it is sent to entropy coder 141.

In optional example, skip mode can be used by Inter prediction module 120.For example, with above-mentioned merge is used Mode is similar, and current block can be predicted with skip mode, to determine exercise data collection, however, be generated without residual error or It sends.Skipping mark can be related to current block.Skipping mark can be transmitted to Video Decoder.In Video Decoder side Place may be used as decoding block based on the determining prediction (reference block) of index is merged, without adding residual signals.

In another example, sub- predicting unit temporal motion vector prediction mode (sub-PU TMVP) may be used as merging The part of mode, to handle current block (in this way, sub- predicting unit temporal motion vector prediction mode is referred to as sub- prediction list Elementary time motion vector prediction merging patterns).For example, merging patterns module 124 may include sub-block merging module 125, quilt It is configured to execute the operation of sub- predicting unit temporal motion vector prediction mode.In sub- predicting unit temporal motion vector prediction mould In the operation of formula, for example, current block can be further partitioned into sub-block collection.Then, the time of each sub-block with position move to Amount can be obtained, scale and be used as the motion vector of sub-block.These obtained motion vectors can be calculated as merging candidate (it is candidate that referred to as sub- predicting unit temporal motion vector prediction merges candidate or sub- predicting unit), and it is candidate to be formulated into merging In list.In addition, in some instances, reference picture index relevant to obtained motion vector is configured to 0, correspond to Reference picture list, i.e. list 0 or list 1.During above-mentioned merging candidate's estimation flow, such as fruit predicting unit candidate's quilt Selection (prediction competition), then corresponding to the candidate merging index of sub- predicting unit merging can be generated and be sent in motion information In 103.Sub- predicting unit candidate can also you can well imagine supply compensating module 121, generate current block based on sub- predicting unit candidate Prediction.

A variety of tupes are described above, for example, intra prediction mode, advanced motion vector prediction mode, merging mould Formula, sub- predicting unit temporal motion vector prediction mode and skip mode.In general, different blocks can use different tupes It is handled, and is made the decision for being used for block needs about which tupe.For example, mode decision can be Based on the test result that different disposal mode is applied to a block.Test result can be distorted based on the rate of respective tupe Performance is evaluated.Tupe with optimum can be determined that the selection for handling this block.In optional example, other Method or algorithm can be used to determine tupe.For example, image and since then image segmentation and the feature of block come can To be considered for determining tupe.

First adder 131 receives the prediction of current block from intra-framed prediction module 110 or motion compensating module 121, with And the current block from inputting video data 101.Then, first adder 131 can extract pre- from the pixel value of current block It surveys, to obtain the residual error of current block.The residual error of current block is sent to residual coder 132.

Residual coder 132 receives the residual error of block, and compresses this residual error, has compressed residual error to generate.For example, residual coding Device 132 can will first convert, such as discrete cosine transform (discrete cosine transform, DCT), discrete sine become (discrete sine transform, DST), wavelet transformation etc. are changed, the received residual error of institute corresponding to transform block is applied to, And generate the transformation coefficient of transform block.The segmentation for dividing the image into transform block can be with the segmentation that divides the image into prediction block It is same or different, to be used for interframe or intra-prediction process.Then, residual coder 132 can be with quantization parameter to compress Residual error.It has compressed residual error (quantization transform coefficient) and has been sent to residual error decoding device 133 and entropy coder 141.

Residual error has been compressed in the reception of residual error decoding device 133, and executes the quantization operation and change executed at residual coder 132 The inverse process of operation is changed, with the residual error of restructuring transformation block.Due to quantization operation, reconstructed residual is given birth to from adder 131 At initial residual error it is similar, but be typically different than original version.

Second adder 134 receives the prediction and autotomy of block from intra-framed prediction module 110 and motion compensating module 121 Poor decoder 133 receives the reconstructed residual of variation block.Then, second adder 134 by reconstructed residual and corresponds to image The received prediction of the institute of middle the same area is combined, to generate reconstructed video data.Reconstructed video data can be stored up In the presence of in decoded picture buffer device 151, the reference picture that can be used for inter prediction operating is formed.

Entropy coder 141 can be received from residual coder 132 has compressed residual error, and connects from Inter prediction module 120 Receive motion information 103.Entropy coder 141 also can receive other parameters and/or control information, for example, intra prediction mode is believed Breath, quantization parameter etc..The received parameter of the coding of entropy coder 141 institute or information, to form bit stream 102.Including to have pressed The bit stream 102 of the data of contracting format can be sent to decoder through telecommunication network, or be sent to storage facilities (example Such as, nonvolatile computer-readable medium), wherein the video data carried by bit stream 102 can be stored.

Fig. 2 shows exemplary video decoder 200 according to an embodiment of the present invention.Decoder 200 may include entropy decoding Device 241, intra-framed prediction module 210, including motion compensating module 221 and the Inter prediction module of merging patterns module 224 220, Residual error decoding device 233, adder 234 and decoded picture buffer device 251.As shown in Fig. 2, these components are coupled together. In one example, 200 self-encoding encoder of decoder receives bit stream 201, such as self-encoding encoder receives bit stream 102, and executes and go Compression process is to generate output video data 202.Exporting video data 202 may include image sequence, for example, it can be shown Show in display equipment, for example, monitor, touch screen etc..

Entropy decoder 241 receives bit stream 201, and executing is the coding process executed in Fig. 1 mark example by entropy coder 141 Inverse process decoding process.Therefore, motion information 203, intraprediction mode information, compressed residual error, quantization parameter, control Information etc. is obtained.Residual error decoding device 233 can be provided to by having compressed residual error.

Intra-framed prediction module 210 can receive intraprediction mode information, and therefore generates and use intra prediction mode coding Block prediction.Inter prediction module 220 can receive motion information 203 from entropy decoder 241, and therefore generate prediction, with For with advanced motion vector prediction mode, merging patterns, sub- predicting unit temporal motion vector prediction mode, skip mode etc. The block of coding.Prediction generated is provided to adder 234.

For example, inter-frame mode module 223 can receive fortune for the current block encoded with advanced motion vector prediction mode Motion vector prediction subindex and difference motion vector corresponding to current block.Frame mode module 223 can with regarded in Fig. 1 mark example The identical mode tectonic movement vector predictor candidate list of frame mode module 123 at frequency encoder 100.Using movement to Amount prediction subindex, and the motion vector predictor candidate list based on construction, motion vector predictor candidate can be determined. Then, motion vector can be derived through in conjunction with motion vector predictor candidate and difference motion vector, and be provided to fortune Dynamic compensating module 221.In conjunction with other motion informations, such as reference key, respective reference picture list, and it is based on being stored in The prediction of current block can be generated in reference picture in decoded picture buffer device 251, motion compensating module 221.

For the block encoded with merging patterns, merging patterns module 224 can be obtained with autokinesis information 203 merges index. In addition, merging patterns module 224 can be with side identical with the merging patterns module 124 at video encoder 100 in Fig. 1 mark example Formula juxtaposition candidate list.It is indexed using merging, and based on the merging candidate list constructed, merging candidate can be true It is fixed, and it is provided to motion compensating module 221.Therefore the prediction of current block can be generated in motion compensating module 221.

In one example, received merging index can indicate that sub- predicting unit temporal motion vector prediction is applied To current block.For example, merging index in predefined scope, to be used to indicate that sub- predicting unit to be candidate, or merge index It is related to special sign.Therefore, the relevant operation of sub- predicting unit temporal motion vector prediction mode can merge mould in sub-block It is performed at block 225, to derive that corresponding to the respective sub- predicting unit for merging index merges candidate.For example, sub-block merges Module 225 can obtain sub- predicting unit in a manner of identical with the sub-block merging module 125 in Fig. 1 mark example at video encoder Merge candidate.Then, the sub- predicting unit derived, which merges candidate, can be provided to motion compensating module 221.Movement is mended It repays module 221 therefore the prediction of current block can be generated.

Residual error decoding device 233, adder 234 can be with the residual error decoding devices 133 and second adder 134 in Fig. 1 example It is similar in terms of function and structure.Specifically, for the block encoded with skip mode, no residual error is generated for these Block.Decoded picture buffer device 251 stores reference picture, is used for the motion compensation executed at motion compensating module 221. For example, reference picture can be by forming from the received reconstructed video data of adder 234.In addition, reference picture can be self Decoded picture buffer device 251 obtains, and is included in output video data 202, for showing that equipment is shown.

In various embodiments, these components of encoder 100 and decoder 200 can with hardware, software or its Combination is realized.For example, merging module 124 and merging module 224 can use one or more integrated circuit (integrated Circuit, IC) it realizes, such as application specific integrated circuit (application specific integrated circuit, ASIC), scene can programming gate array (field programmable gate array, FPGA) etc..In another example merging Module 124 and merging module 224 may be implemented as including the instruction being stored in computer-readable nonvolatile storage medium Software or firmware.These instructions, when being executed by processing circuit, so that processing circuit executes merging module 124 or merges The function of module 224.

It is noted that merging module 124 and merging module 224 can be included in other decoders or encoder, It can have and Fig. 1 or shown in Fig. 2 similar or different structure.In addition, in different examples, 100 He of encoder Decoder 200 can be included in same equipment or autonomous device.

Fig. 3 show according to some embodiments of the invention for advanced motion vector prediction mode derive movement to Amount prediction candidate or the example for deriving candidate spatial candidate position and time position candidate with merging patterns.Figure Position candidate in 3 with for merging patterns or advanced motion vector prediction mode such as the candidate bit specified in HEVC standard It sets similar.As shown, prediction block 310 will be handled with merging patterns.Position candidate collection A0, A1, B0, B1, B2, T0, T1 } it is predefined.Specifically, position candidate { A0, A1, B0, B1, B2 } is spatial candidate position, at expression and prediction block 310 The position of spatial neighboring blocks in same image.Relatively, position candidate { T0, T1 } is time position candidate, indicates ginseng Examine temporally adjacent piece of position in image.In some instances, reference picture can be reference listing (i.e. list 0 or list 1) the first reference picture (reference key is equal to 0), for example, as specified in piece paiagraph header.Position candidate T1 is to be located at prediction Around the center of block 310.

In Fig. 3, each position candidate is indicated by the block of sample, for example, the size with 4x4 sample.The ruler of this block It is very little to permit equal to or less than for the minimum for generating prediction block defined in the splitting scheme based on tree of prediction block 310 Perhaps size (for example, 4x4 sample).In this configuration, in single neighboring prediction block, indicate that the block of position candidate can be total It is capped.In optional example, sample position can be used to indicate that position candidate.

In motion vector predictor candidate list or during merging candidate list construction process, at position candidate The motion information of neighboring prediction block can be selected as motion vector predictor candidate or merge candidate, and be included in movement Vector predictor candidate list merges in candidate list.In some scenes, the motion vector at position candidate is pre- Surveying son candidate or merging candidate can be disabled.For example, the candidate blocks being located at position candidate can be intra prediction , or the outside of the segment including current prediction block 310 can be located at, or be not in same volume with current prediction block 310 In code tree block (coding tree block, CTB) column.In some scenes, the merging candidate at position candidate be can be Redundancy.For example, merging the motion information of candidate and can be considered motion vector predictor candidate list or the conjunction of redundancy candidate And in candidate list another candidate motion information it is identical.In some instances, redundancy merging candidate can be from candidate list It removes.

In one example, in advanced motion vector prediction mode, left side motion vector predictor be can be from position Set first available candidate of { A0, A1 }, tip motions vector predictor can be from position { B0, B1, B2 } first can With candidate, and temporal motion vector prediction can be the first available candidate from position { T0, T1 } (T0 first made With.If T0 is unavailable, T1 is used).If left side motion-vector prediction is unavailable, and tip motions vector prediction It is not to have scaled motion-vector prediction, then if it is pre- in the presence of motion vector has been scaled in the candidate being located at { B0, B1, B2 } Son is surveyed, then the second tip motions vector prediction can be derived with from position { B0, B1, B2 }.As an example, in HEVC standard In, motion vector predictor candidate list is dimensioned to 2.Therefore, in two spaces motion vector predictor and one Between motion vector predictor reasoning flow after, the first two motion vector predictor can be included in motion vector predictor In candidate list.If the quantity of available motion vector prediction is less than 2, then null vector candidate can be with after removing redundancy It is added in motion vector predictor candidate list.

In one example, for merging patterns, up to four spaces merge candidate from position { A0, A1, B0, B1 } derivation Out, and time merges that candidate (T0 is first used from position { T0, T1 }.If T0 is unavailable, T1 is used) it derives Out.If to merge any one of candidate unavailable for four spaces, position B2 is for deriving that merging is candidate, using as replacing Generation.After four spaces merge reasoning flow candidate and that a time merging is candidate, removing redundancy can be applied, to move Except the merging of redundancy is candidate.If after removing redundancy the candidate column of predefined merging can be less than with candidate quantity is merged Table size (for example, in one example 5), then additional candidate can be derived, and be added to and merge in candidate list. In some instances, additional candidate may include following three kinds of candidate types: combined bidirectional prediction merging is candidate, it is double to have scaled It is candidate to merge the merging of candidate and null vector to prediction.

Fig. 4 shows the example of motion vector zoom operations 400 according to some embodiments of the invention.Through executing movement Vector zoom operations 400, having scaled motion vector 432 can derive with position motion vector 422 certainly.Specifically, fortune has been scaled Moving vector 432 is related to present image 430 and current reference image 440.Motion vector 432 has been scaled to be determined for being used for The prediction of current block 431 in present image 430.Relatively, with position motion vector 422 and co-sited image 420 and with position with reference to figure As 410 correlations.With position motion vector 422 be determined for in co-sited image 420 with the prediction of position block 421.In addition, figure As 410- image 440 can each be assigned image sequence count (picture order count, POC) value, i.e. image sequence 1- image sequence count 4 is counted, indicates the output position (or indicating the time) relative to other images in video sequence.

Specifically, it can be temporally adjacent piece of current block 431 with position block 421.For example, in Fig. 3, it can with position block 421 Be positioned at position candidate T0 perhaps position candidate T1 place temporally adjacent piece be used for advanced motion vector prediction mode or Merging patterns.In addition, corresponding to advanced motion vector prediction mode, current reference image 440 be can be by motion estimation operation The reference picture of identified current block 431.Corresponding to merging patterns, current reference image 440 can be pre-configuration to be used for Time merges candidate reference picture, for example, first in the reference picture list of current block 431, i.e. list 0 or list 1 Reference picture.

For motion vector zoom operations, it can be assumed that the value of motion vector two images relevant to this motion vector Between expression the time time gap it is proportional.Based on this it is assumed that having scaled motion vector 432 when can be through based on two Between distance scaling with position motion vector 422 obtain.For example, as shown in figure 4, first time distance 433 can be image sequence meter The difference of number 3- image sequence count 4, the second time gap 423 can be the difference of image sequence count 2- image sequence count 1.Cause This, using following expression, the horizontal displacement value or vertical displacement value for having scaled motion vector MVS_x or MVS_y can be with It is calculated:

Wherein MVC_x and MVC_y is the vertical displacement value and horizontal displacement value of same position motion vector 422.In optional example In, movement zoom operations can be performed in a manner of being different from aforesaid way.For example, being different from the expression formula of above-mentioned expression formula It can be used, and extra factor can be considered.

Fig. 5 shows being worked as with sub- predicting unit temporal motion vector prediction mode treatment according to some embodiments of the invention The example flow 500 of preceding predicting unit 510.Process 500 can be performed, to determine the sub-block for being used for current prediction unit 500 Merging Candidate Set (motion information).Process 500 can be in the 100 neutron merged block module 125 of video encoder in Fig. 1 mark example Place, or executed at the 200 neutron merged block module 225 of Video Decoder in Fig. 2 mark example.

Specifically, current prediction unit 510 can be divided into sub- predicting unit 501.For example, current prediction unit 510 It can have the size of MxN pixel, and be divided into a sub- predicting unit 501 of (M/P) x (N/Q), wherein M is divided by P, and N is divided by Q. The sub- predicting unit 501 each obtained is the size of PxQ pixel.For example, obtained sub- predicting unit 501 can have 8x8 picture The size of element, 4x4 pixel or 2x2 pixel.

Then, reference picture 520, referred to as time co-sited image 520, can be determined.Then, every sub- predicting unit 501 motion vector, referred to as initial sub- predicting unit motion vector, can be determined.As shown, sub- predicting unit 511 has There is initial sub- predicting unit motion vector 531.Sub- predicting unit 512 has initial sub- predicting unit motion vector 532.Hereafter, Time can be determined with the set of seat predicting unit (it is temporally adjacent piece of sub- predicting unit 501).Time same seat The set (each corresponding to sub- predicting unit 501) of predicting unit can be positioned at time co-sited image 520, and by initial Sub- predicting unit motion vector determines.As shown, respectively using initial sub- predicting unit motion vector 531 and initial son prediction Unit motion vector 532, time can be determined with seat predicting unit 521 and time with seat predicting unit 522 respectively.

Then, the identified time is obtained to be used for predicting unit 510 with the motion information of seat predicting unit.Example Such as, the time can be used for deriving the motion vector of sub- predicting unit 511 with the motion information of seat predicting unit 521.For example, Time may include motion vector 541, coherent reference index with the motion information of seat predicting unit 521, and optionally wrap Include the reference picture list for corresponding to coherent reference index.Similarly, motion information (packet of the time with seat predicting unit 522 Include motion vector 542) it can be used for deriving the motion vector of sub- predicting unit 512.For example, motion vector scaling can be held Row is to derive respective motion vector.

In the optional of the process 500 with sub- predicting unit temporal motion vector prediction mode treatment current prediction unit 510 In example, operation can be different from foregoing description.For example, in different examples, different sub- predicting units 501 can be used Different time co-sited images, and the method for determining time co-sited image can change.In addition, determining initial sub- predicting unit The method of motion vector can change.In one example, the initial sub- predicting unit motion vector of sub- predicting unit can be used Same motion vector, for example, the motion vector of the special neighbourhood block of current prediction unit 510.

As can be seen that sub- predicting unit temporal motion vector prediction mode enables the specific movement letter of multiple sub- predicting units Breath, to be derived and make to be configured for coding current block.Relatively, in traditional merging patterns, current block is located as a whole Reason, and merge candidate for entire current block.Therefore, sub- predicting unit temporal motion vector prediction mode can be than for son Traditional merging patterns of predicting unit potentially provide more accurate motion informations, to improve coding and decoding video efficiency.

Fig. 6 shows being worked as with sub- predicting unit temporal motion vector prediction mode treatment according to some embodiments of the invention The example flow 600 of preceding piece (predicting unit).Process 600 can be in the 100 neutron merged block mould of video encoder in Fig. 1 mark example It is executed at 200 neutron merged block module 225 of Video Decoder in block 125 or Fig. 2 mark example.Process 600 starts from S601, and Proceed to S610.

In S610, during search routine, determine that the reference picture of the sub- predicting unit for current prediction unit (claims For main co-sited image).Firstly, sub-block merging module 125 or sub-block merging module 225 can find current prediction unit Initial motion vector.Initial motion vector can be marked as vec_init.In one example, vec_init can be from The motion vector of first available space adjacent block, for example, being located at one of position { A0, A1, B0, B1, B2 } place in Fig. 3 mark example One of adjacent block.

In one example, vec_init is movement relevant to reference picture list searched first during search routine Vector.For example, the first available space adjacent block is located in B segment, it is related to two reference picture lists.It therefore, first can It can have and different reference picture lists, i.e. 1 relevant two motion vector of list 0 and list with spatial neighboring blocks.Two A motion vector is referred to as 1 motion vector of 0 motion vector of list and list.During search routine, in list 0 and list 1 One be searched (as described below) first with for main co-sited image, another is then searched.A first searched (list 0 or list 1) be known as first list, then searched one is known as second list.Therefore, in 0 motion vector of list and column In 1 motion vector of table, one relevant to first list may be used as vec_init.

For example, list X is first list, with for searching for the same position information in main co-sited image, if list X=list 0, then vec_init uses 0 motion vector of list, and if list X=list 1, vec_init using list 1 move to Amount.The value of list X (list 0 or list 1) depends on which list (list 0 or list 1) more preferably for position information. If list 0 is more preferably to be used for position information (for example, how to be constructed depending on list 0 and list 1, the figure of image in list 0 As sequential counting distance than image in list 1 image sequence count distance closer to), then list X=list 0, and anti-mistake Come same.List X distribution can be positioned at slice layer or image layer.In optional example, difference is can be used in vect_init Method and be determined.

After the initial motion vector of current prediction unit is determined, co-sited image search routine can start search master Co-sited image.Main co-sited image is marked as main_colpic.Co-sited image search routine will find current prediction unit The main co-sited image of sub- predicting unit.During co-sited image search routine, the reference picture of current prediction unit it is searched and It studies (investigated), one of reference picture is selected as main_colpic.In different examples, search routine can To implement in different ways.For example, reference picture can be with different methods (for example, scaling with or without motion vector Operation) it studies.Or the sequence of searching for reference image can change.

In one example, search is implemented in the following order.Firstly, the reference that search is selected by the first available space adjacent block Image (for example, reference picture relevant to initial motion vector).Then, in B segment, it may search for current prediction unit All reference pictures, from a reference picture list, i.e. list 0 (or list 1) starts, reference key 0, index 1, index 2 etc. Deng (increasing index order (index order)).If the search to list 0 (or list 1) is completed, without finding effectively Main co-sited image, then may search for another list, i.e. list 1 (or list 0).In P segment, may search in list 0 when The reference picture of preceding predicting unit then indexes 1, index 2 etc. (increasing index order) since reference key 0.

During the search of main co-sited image, reference picture is studied, to determine whether the image being studied has Effect is available.Therefore, this research of each reference picture is also referred to as availability detection.It in some instances, can be with such as Under type executes the research, for having searched for image each of other than reference picture relevant to initial motion vector (image being studied).In the first step, motion vector zoom operations can be executed.It scales and grasps through motion vector Make, initial motion vector is scaled to scale motion vector, is labeled as vec_init_scaled, corresponds to and is ground The reference picture studied carefully.Zoom operations can be based on present image (including current prediction unit and first available space adjacent block) First time distance and present image between reference picture relevant to initial motion vector and the ginseng being studied Examine the second time gap between image.(it is ginseng relevant to initial motion vector to the image being studied for first Examine image), do not execute zoom operations.

In some instances, optionally, it is performed before in motion vector zoom operations, if execute motion vector scaling Decision can be determined.For example, the reference picture being studied in list 0 or list 1 and with initial motion vector phase Whether the reference picture of pass is that same image is examined.When relevant to initial motion vector reference picture and it is being studied When reference picture is same image, motion vector scaling can be skipped, and the research of the image being studied can be with It is completed.In opposite scene, zoom operations can be executed as described above.

Here be two it is exemplary check the reference picture being studied in lists 0 or list 1 and with initial motion to Measure whether relevant reference picture is same image.In the first example, when the initial motion with the first available space adjacent block When the reference key of vector correlation is not equal to the reference key for the reference picture being studied, zoom operations can be performed. In another example, the Picture Order Count value of reference picture relevant to initial motion vector and be studied with reference to figure The Picture Order Count value of picture can be examined.When Picture Order Count value difference, zoom operations can be performed.

In the second step of the research, based on initial motion vector has been scaled, determined in the image being studied Position is detected, and detecting detection position is encoding and decoding in interframe encoding and decoding (being handled with intra prediction mode) or frame (being handled with intra prediction mode).If detection position is interframe encoding and decoding (availability detection is successful), just It may be used as main co-sited image in studied image, and search routine can stop.If detecting position is encoding and decoding in frame (availability detection be failure), then search can continue to study next reference picture.

In one example, the surrounding central position of current prediction unit is added vec_init_scaled, with by Detection position is determined in the image of research.In different examples, surrounding central position can be determined in different ways.One In example, surrounding central position can be center pixel.For example, for the current prediction unit having a size of MxN pixel, in surrounding Heart position can be position (M/2, N/2).In one example, surrounding central position can be the middle center in current prediction unit The center pixel of predicting unit.In one example, surrounding central position can be other than the position in the example of front two Positioned at the pericentral position of current prediction unit.In optional example, it can define in different ways and determine inspection Position.

For reference picture relevant to initial motion vector, the surrounding central position of current prediction unit can be added Vec_init, rather than vec_init_scaled detect position to determine.

In S620, the initial motion vector of the sub- predicting unit of current prediction unit can be determined.For example, having a size of The current prediction unit of MxN pixel can be divided into the sub- predicting unit having a size of PxQ pixel.Sub- predicting unit initial motion Vector can be determined for every sub- predicting unit.The sub- predicting unit initial motion vector of i-th of sub- predicting unit can To be marked as vec_init_sub_i (i=0~((M/P) x (N/Q) -1)).In one example, sub- predicting unit initial motion Vector is equal to the motion vector of scaling (the i.e. vec_init_sub_i=vec_ for corresponding to the main co-sited image found in S610 init_scaled).In one example, sub- predicting unit initial motion vector, i.e. vec_init_sub_i (i=0~((M/P) x (N/Q) -1)) can be mutually different, and can be based on one or more space neighboring prediction units of current block, or use other Suitable method is derived.

In S630, the co-sited image of sub- predicting unit may search for, be known as sub- predicting unit co-sited image.For example, For every sub- predicting unit, arranged from the sub- predicting unit co-sited image of reference picture list 0 and from reference picture The sub- predicting unit co-sited image of table 1 can be searched.In one example, there is only a co-sited images (to use institute as above The main_colpic stated), with the reference picture list 0 of all sub- predicting units for current prediction unit.In an example In, the sub- predicting unit co-sited image of the reference picture list 0 of all sub- predicting units can be different.In one example, it only deposits A co-sited image (using main_colpic as elucidated before), with all sub- prediction lists for current prediction unit The reference picture list 1 of member.In one example, the same bitmap of sub- predicting unit of the reference picture list 1 of all sub- predicting units Picture can be different.The sub- predicting unit co-sited image of the reference picture list 0 of i-th of sub- predicting unit can be marked as Collocated_picture_i_L0, and the sub- predicting unit co-sited image of the reference picture list 1 of i-th of sub- predicting unit It can be marked as collocated_picture_i_L1.In one example, main_colpic is used to list 0 and list 1 Current prediction unit all sub- predicting units.

In S640, it can determine that sub- predicting unit parity bit is set in sub- predicting unit co-sited image.For example, son prediction Parity bit in unit co-sited image, which is set, can be searched for sub- predicting unit.It in one example, can be according to as follows Expression formula determines that sub- predicting unit parity bit is set.

Parity bit sets x=sub-PU_i_x+vec_init_sub_i_x (integer part)+shift_x, and parity bit sets y= Sub-PU_i_y+vec_init_sub_i_y (integer part)+shift_y, wherein sub-PU_i_x indicates current prediction unit The horizontal top-left position (integer position) of the sub- predicting unit of interior i-th, sub-PU_i_y indicate i-th in current prediction unit The vertical top-left position (integer position) of a sub- predicting unit, vec_init_sub_i_x indicate the level of vec_init_sub_i Partially (vec_init_sub_i computationally can have integer part and fractional part, and integer part is used), vec_ Init_sub_i_y indicates that the vertical component (similarly, integer part is used) of vec_init_sub_i, shift_x indicate first Shift value, shift_y indicate second displacement value.In one example, shift_x can be the half of sub- predicting unit width, Shift_y can be the half of sub- predicting unit height.In optional example, shift_x or shift_y can be suitable using other When value.

In S650, the motion information that place is set positioned at sub- predicting unit parity bit can be obtained, to predict for every height Unit.For example, the motion information of time prediction as i-th of sub- predicting unit, is labeled as subPU_MI_i, can be obtained It is able to for every sub- predicting unit from respective sub- predicting unit co-sited image.SubPU_MI_i can be from same Position position x and parity bit set the movement of collocated_picture_i_L0 and collocated_picture_i_L1 on y Information.In one example, subPU_MI_i can be defined as MV_x, MV_y, coherent reference list, coherent reference index and Such as local luminance compensation mark other merging patterns sensitive informations set.MV_x and MV_y indicates to be located at i-th of son in advance Parity bit sets x in the collocated_picture_i_L0 and collocated_picture_i_L1 of survey unit and parity bit is set The horizontal motion vector shift value and vertical motion vectors shift value of motion vector at y.

In addition, in some instances, MV_x and MV_y can be according to co-sited image, present images and with position motion vector Time gap relationship between (motion vector, MV) zooms in and out.For example, sub- predicting unit in present image can be with With the first reference picture (for example, first reference picture in list 0 or list 1), and have including the sub- predicting unit With the sub- predicting unit co-sited image of position motion vector.It can be related to the second reference picture with position motion vector.Therefore, same to position Motion vector can be scaled based on the first time distance and sub- predicting unit between present image and the first reference picture The second time gap between co-sited image and the second reference picture has scaled motion vector to obtain.Process 600 can continue To S699, and end at S699.

Fig. 7 shows the calculating cost relevant to motion vector zoom operations of reduction according to some embodiments of the invention First exemplary method.The sequence 700 of M reference picture of current prediction unit is as shown in Figure 7.The sequence 700 of M reference picture May include reference picture relevant to initial motion vector and include current prediction unit one or two with reference to figure As the image in list (list 0 and list 1).For example, current prediction unit will use sub- predicting unit temporal motion vector prediction Mode is handled, for example, the process 600 based on Fig. 6.The sequence 700 of reference picture will be during co-sited image search routine It is studied, in order to described in the S610 such as Fig. 6, find main co-sited image.

In the scene of worst case, the motion vector zoom operations for scaling initial motion vector may need by It executes to be used for all M reference pictures (other than being identical to the image of reference picture relevant to initial motion vector). For example, current prediction unit there can be 32 reference pictures in list 0, and there are 32 reference pictures in list 1.Cause This, in worst case, general 64 times of motion vector zoom operations may need to be performed.In addition, in one image There may be a large amount of predicting units to be processed.Therefore, the calculating relevant to motion vector zoom operations in an image Cost may be very big in certain scenes.

In order to reduce motion vector scaling cost, in some embodiments, motion vector zoom operations are constrained to top n Reference picture to be searched, for example, at least one of reference picture list.N can be predefined thresholds.For example, N can be with It is 3,4 or 5 etc..For example, motion vector zoom operations can be performed in search top n reference picture to be searched；However, not having There is motion vector scaling to be performed when search is more than the reference picture of top n reference picture to be searched.On the contrary, initial motion Vector can be used for studying the image being studied.In other words, as shown in fig. 7, motion vector zoom operations can search It is turned on when rope top n reference picture to be searched, and is closed when search is more than the reference picture of top n reference picture to be searched It closes.

Assuming that search is more than that the reference picture of top n reference picture to be searched occurs when considering big quantum predicting unit Probability is smaller, and searches for top n reference picture to be searched probability usually with higher.In this way, skipping motion vector scaling behaviour Make that probability of happening is small, and obtained counter productive (loss of correctness) probability of happening is small, but worst case scene is big Meter is counted as originally to be avoided by.

As an example, as shown in fig. 7, when searching for top n reference picture, it is relevant to initial motion vector with reference to figure As can be first studied using initial motion vector.Then, other images in top n reference picture can be studied.Example Such as, whether the image being studied is that reference picture relevant to initial motion vector can be determined.Subsequently, for not being The image of reference picture relevant to initial motion vector being studied, motion vector scaling can be performed to obtain Scale motion vector.Then, having scaled motion vector can be used for studying the image being studied (for example, determining check bit Set the still interframe encoding and decoding of encoding and decoding in simultaneously detection frame).When search is more than the reference picture of top n reference picture 700, Scaling is not performed, and initial motion vector can be used for studying the image being studied.

Here is to realize to reduce the example for calculating cost relevant to motion vector zoom operations.In this example, variable i It is defined, as the search order during the search of main co-sited image.For example, when searching for the first image, i=1, when search second When image, i=2.The motion vector zoom operations of initial motion vector during searching for main co-sited image to top n image (i < =N) open, and searched for image sequence i >=(N+1) when be closed.If some has been searched for image and has been turned off to for just Beginning motion vector scaling, then vec_init_scaled=vec_init can be in sub- predicting unit temporal motion vector prediction stream This image is used in journey.If some searched for image be turned on scaled for initial motion vector motion vector scaling, Then the zoom version of vec_init_scaled=vec_init can be in sub- predicting unit temporal motion vector prediction process For this image.

It will now be described and flexibly define co-sited image search routine (for example, co-sited image described in the example of Fig. 6 is searched for Process) in search order embodiment.In one example, co-sited image search routine can have predefined sequence.For example, As described in the example of Fig. 6, search order be can be fixed: firstly, reference picture quilt relevant to initial motion example Search；Then, the image in list 0 (or list 1) is studied with reference key incremental order；And it is subsequent, list 1 (or List 0) in image be studied with reference key incremental order.However, the result of this permanent order may not be optimal 's.For example, reference picture list, i.e. list 0 and/or list 1, can be arranged differently because of different image or segment. In this way, the position of the nearest reference picture apart from present image in the form of time gap can change.It is solid compared to using Determine search order, the search order being adjusted flexibly in the main co-sited image of search, which can increase, finds the general of more preferable main co-sited image Rate.

Therefore, in some embodiments, the search order in main co-sited image search routine can be defined or be configured It is adaptive and adjustable.For example, the son at sub-block merging module 125 or Video Decoder 200 at video encoder 100 Merged block module 225 can neatly be selected or be determined according to the reference picture arrangement of some conditions, such as reference picture list Search order.In some instances, at encoder determined by search order can be in image sequence layer, image layer, slice layer Or it is transmitted in predicting unit layer to decoder.In some instances, encoder and decoder can be according to formation same search The same rule of sequence determines search order, transmits operation without any.

For example, in order to specify search for sequence, symbol below can be used.Image i to be searched with search order (being expressed as searched_i) can be by syntactic element { USING_INITVEC_REF_PIC, LIST, REFERENCE_INDEX } Distribution is specified.In this syntactic element, USING_INITVEC_REF_PIC can take 0 value or 1 value.If USING_ INITVEC_REF_PIC=1, then the syntactic element specifies image to be searched, for sky can be used by the first of current prediction unit Between the selected reference picture of adjacent block (reference picture i.e. relevant to initial motion vector).If USING_INITVEC_ REF_PIC=1, then the LIST in the syntactic element of searched_i and REFERENCE_INDEX are ignored.If USING_ INITVEC_REF_PIC=0, then the LIST in the syntactic element of searched_i and REFERENCE_INDEX become effective. Specifically, LIST specifies reference picture list, i.e. list 0 or list 1, and REFERENCE_INDEX to specify reference key.Cause This, reference picture can be specified by LIST and REFERENCE_INDEX.

Fig. 8 is shown according to an embodiment of the invention to be worked as with sub- predicting unit temporal motion vector prediction mode treatment The main exemplary search sequence with position reference picture is searched for during the process of preceding predicting unit.In fig. 8 it is shown that including current pre- Survey the present image 810 and two reference picture lists of unit, i.e. list 0 and list 1.List 0 include with reference key 0, Three reference pictures of reference key 1 and reference key 2, i.e. 831-833.List 1 includes having reference key 0, reference key 1 With three reference pictures of reference key 2, i.e. 841-843.For the initial reference of reference picture relevant to initial motion vector Image 820 is also shown as.Initial reference image 820 can be one in image 831-833 and image 841-843.

In the example of fig. 8, for search for the main search order with position reference picture be defined as it is as follows.Firstly, initial Reference picture 820 is searched.Then, the reference picture 831-833 in two reference picture lists and reference picture 841-843 are pressed It is according to alternate mode and searched according to reference key incremental order.Specifically, in sequence: 820,841,833,842,832, Reference picture in 843 and 831, Fig. 8 is searched.Main co-sited image is closer apart from present image 810 in time gap, this two Correlation between a image is higher, and the quality of the same position motion information of the sub- predicting unit for current prediction unit may Better.Therefore, compared to two reference picture lists of continuous search, the i.e. search order of list 0 and list, in the example of Fig. 8 The available better result of the search order of definition.

Search order similar with the search order of Fig. 8 mark example can be designated with syntactic element as follows:

As i=0, image searched_i is designated as { 1, disregard, disregard }；

Work as i=1, when 3,5... (odd-integral number), image searched_i is designated as { 0, LA, (((i+1)/2) -1) }；With And

Work as i=2, when 4,6... (even-integral number), image searched_i is designated as { 0, LB, ((i/2) -1) }.

In the above description, the operation of (/ 2) indicates division of integer (for example, 1/2=0,2/2=1 etc.).It is stated upper In bright, LA and LB can indicate two different reference picture lists.For example, in one example, LA=list 0, LB=list 1, in another example, LA=list 1, LB=list 0.

In one embodiment, main co-sited image is determined at encoder, and is then transmitted to decoder.Determining master Co-sited image can be co-sited image search routine as a result, can be pre-configuration.Alternatively, the main co-sited image determined It can be determined in another appropriate manner.Therefore, syntactic element can be transmitted from encoder to decoder, directly to refer to Show main co-sited image.In other words, main co-sited image can be transmitted from encoder to decoder.Accordingly, there exist a same positions Image selection.This, which is transmitted, can be designed in sequence layer, image layer, slice layer or predicting unit layer.In one example, it transmits It is reference picture relevant to initial motion vector with position master image.Therefore, syntactic element { 1, disregard, disregard } can be sent out It send.

Described herein flow and functionality may be implemented as computer program, execute by one or more processing When one or more processors can to execute respective flow and functionality.Computer program can be stored or be distributed in On medium appropriate, for example, together with other hardware provide or as other hardware a part light-memory medium or consolidate State medium.Computer program can also be distributed otherwise, for example, through internet or other wired or wireless electricity Communication system.For example, computer program can be obtained and load in a device comprising penetrate physical medium or distribution System obtains computer program, it may for example comprise from the server for being connected to internet.

Fig. 9 shows the calculating cost relevant to motion vector zoom operations of reduction according to some embodiments of the invention Second exemplary method.It is similar to the example of Fig. 7, another method is employed for the same bitmap of master in search current prediction unit When picture, the N number of reference picture studied with the motion vector zoom operations (zoom operations opened) being performed is determined. In fig. 9 it is shown that including the present image 910 and two reference picture lists of current prediction unit, i.e. list 0 and list 1. List 0 includes reference picture 931-934, is respectively provided with reference key 0, reference key 1, reference key 2 and reference key 3. List 1 includes reference picture 941-944, is respectively provided with reference key 0, reference key 1, reference key 2 and reference key 3. Initial reference image 920 for reference picture relevant to initial motion vector is also shown as.Initial reference image 920 can be One in image 931-934 and image 941-944.

In a specific example, the search order for main co-sited image search routine is defined as: 920,941-944 and 934-931.In other words, initial reference image 920 is first searched, and is followed by the list 1 according to index incremental order, and press According to the list 0 of index incremental order.In optional example, the search order for main co-sited image search routine is defined as: 920,934-932 and 941-944.After initial reference image 920 is first searched, list 0 is searched according to index incremental order Rope, then, list 1 are searched according to index incremental order.Which of above-mentioned two search order is used can be and be based on Some condition, or can be based on being carried on the respective piece paiagraph header in coding/decoding bit stream or the header in image header Information.

It is defined as the preceding N0 reference picture in list 0 with N number of reference picture that the zoom operations of unlatching are studied, with And the preceding N1 reference picture in list 1.In the example of figure 9, N0=2 and N1=2.Therefore, as reference picture 941-942 and When reference picture 934-933 is studied during search, zoom operations are turned on, and zoom operations can be performed with from initial Motion vector acquisition has scaled motion vector.Optionally, whether the image being studied is that initial reference image 920 can be by It checks.When the image being studied is initial reference image 920, zoom operations are not performed, and the research of this image can To be completed.Conversely, for not including reference picture 943-944 in preceding N0 or preceding N1 reference picture and with reference to figure As 932-931, zoom operations, which can be turned off to save, calculates cost.

In general, N number of reference picture in order to save the cost of motion vector zoom operations, in reference image sequence to be searched It can be determined, to be studied with the zoom operations of unlatching, and other reference pictures can be in reference image sequence to be searched It is studied with the zoom operations of closing.N number of reference picture can be determined that in the image sequence to be searched in Fig. 7 example Top n, or first part (N0) and second part (N1 including two reference picture lists, i.e., in list 0 and list 1 It is a) N number of reference picture.Optionally, N can by different from Fig. 7 or Fig. 9 it is exemplary in a manner of and be determined, for example, based on ginseng Examine how image list is constructed and how search order is used.In addition, N can also be variable based on different scenes , for example, the size etc. of statistics or current prediction unit based on present image.

Computer program can be accessed by the computer-readable medium of offer program instruction, to be used for by computer or appoint What instruction execution system uses, and is either connected to computer or any instruction execution system.Computer-readable medium can wrap Any device is included, storage, communication, transmission transmit computer program perhaps to be used for by instruction execution system, device or set It is standby to use, either it is connected to instruction execution system, device or equipment.Computer-readable medium may include computer-readable non- Temporary storage medium, for example, semiconductor or solid-state memory, tape, removable computer diskette, random-access memory (random access memory, RAM), read-only memory (read-only memory, ROM), disk and CD etc..It calculates The readable non-transitory storage medium of machine may include all types of computer-readable mediums, including the storage of magnetic storage medium, light Medium, flash medium and solid-state storage medium.

Due to have been combined it is of the invention be suggested as the exemplary various aspects that specific examples describe the present invention, These exemplary substitutions, modification and variation can be made.Therefore, embodiment illustrated here is used as illustrative purpose, but does not have to In limitation.In the case where not departing from the range of claims, can make a change.