CN101860754A - Method and apparatus for encoding and decoding motion vector - Google Patents

Abstract

The present invention relates to a video encoding method. In particular, it relates to Motion Vector (MV) coding, wherein MV coding is based on edge matching or adaptive template matching, i.e. motion vector coding based on boundary matching (MVCBM) and motion vector coding based on adaptive template matching (mvcmta). In general, the present disclosure defines a prediction candidate set, where the predictor adaptively changes based on the current distribution of neighboring MVs. Subsequently, matching techniques (boundary matching in MVCBM and adaptive template matching in mvcmat) are used to reduce the size of the prediction candidate set to reduce the number of bits used for indexing. Then, the optimal motion vector predictor is selected among the predictors included in the reduced set. The guessing strategy based on the minimum MVD standard further saves the number of index bits.

Description

The method and apparatus of motion vector encoder and decoding

Technical field

The present invention relates generally to Digital Video Processing.Especially, the present invention relates to comprise the method and apparatus of encoding and decoding motion vector encoder.

Background technology

Estimation (Motion estimation, be called for short ME) and motion compensation (Motioncompensation, be called for short MC) be the modern video coding standard (as MPEG-1, MPEG-2, MPEG-4, H.261, H.263, H.264 and AVS) in important tool, with the press time redundancy, thereby reach high coding efficiency.In ME/MC, present frame is divided into nonoverlapping block.For each current block in the present frame, limit search district in reference frame.Every bit in the field of search is corresponding to a candidate blocks, and this candidate blocks is the good predict of current block potentially.Limiting similarity measurement and search for to search candidate blocks based on similarity measurement in the field of search, even the similitude of this candidate blocks and current block is not a maximum, also is very big.This candidate blocks with maximum comparability will be a best matching blocks.Relative displacement between the fast and current block of optimum Match is called as motion vector, and it need be encoded.The coding of motion vector (motion vector is called for short MV) is also referred to as motion vector encoder (motion vectorcoding is called for short MVC).

On the other hand, for decoding end realizes motion compensation, coding side need expend a large amount of bit streams and be used to transmit motion vector information.For example in H.264, when bit rate very low (for example QP=40), the percentage that the spent bit number of motion vector information accounts for editor-in-chief's number of codes is very high, even reaches 50%.So this area presses for more efficient methods and carries out motion vector encoder and improve its code efficiency.

In up-to-date motion vector encoder method, many instruments have been introduced to improve the efficient of motion vector encoder.For example, for the quantity of the compressed-bit that reduces motion vector, the H.264 simple predictive coding encoding motion vector of standard utilization.For each current block, H.264 utilize a formation motion vector predictor (motion vector predictor is called for short MVP) of the centre of three adjacent motion vectors (MV).The motion vector difference of the true MV of MVP and current block (motion vector difference is called for short MVD) is encoded in the bit stream.Should " middle MVP " reduce the position that is used for the MV coding effectively, because it in most of the cases is similar to the true MV of current block.But middle MVP can not always optimally predict current MV.Need more effective MVP to be used for the coding figure place of MV with further minimizing.

Also once there was additive method to attempt to obtain the effective motion vector coding.For example, in the following discloses file, put down in writing to some extent: S.Kamp, " the Decoder side motion vectorderivation for inter frame video coding " of M.Evertz and M.Wien, come from the international image processing meeting (IEEEInternational Conference on Image Processing) of IEEE, pp.1120-1123,2008, (hereinafter referred to as " people's such as S.Kamp articles ") and the 7th, 023, No. 919, name is called the United States Patent (USP) of " Method for coding motionvector using 2-dimentional minimum bitrate predicting technique ".

In people's such as S.Kamp article, a kind of method of obtaining motion vector has been described.This method is carried out the method for similar estimation to obtain motion vector and reference key in decoding end.But, this method can not adaptively modifying MV fallout predictor Candidate Set, and this method abandon using existing accurate ME method to obtain MV but directly in decoding end with not being that the template matches mode obtains MV very accurately, and use conjecture tactful in this method.In addition, template matches is used the template of fixed size and shape.Do not use reference key or MVD in this method.Generally speaking, this method is very complicated, and the use template matches obtains MV and reference key self is always not reliable.

In No. 7023919 U.S. Patent application, a kind of method of motion vector encoder has been described.This method does not use matching technique to reduce MV fallout predictor Candidate Set, and the self-adapted candidate collection is not provided yet.Even use the conjecture strategy, it is based on minimum bit rate but not minimum MVD.Time-based MV fallout predictor is not used as the candidate in this method.This method lacks efficient, and is particularly all the more so under the very little situation of MVD.Different in addition MVD may have identical bit rate, and feasible standard based on minimum bit rate lacks efficient.

Also have method for distinguishing to utilize the border coupling to attempt.For example in the following discloses file: YanChen, Yang Hu, Oscar C.Au, Houqiang Li, " the Video ErrorConcealment Using Spatio-Temporal Boundary Matching and Partial DifferentialEquation " of Chang Wen Chen, IEEE multimedia journal (IEEE TRANSACTIONS ON MULTIMEDIA), VOL.10, NO.1, in January, 2008 (hereinafter referred to as " people's such as Yan Chen articles "), and the 5th, 596, No. 370, name is called the United States Patent (USP) of " Boundary matching motion estimation apparatus ".

In people's such as Yan Chen article, provide a kind of system and method that is used for the mistake of hiding vision signal.The border coupling only limits to be used for error concealing.Thereby the method for the disclosure is not pointed out the accuracy of utilizing the border match selection can improve the MV fallout predictor and is reduced MVD, does not utilize the existing MV that obtains from the accurate estimation of coding side yet.In addition, do not use template matches in this method.The shortcoming of this method is to utilize the border coupling always not reliable to obtain MV itself.

The 5th, 596, in No. 370 United States Patent (USP)s, provide a kind of device that carries out estimation.The border coupling only limits to obtain motion vector, and the purpose of this method provides a kind of alternative method of estimation.This method does not have to use the MV that obtains in the coding side estimation, shortcoming is to utilize the border coupling but not estimation acquisition MV itself is always not reliable.The border coupling only limits in order to obtain motion vector, and this method does not provide the information of motion vector encoder.Because not self-adapted candidate setting, so this mode is inefficent, and this method do not consider other matching techniques, for example the adaptive template coupling.

Thereby, need provide a kind of system and method for motion vector encoder, to solve above-mentioned one or more problem.

Other background technologies of relevant video coding can be with reference to file once: the H.264and M-PEG-4VideoCompression of Iain Richardson, Wiley﹠amp; Sons,

2003 and the The Essential Guide to Video Processing of Alan Bovik, Academic Press

2009.These disclosed files all are introduced into wherein as a reference.

Summary of the invention

In view of above-mentioned, the present invention needs the method and apparatus of a kind of motion vector encoder and decoding, and method efficient in the above-mentioned technology is low to overcome, the problem of poor reliability.

Particularly, the invention provides a kind of motion vector encoder method, may further comprise the steps:

Step 1: limit the Candidate Set of fallout predictor, wherein said fallout predictor is based on the motion vector adaptive variation adjacent with described motion vector;

Step 2: utilize matching process to reduce the number of the fallout predictor in the described Candidate Set to reduce the figure place that index uses;

Step 3:, and obtain current motion vector difference based on selected optimum fallout predictor in the remaining predicted device of standard in described Candidate Set of minimum movement phasor difference;

Step 4:, utilize the conjecture strategy to survey motion vector predictor is used for code index with further saving figure place based on remaining predicted device and current motion vector difference;

Step 5: judge whether to detect correct motion vector predictor, if detect correct motion vector predictor, then the current at least motion vector difference of coding if can't detect correct motion vector predictor, is then encoded to current at least motion vector difference and index.

Wherein, described matching process is border matching process or adaptive template matching process.

Wherein, described border matching process is, will paste the current data space corresponding to a part of data of the fallout predictor in the described Candidate Set, obtains the border matching error, removes one or more fallout predictors of matching error maximum in border in the described Candidate Set.

Wherein, described adaptive template matching process is to be identified for the shape of the adaptive template in current data space based on correlation possible between adaptive template and the current data; Perhaps be identified for the size of the adaptive template in current data space based on the consistency in the adaptive template.

Wherein, described adaptive template matching process also comprises the displacement of utilization based on fallout predictor correspondence in the Candidate Set, calculates poor between the template of the data space after the displacement in the template in current data space and the reference picture.

Wherein, in the adaptive template matching process, the width and the shape of described adaptive template are specified.

Wherein, after the described step 5,, then send the current motion vector difference of having encoded at least if detect correct motion vector predictor; If can't detect correct motion vector predictor, the position that then sends the current motion vector difference of having encoded at least and encoded as index.

Wherein, the Candidate Set of described fallout predictor comprises some proportional room and time fallout predictors.

Wherein, in the described step 1, the motion vector adjacent with described motion vector be adjacent with the described motion vector time, the space is adjacent, the time and space is adjacent or based on the adjacent motion vector of analysis strategy.

Wherein, conjecture strategy in the step 4, the number of described remaining predicted device is at least 1, each remaining predicted device and current motion vector difference are produced an intermediate parameters, standard based on the minimum movement phasor difference is reselected fallout predictor to be selected for each intermediate parameters, if the fallout predictor of reselecting to be selected is identical with the remaining predicted device that produces this intermediate parameters, judge that then this remaining predicted device is optional fallout predictor, if the number of optional fallout predictor is 1, then this optional fallout predictor is correct motion vector predictor.

The present invention also provides the method for the motion vector that a kind of decoding encoded, and described motion vector of having encoded is encoded by the method for claim 1, may further comprise the steps:

Step 1: decoding motion vectors is poor;

Step 2: limit the Candidate Set of fallout predictor, wherein said fallout predictor is based on the motion vector adaptive variation adjacent with described motion vector;

Step 3: utilize matching process to reduce the number of the fallout predictor in the described Candidate Set;

Step 4:, utilize the tactful judgement of conjecture can determine motion vector predictor based on the remaining predicted device in the described Candidate Set and the motion vector difference of decoding;

Step 5:, then utilize the motion vector difference and the fixed motion vector predictor of decoding to obtain motion vector if can determine motion vector predictor; If can not determine motion vector predictor, the index and judge motion vector predictor of then decoding utilizes the motion vector difference of decoding and the motion vector predictor of having judged to obtain motion vector.

The matching process that described matching process and when coding adopt is identical.

The present invention also provides a kind of motion vector encoder device, comprises with the lower part:

First selects module, in order to select to form the fallout predictor of Candidate Set;

First processing module receives described first Candidate Set of selecting the fallout predictor of module, reduces the number of the fallout predictor in the described Candidate Set in order to utilize matching process;

Second selects module, is connected with described first processing module, receives the remaining predicted device in the described Candidate Set, based on standard selected optimum fallout predictor in described remaining predicted device of minimum movement phasor difference;

Second processing module selects module to be connected with described second, obtains current motion vector difference according to the fallout predictor of the optimum of selecting;

Detecting module is connected with first processing module with described second processing module, based on remaining predicted device and the current motion vector difference in the described Candidate Set that receives, utilizes the conjecture strategy to survey motion vector predictor;

Judge module is connected with described detecting module, if the result of detection of detecting module is to detect correct motion vector predictor, then judged result is sent to coding module;

Coding module receives described judged result, the current at least motion vector difference of encoding or current at least motion vector difference of coding and index.

The present invention also provides a kind of motion vector decoding apparatus, comprises with the lower part:

Receiver module receives the motion vector difference of having encoded at least;

First decoder module connects described receiver module, and according to the motion vector difference of having encoded at least that receives, decoding motion vectors is poor;

First selects module, in order to select to form the fallout predictor of Candidate Set;

First processing module receives described first Candidate Set of selecting the fallout predictor of module, reduces the number of the fallout predictor in the described Candidate Set in order to utilize matching process;

Can judge module connects described first processing module and described first decoder module, based on the remaining predicted device in the described Candidate Set and the motion vector difference of decoding, utilize the tactful judgement of conjecture determine motion vector predictor;

Second decoder module connects described judge module, according to described judged result decoding index;

Analysis module connects described first processing module and second decoder module, judges motion vector predictor according to the index of decoding;

Second processing module connects described judge module, described first decoder module and described analysis module, according to judged result, utilizes motion vector difference and definite motion vector predictor of having decoded of decoding to obtain motion vector; Perhaps utilize the motion vector difference of decoding and the motion vector predictor of having judged to obtain motion vector.

The method of motion vector encoder provided by the invention and decoding can be carried out the processing of encoding and decoding rapidly, reliably, effectively to motion vector, improved the efficient of system.And the present invention can be used for the video coding system of any employing estimation, and movable information need be sent to decoder, such as but not limited to H.264, KTA, AVS, MPEG.In addition, the present invention also can be used for the encoder of any employing hybrid predicting coding.

By the embodiment of following explanation, other aspects of the present invention are also disclosed.

Description of drawings

Below with reference to the accompanying drawings, describe these and other purpose of the present invention, aspect and execution mode in detail, wherein:

Fig. 1 represents the exemplary expression according to the current block of the embodiment of the invention and adjacent block thereof.

Fig. 2 represents according to the inner boundary pixel of the piece of the embodiment of the invention and outer boundary pixel.

Fig. 3 represents the exemplary template according to the template matches of the embodiment of the invention.

Fig. 4 represents the exemplary template that the template width according to the embodiment of the invention equals 4.

Fig. 5 represents the flow chart according to the MVCBM of the embodiment of the invention.

Fig. 6 represents the flow chart according to the MVCATM of the embodiment of the invention.

Fig. 7 represents the block diagram according to the embodiment of a video coding system of the embodiment of the invention.

Fig. 8 represents the block diagram according to the embodiment of a video decoding system of the embodiment of the invention.

Fig. 9 represents the motion vector encoder schematic representation of apparatus according to the embodiment of the invention.

Figure 10 represents the schematic diagram according to the motion vector decoding apparatus of the embodiment of the invention.

Embodiment

The present invention relates to method for video coding.Especially, it relates to motion vector (MV) encodes, and wherein the MV coding is based on border coupling or adaptive template coupling, just based on the motion vector encoder (MVCBM) of border coupling and the motion vector encoder (MVCATM) that mates based on adaptive template.In general, the present invention limits the fallout predictor Candidate Set, and wherein fallout predictor changes adaptively based on the current distribution of adjacent MV.Subsequently, matching technique (the border coupling among the MVCBM and the adaptive template coupling among the MVCATM) is used to reduce the size of fallout predictor Candidate Set to reduce the coding figure place that index uses.Then, select optimum motion vector predictor in the fallout predictor among being contained in the collection of minimizing.Further saved the figure place of index based on the conjecture strategy of minimum MVD standard.

Particularly, possible MVP Candidate Set comprises some proportional room and time fallout predictors.In order to increase the diversity of fallout predictor, spatial predictors is based on the feature adaptively modifying of adjacent motion vectors.In order to select good fallout predictor from Candidate Set (CS), the matching technique of use is such as but not limited to border coupling (BM) and adaptive template coupling (ATM).From the fallout predictor of selecting by matching technique, select the optimum MVP of current block so that the MVD minimum.Because optimum MVP is based on the standard of minimum MVD and selects, thus the conjecture strategy introduced, thus can save the bit number that the MVP index signal is transferred to decoder in some cases.The present invention can reduce bit rate significantly with respect to standard H.264.

A. proportional motion vector predictor

The motion vector of adjacent block is used as the fallout predictor of the motion vector of current block.The different reference frames of the motion vector of adjacent block in can corresponding embodiment, a plurality of reference frames in for example allowing H.264 among this embodiment.During the corresponding different reference frame of the motion vector of two adjacent pieces, the relative present frame of its time gap is different.Thereby in using the embodiment of a plurality of reference frames, before being used as the motion vector of predict current block, the motion vector of adjacent block is according to its time gap quilt convergent-divergent pro rata.Fig. 1 represents the exemplary expression of current block and adjacent block thereof.With mv _A(MV of the piece A in the left side of current block) is example, supposes that the time gap between piece A and its reference block is d _p, the time gap between current block and its reference block is d _cCalculate proportional fallout predictor mv by following mode _SA

$m{v}_{\mathrm{SA}}={\mathrm{mv}}_A\×\frac{d_c}{d_p}---\left(1\right)$

Utilize the mv that identical method can computing block B _SB, the mv of piece C _SCDeng.Therefore, each fallout predictor all proportional convergent-divergent before using.

The B.MVP Candidate Set

One aspect of the present invention is to provide fallout predictor and coding method for the parameter generating in the data compression and expression, for example encoding motion vector in video compression.Select two or more candidates adaptively according to adjacent data message.The data message of " adjacent " comprises but is not limited to the space is adjacent, the time is adjacent, adjacent combination and based on some adjacent functions of analysis strategy.Analysis strategy includes but not limited to current relevant, distribution, for example selects the similarly not adjacent motion vector of adjacent vector with other as the candidate, in order to the current motion vector of coding in video compression.

Possible MVP Candidate Set comprises some proportional room and time fallout predictors.In order to increase the diversity of fallout predictor, spatial predictors is based on the feature adaptively modifying of adjacent motion vectors.

In one embodiment, reach high accuracy, comprise the MVP in time and space in the MVP Candidate Set in order to make MVP.The MVP Candidate Set comprises a plurality of such MVP, and its quantity according to circumstances becomes, such as but not limited to 5 MVP or 8 MVP.

In further embodiment, for a given Candidate Set, the fallout predictor that comprises changes adaptively based on the current distribution of MVP.For example, do not lose generality, the size of Candidate Set is selected as 3, and size equals to comprise mv in 3 the Candidate Set _Scol, mv _SH.264And mv _SneiMv _ScolBe common location MV (referring to have the motion vector of the piece of same position) at former frame and current block according to the proportional convergent-divergent of formula (1).It is time MVP candidate.Mv _SH.264And mv _SneiBe space M VP candidate.

Mv _SH.264Be one (these 3 adjacent motion vectors are according to the proportional convergent-divergent of formula (1)) of the centre of 3 adjacent motion vectors, for example it can be from mv _SA, mv _SBAnd mv _SCAn acquisition of centre.

Mv _SneiBe three proportional adjacent motion vectors mv _SA, mv _SBAnd mv _SCIn one, it is from mv _SH.264Farthest:

${\mathrm{mv}}_{\mathrm{Snei}}=\arg \underset{{\mathrm{mv}}_x}{\max }{\left|\right|{\mathrm{mv}}_x-{\mathrm{mv}}_{\mathrm{SH},264}\left|\right|}^2,x=\mathrm{SA},\mathrm{SB},\mathrm{SC}---\left(2\right)$

This fallout predictor mv is set _SneiReason be, in adjacent piece A, B and C only one when belonging to the object that comprises current block, try to obtain more effective MVP.In this case, the correlation of the actual MV of Zhong Jian MVP and current block may be very low.

So MVP Candidate Set (Candidate Set, be called for short CS) is made up of three MVP under this concrete exemplary condition, wherein under this condition the size of Candidate Set equal 3 (size-3CS):

CS＝{mv _SH.264，mv _Scol，mv _Snei} (3)

C.CS reduces and optimizes MVP

Initial MVP Candidate Set can comprise many MVP, and one the index of selecting needs many positions with explanation.In order to reduce the position of the index that is used for optimum MVP, reduce CS to comprise the fallout predictor of lesser number.Among the embodiment, use the method that is called the border coupling.Among another embodiment, use the method that is called the adaptive template coupling.Make Candidate Set (Reduced CandidateSet the is called for short RCS) fixed size or variable of minimizing based on matching status.For example select the optimum prediction device, also select the several good fallout predictor of back if the fallout predictor of back is close to well.If perhaps best fallout predictor is much better than other fallout predictors, then we also can only select best fallout predictor.Has mv at this _Scol, mv _SH.264And mv _SneiThe special case of size-3CS in, we can use the RCS that is of a size of 2 fixed dimension.

D. mate on the border

One aspect of the invention is: determine the self-adapted candidate collection; Reduce Candidate Set to save the figure place of index by matching technique; Execution is based on conjecture strategy and the whole MV cataloged procedure of minimum MVD; In the MV coding, utilize the border coupling as matching technique.

Further aspect of the present invention also comprises: carry out the border coupling, for example will copy to corresponding to a part of data of each candidate in the current data space, harmony and continuity between the adjacent data in measurement current data space and the data division of stickup.

In following article, (Boundary matching is mated on the border, be called for short BM) existing explanation: Yan Chen, Yang Hu, Oscar C.Au, Houqiang Li, " the Video ErrorConcealment Using Spatio-Temporal Boundary Matching and Partial DifferentialEquation " of Chang Wen Chen, come from IEEE multimedia journal (IEEE TRANSACTIONS ONMULTIMEDIA), VOL.10, NO.1, in January, 2008, be incorporated herein its disclosure.

The border coupling is widely used in error concealment to find the most probable motion vector that points in the reference frame, to recover the loss piece.BM estimates based on the search of the spatial continuity between piece and the adjacent block thereof fallout predictor.If use incorrect fallout predictor, then corresponding reconstructed block is adjacent piece, and not have the successional possibility of high spatial very big.If but used correct fallout predictor, then piece of Chong Jianing and adjacent block should be continuous.

Match-on criterion is border matching error (boundary matching error is called for short BME), the space discontinuity between the inner boundary pixel of its measurement candidate blocks and the outer boundary pixel of current block, as shown in Figure 2.Fig. 2 represents the inner boundary pixel and the outer boundary pixel of piece.Letter N, W, E, S represent direction north, west, Dong Henan.BME has two parts: level and smooth discontinuous distortion (smoothdiscontinuity distortion is called for short SDD) and the discontinuous distortion in edge (edge discontinuitydistortion is called for short EDD):

BME＝α×SDD+(1-α)×EDD (4)

Wherein weight factor α is the real number between 0 to 1.

BM can both carry out to find suitable fallout predictor in encoder in the MV coding method of noticing in this instantiation.Because the boundary information that obtains easily, for example information on north side and/or border, west side usually as yet not at decoder decode, are only used in the east side of current block and southern side.SDD and EDD are calculated as follows:

$\mathrm{SDD}=\underset{k=0}{\overset{N_X-1}{\mathrm{\Σ}}}|p_{\mathrm{rec}}(x+k,y)-p_{\mathrm{rec}}(x+k,y-1)|+\underset{k=0}{\overset{N_W-1}{\mathrm{\Σ}}}|p_{\mathrm{rec}}(x,y+k)-p_{\mathrm{rec}}(x-1,y+k)|---\left(5\right)$

$\mathrm{EDD}=\underset{k-0}{\overset{N_w-1}{\mathrm{\Σ}}}|\frac{\▿\left({\mathrm{\Δp}}_{\mathrm{rec}}(x+k,y)\right)}{|\▿\left(\mathrm{\Δ}{p}_{\mathrm{rec}}(x+k,y)\right)|}\·\frac{{\▿}^{\⊥}{p}_{\mathrm{rec}}(x+k,y)}{\left|{\▿}^{\⊥}{p}_{\mathrm{rec}}(x+k,y)\right|}|\×|\▿p_{\mathrm{rec}}(x+k,y)|$

$+\underset{k-0}{\overset{N_h-1}{\mathrm{\Σ}}}|\frac{\▿\left({\mathrm{\Δp}}_{\mathrm{rec}}(x,y+k)\right)}{|\▿\left(\mathrm{\Δ}{p}_{\mathrm{rec}}(x,y+k)\right)|}\·\frac{{\▿}^{\⊥}{p}_{\mathrm{rec}}(x,y+k)}{\left|{\▿}^{\⊥}{p}_{\mathrm{rec}}(x,y+k)\right|}|\×|\▿p_{\mathrm{rec}}(x,y+k)|---\left(6\right)$

If wherein (x, y) directed outwards border, then p _Rec(x refers to y) that (x, y) the current reconstruction pixel value of position is if (x y) points in the inner boundary, then p _Rec(x y) refers to possible candidate blocks. Be gradient operator,

Be the normal operator with the direction of gradient direction quadrature,

It is Laplacian.In an exemplary embodiments, these relevant operators can be calculated as follows:

$\left|{\▿}^{\⊥}{p}_{\mathrm{rec}}(x,y)\right|=|\▿p_{\mathrm{rec}}(x,y)=\sqrt{{\left[\frac{\∂p_{\mathrm{rec}}(x,y)}{\∂x}\right]}^2+{\left[\frac{\∂p_{\mathrm{rec}}(x,y)}{\∂y}\right]}^2}$

$\frac{\∂p_{\mathrm{rec}}(x,y)}{\∂x}=\frac{p_{\mathrm{rec}}(x+1,y)-p_{\mathrm{rec}}(x-1,y)}{2}$

$\frac{\∂p_{\mathrm{rec}}(x,y)}{\∂y}=\frac{p_{\mathrm{rec}}(x,y+1)-p_{\mathrm{rec}}(x,y-1)}{2}$

$\frac{{\∂}^2{p}_{\mathrm{rec}}(x,y)}{{\∂}^{2}x}=p_{\mathrm{rec}}(x+1,y)+p_{\mathrm{rec}}(x-1,y)-2p_{\mathrm{rec}}(x,y)$

$\frac{{\∂}^2{p}_{\mathrm{rec}}(x,y)}{{\∂}^{2}y}=p_{\mathrm{rec}}(x,y+1)+p_{\mathrm{rec}}(x,y-1)-2p_{\mathrm{rec}}(x,y)---\left(7\right)$

In the equation (6), (▽ (Δ))/(| (▽ (Δ)) |) represented the normalization gradient of Laplace estimation, (▽ ^⊥)/(| ▽ ^⊥|) be tangential normalized vector.If cross-border structure is mated suitably, two should be orthogonal and inner product be zero.But if mate improperly, it is big that the absolute value of two inner product tends to become, and causes that inner product becomes big.In addition, for each pixel in the equation (6), inner product is multiplied by gradient magnitude | ▽ | and, this tends to make that the number range of EDD and the number range of SDD have comparativity.Inner product has also reflected the situation of the piece of boundary.

For each fallout predictor among the CS, the corresponding reference piece is used to calculate BME.Those fallout predictors with less BME are included among the RCS.For example, has mv _Scol, mv _SH.264And mv _SneiThe special case of CS of size-3 in, remove fallout predictor with maximum BME, comprise other two fallout predictors among the RCS.

E. adaptive template mates

Another aspect of the invention is: determine the self-adapted candidate collection; Reduce Candidate Set to save the figure place of index by matching technique; Exploitation adaptive template shape and width strategy; Execution is based on conjecture strategy and the whole MV cataloged procedure of minimum MVD; Utilize the adaptive template coupling as matching technique in the MV coding.In addition, the invention provides the strategy of determining shape of template and width based on correlation and similitude standard, this strategy is applicable to multiple application, such as but not limited to motion vector encoder.

Further aspect of the present invention also comprises:

Be identified for the shape of template in current data space based on correlation possible between template and the current data, for example, as block-based video compression system H.264 in, if macro block is divided into different subregions (this means that the correlation between the different subregions can be very low) by rate-distortion criterion, the template of this subregion may not comprise the pixel in another subregion;

Be identified for the template size in current data space based on the consistency in the template, for example, calculate the poor of possible template inside, if with current data space pixel a long way off be not similar to the current data space near pixel, then can from template, remove pixel a long way off; And

The template matches technology can include but not limited to, utilizes the displacement based on the candidate, calculates in the template in current data space and the reference picture poor based between the template of displacement.

Fig. 3 represents the exemplary template of template matches.Template matches illustrates in following article: S.Kamp, " the Decoder side motion vector derivation for inter framevideo coding " of M.Evertz and M.Wien, come from the international image processing meeting (IEEE International Conference onImage Processing) of IEEE, pp.1120-1123,2008, introduce by reference at this.In order to obtain the good predict of current block, limit L shaped template region (TR) around the left margin and the coboundary of object block.Limit template width M_left and M_up and be the width of the pixel of the target left side that extends to template region and top.For the template of testing current block and corresponding to the similitude between the template of the candidate blocks of MVP, the absolute difference and (sum of absolute difference is called for short SAD) that only need to calculate two templates.When template belongs to the object identical with current block,, can rationally infer the good predict that current block also can be provided corresponding to the piece of MVP if the template SAD that MVP provides is very little.

But if the some parts in the template region belongs to another object with different motion, even MVP is good, this still can draw a large form SAD.Among the embodiment,, do not use the M_left and the M_up of fixed dimension, and be based on the possible correlation between template and the object block and the similitude of template inside, use adaptive shape of template and width for avoiding this situation.In preferred embodiment,, use large form owing to can provide more significant template SAD.But any affiliated object pixels of different current blocks that belongs to should be not included in this template.

L shaped template comprises two parts: upper section and left part.Term " left side template " refers to the left part of L shaped template.Similarly, term " top template " refers to the upper portion of L shaped template.Typically, we to allow template be L shaped template, left side template or top template.

Has mv _Scol, mv _SH.264And mv _SneiThe one exemplary embodiment of CS of size-3 in, we limit the shape of template selection strategy as follows according to the correlation that current block is adjacent between the piece:

If (blocktype=P16*16)

Then use left side template and top template;

The subregion * of the subregion of 16x8 and below 16x8 above (blocktype=P16*8)/* has else if/

Then the top subregion uses left side template and top template, and the below subregion only uses the left side template;

(blocktype=P8*16)/* has the subregion of left side 8x16 else if, the subregion * of right side 8x16/then

The left side subregion only uses the top template with use left side template and top template, right partition;

(blocktype=P8*8)/* has the subregion of four 8x8 else if, some of them can be further divided into the subregion * of the littler 4x8 of being of a size of, 8x4,4x4/

Then the left upper subregion of current MB uses left side template and top template,

The left margin punishment district of current MB only uses the left side template,

The punishment district, coboundary of current MB only uses the top template,

Other subregions use left side template and top template.

This tactful reason also is the reason that we wish macro block (MB) is divided into different subregions.Often have relatively low correlation or different motion conditions between the different subregions.So when using template SAD to select good MVP, the pixel in another subregion is included in the template and may brings risk to the correctness of selected good MVP.But also can use other shape of template selection strategy.

If current block is level and smooth, then need the pixel in the template to have similar brightness and flatness.If current block is a texture block, then need the pixel in the template to have similar texture.If current block has the edge, then need the pixel in the template to have the edge, this edge is the extension at the edge in the piece, and the pixel on the both sides of edges is similar to the interior pixel of piece in the template.In other words, according to consistency between template and the piece or similarity the template width is set.

Has mv _Scol, mv _SH.264And mv _SneiThe one exemplary embodiment of CS of size-3 in, limit the structure standard of template width as follows:

Fig. 4 represents the exemplary template that the template width equals 4.

The Breadth Maximum of template is N=4; / * just template have N row pixel */

Poor in the calculation template between the adjacent row.With the top template is example, as shown in Figure 4, calculates:

Absolute value difference sum (SAD) between SAD12=1 row and 2 rows;

SAD between SAD23=2 row and 3 rows;

SAD between SAD34=3 row and 4 rows;

(if SAD12＞SAD23+ threshold value 1)

The top template only comprises 1 row; Otherwise

(if SAD23＞SAD12+ threshold value 2 ‖ SAD12＞SAD34+ threshold value 3)

The top template comprises 1 row and 2 rows; Otherwise

(if SAD34＞SAD12+ threshold value 2)

The top template comprises 1 row, 2 rows, 3 rows;

Otherwise

The top template comprises 1 row, 2 rows, 3 rows, 4 rows;

Same, use identical method to obtain the width of left side template.In case current block has been set shape of template and width, the template of reference block with identical shaped and width is just in order to calculation template SAD.Better matching template zone in the reference picture has and less template SAD between the front template.Because the spatial coherence of template and current block is very high, so have reason to suppose that the piece corresponding to better matching template can provide good predict for current block.

In the present invention, all calculate corresponding template SAD for each fallout predictor among the CS.Those fallout predictors with less template SAD are included among the RCS.For example, has mv _Scol, mv _SH.264And mv _SneiThe special case of CS of size-3 in, remove fallout predictor with maximum template SAD, comprise other two fallout predictors among the RCS.

F. final MVP decision

Another aspect of the present invention is that all candidates in the Candidate Set that reduces are carried out comparison and counting process, and has all compared the difference of candidate and parameter current and the poor of minimum of encoding in each comparison and the counting process.

In the present invention, according to the strategy that is provided with, the Candidate Set after matching technique reduces may comprise one or more MVP.For example, has mv _Scol, mv _SH.264And mv _SneiThe special case of CS of size-3 in, RCS comprises two fallout predictors.So also need from two fallout predictors, to select final fallout predictor.Now with the pmv of the final MVP of current block _OptLimit as follows:

${\mathrm{pmv}}_{\mathrm{opt}}=\arg \underset{{\mathrm{pmv}}_i}{\min }D(\mathrm{mv}-{\mathrm{pmv}}_i)---\left(8\right)$

Pmv wherein _iRefer to i the MVP that is included among the RCS, D () is the function in order to measuring distance, and mv represents the motion vector of current block.The final MVD of current block is (mv-pmv _Opt).

G. index is encoded

Another aspect of the present invention is by guessing strategy coding candidate's index, this candidate compare with parameter current has minimum poor.Conjecture strategy is based on the difference of the minimum of having encoded and the Candidate Set of minimizing: generally speaking, N candidate produced N possible parameter, reselect optimal candidate and the new lowest difference of acquisition based on the standard of the difference of minimum for each possible parameter subsequently.If this optimal candidate identical with the candidate who produces this possibility parameter (new lowest difference also equates with original encoded lowest difference simultaneously), then this candidate may; Otherwise set this candidate for impossible.Candidate if possible only has one, does not then need code index, otherwise needs code index.

When all MVP are identical among the RCS, do not need the index notice selected.Otherwise, need one or more the index of notifying final MVP.But optimum MVP can obtain at the decoder place in some cases, does not need to transmit index.Illustrated below how optimum MVP obtains: HyunDuk Cho, " the Method for coding motion vector using 2-dimentionalminimum bitrate predicting technique " of Sung Deuk Kim, the 7023919B2 United States Patent (USP), in April, 2006, introduce its disclosure by reference herein.The conjecture strategy has been proposed.Different is that the present embodiment use minimizes MVD as standard, and as standard, efficient was not high when the latter had identical bits speed at different MVD with minimum bit rate in replacement.

In the present embodiment, suppose that the number of remaining predicted device is N in the Candidate Set, and has determined optimum prediction device and MVD through after the matching technique.For each fallout predictor MVP_i, suppose that at first it is final fallout predictor, thereby can obtain a new motion vector (value of MVP_i adds MVD).This new motion vector is selected again optimum one according to the minimum standard of difference in N fallout predictor,, illustrate that then MVP_i is possible, otherwise MVP_i is impossible if that selection is MVP_i.

To have mv _Scol, mv _SH.264And mv _Snei(have 2 MVPmv among the RCS _BM1And mv _BM2) size-3 CS one exemplary embodiment as an example, the expression how to work:

Step 1: obtain two possible motion vector MV1=MVD+mv _BM1And MV2=MVD+mv _BM2

Step 2: suppose that MV1 is real motion vector, select to be used for the optimum prediction device of MV1 according to equation 8.If being used for the optimum prediction device of MV1 is mv _BM1, flag_BM1=1 then is set, otherwise flag_BM1=0;

Step 3: suppose that MV2 is real motion vector, select to be used for the optimum prediction device of MV2 according to equation 8.If being used for the optimum prediction device of MV2 is mv _BM2, flag_BM2=1 then is set, otherwise flag_BM2=0;

Step 4: if ((flag_BM1=1 ﹠amp; ﹠amp; Flag_BM2=1) ‖ (flag_BM1=0﹠amp; ﹠amp; Flag_BM2=0)), the index of final MVP will be transmitted.Otherwise do not transmit index, can obtain the optimum prediction device according to flag_BM1 and flag_BM2.

Fig. 5 represents to be used for the flow chart of MVCBM.With border coupling as matching technique with the one exemplary embodiment of carrying out motion vector encoder in:

In coder side

MVCBM (encoder):

Step 501: be calculated to be the medium range forecast device (mv_SH.264) of proportional zoom, altogether location fallout predictor (mv_Scol) with contiguous fallout predictor in the least identical fallout predictor (mv_Snei) of medium range forecast device, but by with equation (1) similarly method will be total to location fallout predictor (mv_Scol) and the proportional convergent-divergent of vicinity fallout predictor.

Step 503: utilize two fallout predictors of border match selection (mv_bm1 and mv_bm2) to enter the Candidate Set (RCS) of minimizing with less border matching error.Can certainly select more than two detectors.

Step 505: based on the MVD standard of minimum, between mv_bm1 and mv_bm2, select final MVP, obtain MVD for current block.

Step 507:, utilize the conjecture strategy to judge whether to detect correct MVP based on current MVD; If can detect correct MVP, need not encode subsequently and send the position as index; Otherwise need encode subsequently and send 1 as index (at the number of coupling back, border fallout predictor greater than 2, then can be greater than 1) as the position of index;

Step 509: coding MVD and other information also need code index under the situation that can not detect correct MVP.Other information for example comprise the residual error of coded block pattern (Coded Block Pattern is called for short CBP), quantification etc.

In the described step 509, in the coding flow process of reality, the MVP index of all subregions is put together to encode and is exported in the macro block, and is in whole macroblock coding output backmost.The reason of doing like this is in decoding end, just consider decoding MVP index in process of reconstruction, and other information of macro block is all decoded before.

At decoder-side

MVCBM (decoder):

Step 511: decoding MVD and other information begin to rebuild current block.Other information for example comprise residual error that for example comprises CBP, quantification etc.

The step 501 of step 513 and step 515 and encoder, 503 identical.

Step 517:, utilize the tactful judgement of conjecture can determine MVP based on the MVD of decoding; If which can't be determined is MVP, then decoding 1 is index.

Step 519: utilize MVP and MVD to obtain the MV of current block.

Fig. 6 represents the flow chart of MVCATM.In with adaptive template coupling another one exemplary embodiment as the matching technique of motion vector encoder:

Coder side

MVCATM (encoder)

Step 601: be calculated to be the medium range forecast device (mv_SH.264) of proportional zoom, altogether location fallout predictor (mv_Scol) with contiguous fallout predictor in the least identical fallout predictor (mv_Snei) of medium range forecast device, but by with equation (1) similarly method will be total to location fallout predictor (mv_Scol) and the proportional convergent-divergent of vicinity fallout predictor.

Step 603:, have the template of self adaptation shape and width for current block limits based on similarity possible between template and the object block and correlation;

Step 605: utilize the adaptive template match selection to have two fallout predictors (mv_tm1 and mv_tm2) of less template SAD.

Step 607: based on the MVD standard of minimum, between mv_tm1 and mv_tm2, select final MVP, obtain MVD for current block.

Step 609:, utilize the conjecture strategy to judge whether to detect correct MVP based on current MVD; If can detect correct MVP, need not encode subsequently and send the position as index; Otherwise need encode subsequently and send 1 position as index;

Step 611: coding MVD and other information also need code index under the situation that can not detect correct MVP.Other information for example comprise the residual error of CBP, quantification etc.

Decoder-side

MVCATM (decoder):

Step 621: decoding MVD and other information begin to rebuild current block.Other information for example comprise the residual error of CBP, quantification etc.

Step 623, step 625 and step 627: with in the step 601,603 of coder side, 605 identical.

Step 629:, utilize the tactful judgement of conjecture can determine MVP based on the MVD of decoding; If which can't be determined is MVP, then be 1 of index decoding.

Step 631: utilize MVP and MVD to obtain the MV of current block.

Fig. 7 represents the block diagram of an embodiment of video coding system.The present invention carries out motion vector encoder (MVC) in estimation (ME) back.Motion vector will offer macro block, if this macro block is inter prediction (inter-prediction is called for short an Inter-Pred) type.If macro block is infra-frame prediction (Intra-Pred) type, then a motion vector does not need yet.Quantize the residual amount that residual amount provides quantification by the residual amount of conversion (T) conversion and by quantizing (Q).By entropy coding MVD is encoded with the residual amount of quantification and the index that needs to transmit, decoder is delivered in the output that will be used to transmit.On the residual amount that quantizes, carry out de-quantization (Q-1) and inverse transformation (T-1).The residual amount of rebuilding joins among the result of the prediction piece that ME obtains to provide reference picture.

Fig. 8 represents the block diagram of an embodiment of video decoding system.When the macro block of decoding was inter prediction (Inter-Pred) type, MVD further decoded.Decode bit stream to obtain the residual error of decoding by entropy.By the residual amount of de-quantization (Q-1) de-quantization decoding and by inverse transformation (T-1) inverse transformation to provide signal.In process of reconstruction, carry out the decoding (DMVC) of motion vector encoder method.If the conjecture strategy among the DMVC can't be determined motion vector predictor (MVP), then may need the further extra index of decoding.The MV that DMVC produces is used for motion compensation (MC).The result of MC adds the frame that signal provides reconstruction.

Fig. 9 has represented a kind of motion vector encoder device according to the embodiment of the invention, comprises with the lower part:

901, first select module, in order to select to form the fallout predictor of Candidate Set;

903, first processing module receives described first Candidate Set of selecting the fallout predictor of module, reduces the number of the fallout predictor in the described Candidate Set in order to utilize matching process;

905, second select module, be connected, receive the remaining predicted device in the described Candidate Set, based on standard selected optimum fallout predictor in described remaining predicted device of minimum movement phasor difference with described first processing module;

907, second processing module selects module to be connected with described second, obtains current motion vector difference according to the fallout predictor of the optimum of selecting;

909, detecting module is connected with first processing module with described second processing module, based on remaining fallout predictor that obtains from first processing module and current motion vector difference, utilizes the conjecture strategy to survey motion vector predictor;

911, judge module is connected with described detecting module, if the result of detection of detecting module is to detect correct motion vector predictor, then judged result is sent to coding module;

913, coding module receives described judged result, the current at least motion vector difference of encoding or current at least motion vector difference of coding and index.

Figure 10 has represented a kind of motion vector decoding apparatus of the embodiment of the invention more, comprises with the lower part:

1001, receiver module receives the motion vector difference of having encoded at least;

1003, first decoder module connects described receiver module, and according to the motion vector difference of having encoded at least that receives, decoding motion vectors is poor;

1005, first select module, in order to select to form the fallout predictor of Candidate Set;

1007, first processing module receives described first Candidate Set of selecting the fallout predictor of module, reduces the number of the fallout predictor in the described Candidate Set in order to utilize matching process;

Can 1009, judge module connects described first processing module and described first decoder module, based on the motion vector difference of remaining predicted device and decoding in the described Candidate Set, utilize the tactful judgement of conjecture determine motion vector predictor;

1011, second decoder module connects described judge module, according to described judged result decoding index;

1013, analysis module connects described first processing module and second decoder module, judges motion vector predictor according to the index of decoding;

1015, second processing module connects described judge module, described first decoder module and described analysis module, according to judged result, utilizes motion vector difference and definite motion vector predictor of having decoded of decoding to obtain motion vector; Perhaps utilize the motion vector difference of decoding and the motion vector predictor of having judged to obtain motion vector.

In general, the numerous embodiments of coding and decoding video and motion vector encoding and decoding can be implemented in multiple terminal equipment or subscriber equipment, and terminal equipment or subscriber equipment include but not limited to movably unit or the device that mobile phone and other radio communication devices, PDA(Personal Digital Assistant), portable and desktop computer, image/video device (as digital camera), audio-video (AV) equipment (as video player), game device, permission visit and the possible internet of browsing or Local Area Network device and these functions combine.

Embodiments of the invention can be implemented in the mode of software, hardware, applied logic or software, hardware and applied logic combination.Software, applied logic and/or hardware can retain on integrated circuit (IC) chip, module or the memory.If necessary, part software, hardware and/or applied logic can retain on the integrated circuit (IC) chip, part software, hardware and/or applied logic can retain on the module, and a part of software, hardware and/or applied logic can retain on the memory.In one example, applied logic, software or instruction set remain in any in the various conventional computer-readable mediums.In the context of the present invention, " computer-readable medium " can be any medium or device, and it comprises, storage, communication, propagation or move instruction uses for instruction execution system, device or equipment (as computer) or and instruction executive system, device or equipment use jointly.Computer-readable medium can comprise the computer-readable storage media, and it can be to contain or any medium of store instruction or device uses for instruction execution system, device or equipment (as computer) or and instruction executive system, device or equipment use jointly.

If necessary, difference in functionality discussed herein can be with order and/or the execution simultaneously that differs from one another.In addition, if necessary, the one or more of above-mentioned functions can be maybe can making up of choosing wantonly.

Although the many aspects of pointing out in the independent claims of the present invention, other aspects of the present invention comprise feature and combination thereof in the foregoing description and/or the dependent claims with feature of independent claims, and are not only the combination that clearly provides in the claim.

Should also be noted that herein that in above explanation one exemplary embodiment of the present invention these explanations should not regarded the meaning of restriction as.Can also make multiple conversion and improvement and not deviate from the scope of the present invention that claim limits.

Claims (15)

1. A motion vector coding method, is characterized in that, comprises the following steps: Step 1: defining a candidate set of predictors based on adaptive changes in motion vectors adjacent to the motion vector; Step 2: using a matching method to reduce the number of predictors in the candidate set to reduce the number of bits used by the index; Step 3: Select the best predictor among the remaining predictors in the candidate set based on the minimum motion vector difference criterion, and obtain the current motion vector difference; Step 4: Based on the difference between the residual predictor and the current motion vector, use a guessing strategy to detect the motion vector predictor to further save the number of bits used to encode the index; Step 5: Judging whether the correct motion vector predictor can be detected, if the correct motion vector predictor is detected, encode at least the current motion vector difference, if the correct motion vector predictor cannot be detected, encode at least the current motion vector The difference and index are encoded. 2. A motion vector coding method according to claim 1, wherein the matching method is a boundary matching method or an adaptive template matching method. 3. A kind of motion vector coding method according to claim 2, it is characterized in that, described boundary matching method is, a part of data corresponding to the predictor in described candidate set is pasted into current data space, obtains boundary matching error , removing one or more predictors with the largest boundary matching error in the candidate set. 4. A kind of motion vector encoding method according to claim 2, it is characterized in that, described adaptive template matching method is, based on the possible correlation between adaptive template and current data, determine the automatic Adapting the shape of the template; or determining the size of the adaptive template for the current data space based on consistency within the adaptive template. 5. A kind of motion vector encoding method according to claim 4, it is characterized in that, described adaptive template matching method also comprises using the displacement based on the corresponding displacement of predictor in candidate set, calculates the displacement in the template of current data space and reference picture The difference between the templates after the data space. 6. A motion vector coding method according to claim 4, characterized in that, in the adaptive template matching method, the width and shape of the adaptive template have been specified. 7. A kind of motion vector coding method according to claim 1, is characterized in that, after described step 5, if detect correct motion vector predictor, then send at least the current motion vector difference of encoding; If can't detect to the correct motion vector predictor, then send at least the encoded current motion vector difference and the encoded bit as an index. 8. A motion vector coding method according to claim 1, wherein the candidate set of predictors includes several proportional spatial and temporal predictors. 9. A kind of motion vector coding method according to claim 1, is characterized in that, in described step 1, the motion vector adjacent to described motion vector is time adjacent, space adjacent, Motion vectors adjacent in time and space or based on analysis strategies. 10. A kind of motion vector coding method according to claim 9, it is characterized in that, the guessing strategy in step 4, the number of described remaining predictor is at least 1, for each remaining predictor and current motion vector difference Generate an intermediate parameter, reselect the candidate predictor for each intermediate parameter based on the minimum motion vector difference criterion, if the reselected candidate predictor is the same as the residual predictor that generated the intermediate parameter, then decide the residual prediction is an optional predictor, and if the number of optional predictors is 1, the optional predictor is a correct motion vector predictor. 11. A method of decoding encoded motion vectors encoded by the method of claim 1, comprising the steps of: Step 1: Decoding the motion vector difference; Step 2: defining a candidate set of predictors based on adaptive changes in motion vectors adjacent to the motion vector; Step 3: using a matching method to reduce the number of predictors in the candidate set; Step 4: Based on the remaining predictors in the candidate set and the decoded motion vector difference, use a guessing strategy to determine whether the motion vector predictor can be determined; Step 5: If the motion vector predictor can be determined, use at least the decoded motion vector difference and the determined motion vector predictor to obtain the motion vector; if the motion vector predictor cannot be determined, then decode the index and determine the motion vector predictor, using At least the decoded motion vector difference and the determined motion vector predictor result in a motion vector. 12. A method for decoding encoded motion vectors according to claim 11, characterized in that the matching method is the same as the matching method used in encoding. 13. A method for decoding encoded motion vectors according to claim 12, wherein the matching method is a boundary matching method or an adaptive template matching method. 14. A motion vector encoding device, characterized in that it comprises the following parts: The first selection module is used to select the predictors that form the candidate set; The first processing module receives the candidate set of predictors of the first selection module, and uses a matching method to reduce the number of predictors in the candidate set; The second selection module is connected to the first processing module, receives the remaining predictors in the candidate set, and selects the optimal predictor among the remaining predictors based on the minimum motion vector difference criterion; The second processing module is connected to the second selection module, and obtains the current motion vector difference according to the selected optimal predictor; A detection module, connected to the second processing module and the first processing module, detects a motion vector predictor using a guessing strategy based on the received remaining predictors in the candidate set and the current motion vector difference; A judgment module, connected to the detection module, if the detection result of the detection module is that a correct motion vector predictor is detected, the judgment result is sent to the encoding module; The encoding module receives the determination result, and encodes at least the current motion vector difference or encodes at least the current motion vector difference and an index. 15. A motion vector decoding device, characterized in that it comprises the following parts: a receiving module, receiving at least the encoded motion vector difference; The first decoding module is connected to the receiving module, and decodes the motion vector difference according to the received at least coded motion vector difference; The first selection module is used to select the predictors that form the candidate set; The first processing module receives the candidate set of predictors of the first selection module, and uses a matching method to reduce the number of predictors in the candidate set; A judging module, connected to the first processing module and the first decoding module, based on the remaining predictors in the candidate set and the decoded motion vector difference, using a guessing strategy to judge whether the motion vector predictor can be determined; The second decoding module is connected to the judgment module, and decodes the index according to the judgment result; An analysis module, connected to the first processing module and the second decoding module, determines the motion vector predictor according to the decoded index; The second processing module is connected to the judgment module, the first decoding module and the analysis module, and according to the judgment result, obtains the motion vector by using the decoded motion vector difference and the decoded and determined motion vector predictor; or by using the decoded The motion vector difference and the determined motion vector predictor result in the motion vector.

Images