CN101090491A - Enhanced block-based motion estimation algorithms for video compression - Google Patents

Abstract

Method, systems and software are proposed for obtaining for blocks of a first image similar blocks of a second image (the ''reference image''). The blocks of the first image are processed sequentially, for each block trying out a number of candidate locations in the second image and evaluating a cost function for each. Each candidate location in the second image is displaced by a respective motion vector from the block of the first image. In a first aspect of the invention the cost function is a function of a predicted motion vector for future blocks of the first image (i.e. blocks of the first image which have not yet been processed). In a second aspect of the invention the motion vectors are given by location values which are not all whole pixel spacings, halves of the pixel spacing, or quarters of the pixel spacing.

Description

The block-based motion estimation algorithm that is used for the enhancing of video compression

Related application

The application's request in the priority of the U.S. Provisional Patent Application 60/691,181 of on June 17th, 2005 application, at this its integral body is incorporated herein by reference.

Technical field

The present invention relates generally to the method and system that is used for digital signal compression, coding and expression, and more exactly, the present invention relates to use the method and system of multiframe estimation (ME).The invention further relates to a kind of computer program, such as recording medium, the program command that carrying can be read by computing equipment is so that described computing equipment is carried out a method according to the present present invention.

Background technology

Owing to the huge size of using the raw digital video data of using (perhaps image sequence) by modern multimedia, must compress so that can transmit and store these data this data.Have many important video compression standards, comprise ISO/IEC MPEG-1, MPEG-2, MPEG-4 standard and ITU-T H.261, H.263, standard H.264.ISO/IEC MPEG-1/2/4 standard applies to show business widely with distribution film, the digital video broadcasting, digital video disk or the digital multi-purpose CD that comprise video compact disc or VCD (MPEG-1) or DVD (MPEG-2), recordable DVD (MPEG-2), digital video broadcasting, digital video broadcasting or DVB (MPEG-2), video request program or VOD (MPEG-2), high definition TV in US or HDTV (MPEG-2) or the like.The MPEG-4 standard is more advanced more than MPEG-2, can realize high-quality video under lower bit rate, this makes it be very suitable for the video stream in internet, digital wireless network (for example 3G network), MMS (Multimedia Message Service) (from the MMS standard of 3GPP) or the like.MPEG-4 is accepted by high definition DVD of future generation (HD-DVD) standard and MMS standard.The ITU-TH.261/3/4 standard design is used for low deferred telegram phonovision and video conferencing system.Early stage H.261 standard design is at p ^*64kbit/s is work down, p=1, and 2 ..., 31.H.263 the standard in later stage is extremely successful, be widely used in the advance TV conference system, and the video stream that is used for broadband network and wireless network, wherein wireless network is included in the MMS (Multimedia Message Service) (MMS) in 2.5G and 3G network and other networks.Newest standards, H.264 (being also referred to as MPEG-4 version 10, perhaps MPEG-4 AVC) is the video compression standard of current state-of-the-art.It is so powerful so that the ITU-T joint development in the framework of MPEG decision and Joint Video Team (JVT).H.264 new standard is called in ITU-T, and is known as MPEG-4 advanced video coding (MPEG-4AVC), and perhaps the MPEG-4 version 10.H.264 be used for HD-DVD standard, direct video broadcasting (DVB) standard and may be used for the MMS standard.Based on H.264, the current relevant criterion that is called audiovisual standard (AVS) of developing in China.AVS 1.0 is designed for high definition TV (HDTV).AVS-M is designed for to move and uses.H.264 have above MPEG-1/2/4 and reach the H.261/3 target and the subjective video quality of standard.Except that using 8 * 8 traditional DCT of integer 4 * 4 discrete cosine transforms (DCT) replacement, H.264[1] H.263 the basic coding class of algorithms is similar to or MPEG-4, and also have extra characteristics, comprise deblocking effect (in-loopdeblocking) filter in the inter-frame forecast mode of I frame, the multiple block size that is used for motion estimation/compensation and multiple reference frame, 1/4th pixel precisions that are used for estimation, the loop, context adaptive binary arithmetic coding (context adaptivebinary arithmetic coding), or the like.

Estimation is the core of most of video compression standards (such as MPEG-1/2/4 and H.261/3/4), its redundancy that employs one's time to the best advantage, so its performance directly influences compression efficiency, well as subjective video quality and the coding rate of video coding system.

In block matching motion is estimated (BMME), in ME, to the most general measurement of the distortion between current block and the reference block be absolute difference and (SAD), for a N * N piece, be defined as

$\mathrm{SAD}(\mathrm{mvx},\mathrm{mvy})=\underset{m=0,n=0}{\overset{N-1}{\mathrm{\Σ}}}|F_t(x+m,y+n)-F_{t-1}(x+m+\mathrm{mvx},y+n+\mathrm{mvy})|$

F wherein _tBe present frame, F _T-1Be standard frame, (mvx mvy) represents current motion vector (MV).For width=X, highly=Y, and the frame of block size=N * N, in hunting zone ± W, need the search point sum that SAD assesses to search the optimal motion vector is equaled:

$\left(\frac{X}{N}\right)\left(\frac{Y}{N}\right){(2W+1)}^2,$

For X=352, Y=288, under the situation of N=16 and W=32, it equals 1673100.This is the huge numerical value of consume significant computing capability in video encoder.Proposed many fast algorithms [2]-[9] and reduced the number of searching for point in ME, for example three steps were searched for (TSS) [11], 2D logarithm search [12], new three go on foot search (NTSS) [3], MVFAST[7], and PMVFAST[2].MVFAST and PMVFAST are better than first three algorithm significantly, because they use the intermediate value motion vector predictor to depart from ME as search center implementation center, have reduced the number of the position of MV coding thus by the smooth motion vector field.

(it is the important improvement to MVFAST and other fast algorithm to the PMVFAST algorithm, and thereby accepted by mpeg standard [10]) consider one group of MV fallout predictor at first, comprise intermediate value, zero point, the left side, top, top-right and previous common position (co-located) MV fallout predictor.Fig. 1 for example understands the position of current block, left piece, top block, upper right square, upper right square and right piece (it is " a following piece ", that is, processed piece after current block).It calculates the SAD cost of each prediction.In later development, PMVFAST is made amendment to calculate the SAD cost that RD (rate distortion) cost [13] substitutes the following cost function of use:

J(m，λ _motion)＝SAD(s，c(m))+λ _motion(R(m-p)) (1)

Wherein s is a raw video signal, and c is a reference video signal, and m is current MV, and p is the intermediate value MV fallout predictor of current block, λ _MotionBe the Lagrange multiplier, R (m-p) expression is used for the position of coded motion information.Next step in PMVFAST is the MV fallout predictor that selection has minimum cost, and comes big diamond search of number of executions (diamond search) or little diamond search according to the value of the minimum cost that obtains from the MV fallout predictor.

Independent still major issue is to use the sub-pix pr-set vector in definition current video coding standard, comprise half-pix, 1/4 pixel or possibility even 1/8 pixel motion vector, it provides the description more accurately to motion, and the PSNR gain of about 1dB of whole pixel motion estimation can be provided.Adopt half-pixel accuracy, motion vector can adopt equidistant positional value, such as 0.0,0.5,1.0,1.5,2.0 or the like.Adopt 1/4 pixel precision, motion vector can adopt such as 0.00,0.25,0.50,0.75,1.00,1.25,1.50,1.75,2.00 or the like and so on positional value.Adopt 1/8 pixel precision, motion vector can adopt such as 0.000,0.125,0.250,0.375,0.500,0.625,0.750,0.875,1.000,1.125,1.250,1.375,1.500,1.625,1.750,1.875,2.000 or the like and so on positional value.

As everyone knows, motion vector distribution trends towards off-centring, this means that motion vector trends towards approaching in the extreme (0,0).In Fig. 6 (a) this situation has been shown, it is depicted as (0,0) MV and uses the motion vector distribution in the Foreman sequence of search (FS) algorithm fully.In addition, shown in Fig. 6 (b), motion vector distribution is also to median prediction device (intermediate value MV) skew, and it is the intermediate value of motion vector left side piece, top block and upper right square shown in Figure 1.In addition, shown in Fig. 6 (c), motion vector also to the adjacent motion vectors in present frame (leftMV, topMV, topRightMV) and set motion vector (preMV) skew in the previous frame, shown in Fig. 6 (d).These can all be the fallout predictors of considering that is used for the motion vector of current vector, and they can be used to PMVFAST.

Summary of the invention

The present invention seeks to be provided for the new and useful technology of estimation, it is applicable to digital signal compression, coding and the method and system of expressing.

Particularly, the present invention manages to provide new and useful effective exercise estimation technique, its can for example be applied in MPEG-1, MPEG-2, MPEG-4, H.261, H.263, H.264 or in AVS or other associated video coding standard.

First aspect present invention is based on such realization: although the estimation of PMVFAST algorithm has advantage really with comparing in preceding technology, it is not best.On principle,, exist to make the minimized motion vector field { m of RD cost of entire frame on the whole for each frame in the video _Ij, i=0..M-1, j=0..N-1}:

$\mathrm{total}\_\mathrm{RD}\_\mathrm{Cost}=\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}[\mathrm{SAD}(s_{i,j},c\left(m_{i,j}\right))+{\mathrm{\λ}}_{i,j}\left(R(m_{i,j}-p_{i,j})\right)]---\left(2\right)$

Wherein (i j) is illustrated in (i, j) individual piece in the frame that comprises M * N piece.For fixing Q _p(it is a quantization parameter), λ _{I, j}=λ=constant, and p _{I, j}=median (m _{I, j-1}, m _{I-1, j}, m _{I-1, j+1}) (3)

Yet, consider that whole RD costs of entire frame need exponential computational complexity simultaneously, this is unpractiaca.Thereby PMVFAST and other algorithm known are each only considers the only RD cost of a piece, rather than whole pieces in the frame.

Particularly, MVFAST or PMVFAST do not have to consider that this causes the variation in intermediate value MV fallout predictor of next piece when derivation during with respect to the motion vector of current block.This can influence the level and smooth of whole motion vector field.

In general, a first aspect of the present invention defines by the cost of improving PMVFAST and motion predictor candidate's selection has proposed a kind of new ME algorithm.Particularly, (current block can be 16 * 16 for each current block of first image, 16 * 8,8 * 16,8 * 8,4 * 8,8 * 4,4 * 4 or other rectangle length, even non-rectangle), select similar of second image (reference picture) according to a cost function, described cost function comprises two:

(i) current block and similar dissimilarity measure (SAD for example, item SAE), and

(ii) as being item at least to the function of the prediction of the motion vector of the following piece of first image.

Particularly, the algorithm that is proposed improves the sports ground flatness and becomes possibility by comprising current intermediate value MV fallout predictor and comprise the estimation intermediate value MV fallout predictor of the encoding block in future (promptly being untreated so far), making.

Many variations of the present invention are possible.Particularly, described can have any size and Any shape.

Can there be a plurality of second images (being a plurality of reference values) and described search can be included in position candidate in all second images.

In addition, can carry out described new cost function for big zone and a plurality of sub-piece that employing is encoded by the defined coded sequence of coded digital that constitutes together in described first image.This a little need not have identical size or shape.

The present invention has on the other hand, and it can or independently use with the first aspect present invention combination.In general, second aspect present invention proposes when (current block can be 16 * 16 to first image, 16 * 8,8 * 16,8 * 8,4 * 8,8 * 4,4 * 4, or other rectangle size, when current block even non-rectangle) is encoded, use selected, as to have the positional value from a class value different, selected (i.e. assembly separately in two axis directions) motion vector to encode with employed value the known technology.

Consider a possible motion vector predictor: (0,0).And the routine techniques of whole pixel allows motion vector to adopt such as-2.0, the positional value of-1.0,0,1.0,2.0 grades, and second aspect present invention proposes to revise the one group of possible position value that approaches described fallout predictor.Near 0 positional value 1.0, we can use the another location value such as 0.85, thereby make admissible positional value will comprise-2.0 for ,-0.85,0,0.85,2.0 or the like.Its advantage is that on statistics, the motion vector trend approaches 0.And, thereby by selecting to approach more 0 position, we will be more near true motion vector, and thereby can provide the better motion compensation that can cause high compression efficiency.

Thereby, in particular expression of the present invention, can be chosen as at least one direction of principal axis and select one group of possible position value, thereby make them can't all write Lm, m=-... wherein, 2 ,-1,0,1,2..., and L is constant (1 pixel separation for example, 1/2 pixel separation, or 1/4 pixel separation); That is, positional value is uneven.Particularly, can be chosen as at least one direction of principal axis and select one group of possible position value, thereby make them can't all write m/n, m=-... wherein, 2 ,-1,0,1,2 ... and n is 1 or 2 power.

Notice that a second aspect of the present invention is not limited to chosen position value from one group of locus heterogeneous value; Compare with traditional positional value group, also be not limited to and only select to approach most two positional values of zero.As example, in another example of a second aspect of the present invention, positional value 2.0 becomes 1.9, thereby admissible positional value will comprise -1.9, -0.85, 0, 0.85, 1.9Deng.

Thereby in the interchangeable particular expression of second aspect present invention, (being at least one described direction) selects one group of possible position value, can write as LA to comprise _mOne or more positional values of m/n, m=-... wherein, 2 ,-1,0,1,2.., n are 1 or 2 power, L is constant (for example 1 pixel separation, 1/2 pixel separation, or 1/4 pixel separation), and A _mBe less than 1 but be 0.75 value (for different m values difference optionally) at least, more preferably be 0.80 at least, and most preferably be 0.85 at least.

We have found that A _mOptimum value depend on video.

An advantage of the specific embodiment of second aspect present invention is, the motion vector that they produced can adopt the form type code identical with conventional algorithm to encode, except the regular code of this positional value should be interpreted as respectively the employed possible position value of this embodiment.For example, if be by the employed positional value of specific embodiment -1.9, -0.85, 0, 0.85, 1.9Deng, then the regular code of positional value 1.0 should be interpreted as 0.85, and the regular code of positional value 2.0 should be interpreted as 1.9, etc.

Method according to a second aspect of the invention can may further comprise the steps: definition of search region, the a plurality of position candidate of definition in described region of search, described a plurality of position candidate comprise by the defined one group of a plurality of position of the new positional value of second aspect present invention.These positional values be from key position (for example, (0,0) motion vector position, motion vectors, etc.) the corresponding shift value of the position candidate located.For each candidate motion vector, our calculation cost function, described cost function are similarity measurement (SAD for example, functions SAE) between the piece at current block in first image and the described candidate motion vector place in second image.Alternatively, it can also be the function of following motion vector: described candidate motion vector, current motion vectors, and alternatively, as first aspect present invention, one or more future anticipation motion vectors.For example, alternatively, as first aspect present invention, can given this cost function.

Description of drawings

Now will only the embodiment of the invention be described with reference to the following drawings as example, wherein:

Fig. 1 shows current block, left piece, top block, upper right square, upper right square and right piece;

Fig. 2 shows the exemplary distribution of Diff, its be | m _{I, j+1}-p _{I, j+1}| with | m _{I, j}-p _{I, j+1}| between difference;

Fig. 3 a shows the way of search as the employed huge diamond search of first embodiment of the invention;

Fig. 3 b shows the way of search as the huge diamond search of employed modification in the first embodiment of the invention;

Fig. 4 shows the way of search as employed little diamond search in the first embodiment of the invention;

Fig. 5 a and 5b compare the flatness in the MV territory of PMVFAST and first embodiment of the invention;

Fig. 6 shows the motion vector distribution in the Foreman sequence of using search (FS) algorithm fully to obtain of the bottom-right MV (PreBottomRightMV) in the frame formerly to the MV that is disposed in (a) (0,0) MV, (b) intermediate value MV, (c) adjacent MV and the previous frame in the present frame, (d);

Fig. 7 is the flow chart of first embodiment of the invention.

Embodiment

First embodiment of the invention has adopted many features of PMVFAST algorithm, but be by consider several contiguous blocks rather than only piece come PMVFAST (and other existing algorithm) in addition improved.Equation (2) and (3) illustrate, and the selection of current block MV directly influences the RD cost of contiguous block, and described contiguous block comprises right piece (or (i, j+1) piece), lower-left square (or (i+1, j-1) piece), and lower block (or (i+1, j) piece).This is because current MV will influence the MV that predicts of these contiguous blocks, and so and then influences the optimum movement vector of those pieces.These are " future " pieces, because when handling current block, also they are not carried out estimation.We can't side by side calculate the optimum movement vector of these following pieces with current block, because we need calculate the optimum movement vector of all pieces in the entire frame simultaneously according to equation (2), this will be very complicated.

As an alternative, for evaluate right piece or the (we can increase by one for the RD cost function of the current block of equation (1) for i, the j+1) implication of the selection of the current MV of the current block on the piece:

R(|m _i，j+1-p _i，j+1|) (4)

Wherein, m _{I, j+1}Be based on the optimum movement vector of right piece of the current MV of current block, and p _{I, j+1}Be based on the intermediate value MV fallout predictor of right piece of the current MV of current block, promptly

p _i，j+1＝median(m _i，j，m _i-1，j+1，m _i-1，j+2) (5)

Yet, the m in the equation (4) _{I, j+1}Be unknown, because the estimation of right piece (following piece) is not also carried out.Yet we notice, | m _{I, j+1}-p _{I, j+1}| can pass through | m _{I, j}-p _{I, j+1}| approximate well.Allow Diff be | m _{I, j+1}-p _{I, j+1}|-| m _{I, j}-p _{I, j+1}|.We experimentize in proper order to many video measurements and study the distribution of Diff.Fig. 2 has shown the probability density function (probability-distribution function) of Foreman sequence, and as shown in Figure 2, typical results seriously is partial to zero.This means that these two amounts are essentially identical (identical under about 70% situation, and only differ 1 under about 23% situation) in most of situation.This means, | m _{I, j}-p _{I, j+1}| be | m _{I, j+1}-p _{I, j+1}| good approximate.Therefore, R (| m _{I, j+1}-p _{I, j+1}|) can by W*R (| m _{I, j}-p _{I, j+1|}) next being similar to, wherein W＞0.Equally, we can add addition Item in cost to lower-left square and lower block.

(i, j) piece allow medianMV represent to provide intermediate value MV fallout predictor by equation 3 for.Allow FmedianMV represent the following intermediate value MV fallout predictor (the intermediate value MV fallout predictor that is used for right piece) that provides by equation (5).Thereby FmedianMV is MV candidate's a function.At this, first embodiment is called as for that (i, j) piece " strengthening motion vectors field adaptable search technology " (E-PMVFAST).The step of this embodiment is as follows.

Each step of embodiment is as follows, is shown among Fig. 7.

To any candidate MV, as the cost of giving a definition.cost(MV)＝SAD+λ*[w*R(MV-medianMV)+(1-w)*R(MV-FmedianMV)]

(6)

1. calculate the cost of three motion vector predictor: (i) intermediate value MV fallout predictor (" medianMV "), the (ii) estimated motion vector of right piece (" futureMV "), it is defined as follows:

FutureMV ≡ median (TopMV, TopRightMV, TopRightRightMV) and (iii) from the MV fallout predictor of past piece (" pastMV "), it is among previous common position MV (" PreMV ") and the previous bottom-right MV (" PreBottomRightMV ") that is away from medianMV one, promptly

pastMV≡arg max _{MV∈{PreMv，PreBottomRightMV}}{abs(MV-medianMV)}

Note, item (ii) can by be used for another contiguous following piece (such as the lower left, the bottom, and/or bottom-right) estimated motion vector replenish or replace.

Shall also be noted that item (iii) in, previous bottom-right MV can be replenished or be replaced by the previous MV fallout predictor that is used for another contiguous block.

Notice that item is the independent aspects of shape cost invention (ii) and (iii).

If above any one MV fallout predictor unavailable border of frame (for example) is then skipped this fallout predictor.

2., then stop search and forward step 7 to if the minimum cost of motion vector predictor is less than threshold value T1.Otherwise the motion vector of selecting to have minimum cost is as currentMV (current MV) and forward next step to.Notice that the cost of 3 motion vector predictor can adopt predefined procedure to calculate that (for example medianMV then is futureMV, then be pastMV), and at any time,, the cost of any motion vector predictor may stop and forwarding to step 7 if less than certain threshold level, then searching for.

3. carry out an iteration of directed little diamond search around currentMV.The notion of the directed little diamond search of explained later.

4., then stop search and forward step 7 to if minimum cost is less than threshold value T2.Otherwise the motion vector of selecting to have minimum cost is as currentMV and forward next step to.

If (currentMV=medianMV) and current minimum cost less than threshold value T3, then carry out little diamond search and forward step 7 to.

6., then carry out big diamond search, shown in Fig. 3 (a) if video is not interlacing scan; Otherwise, the big diamond search of the modification of execution shown in Fig. 3 (b).In each step in these steps, to each index point assessment cost function of rhombus.

7. select MV with minimum cost.

In our experiment, find that the value about 0.8 of w is effective.

Explain the step of directed little diamond search now.Suppose that centerMV is the current search center, and MV1, MV2, MV3 and MV4 be four around the search point, as shown in Figure 4.For each MVi calculates R (MVi-medianMV).If R (MVi-medianMV)＜R (centerMV-medianMV) then calculates SAD and the cost of MVi.Otherwise, ignore this MVi.Selection has the MV of lowest costs as currentMV.Notice that the notion of directed square search is considered to new, and independent aspects of structure cost invention, it need not combine with the notion of using futureMV and carry out.

The step of the big diamond search of big diamond search and modification is identical, but described search is finished for all set of the point that illustrates respectively among Fig. 3 (a) and 3 (b).

We consider a plurality of possible variant of embodiment within the scope of the present invention now.

At first, notice that for different pieces, the weight coefficient w in cost function can be different.In addition, selectively, can be different for the different described w of MV candidate.In particular, whether identical with the X-axis component of described FmedianMV or Y-axis component the definition of w can depend on such as the MV candidate whether approach medianMV and/or futureMV or MV candidate's X-axis component or the Y-axis component situation.

In addition, cost function can be not restricted to the form of equation (6).It can be to comprise distortion measurement item (for example SAD, distortion quadratic sum (SSD), average deviation distortion (MAD), MSD etc.) and considered encode any function of item of necessary position of the motion vector of current block and some contiguous block (for example right piece, lower block, lower-left piece etc.).

In addition, in step 1, the definition of futureMV is not limited to the form that above step 1 provides.The replacements definition possible for two of futureMV is: and futureMV ≡ median (leftMV, TopRightMV, TopRightRightMV) and futureMV ≡ median (medianMV, TopRightMV, TopRightRightMV)

In addition, in the step 1 of above expression, farthest one of the list collection middle distance medianMV that pastMV is chosen as the possible MV (preMV and preBottomRlghtMV) in the frame formerly.Yet the MV that will consider tabulation can comprise MV possible more than two (preMV for example, preLeftMV, preRightMV, preTopMV, preTopLeftMV, preTopRightMV, preBottomMB, preBottomLeftMV, preBottomRightMV, etc.).In addition, from the MV of an above previous coding frame can be included in the described tabulation (if for example present frame is frame N, then described tabulation can comprise frame N-1, N-2, N-3 ...).If present frame is the B frame, then the tabulation of previous coding frame can comprise following P frame.

In addition, in step 1, selecting pastMV is distance reference MV (medianMV in step 1) possible MV farthest.Other also is fine with reference to MV, comprises leffMV or TopMV or TopRightMV or certain combination.Other method of selecting from the tabulation of possible MV also is possible.

In step 2, the cost of described 3 motion vector predictor obtains in according to certain predefine order.Possible predefine comprises in proper order:

A) medianMV is followed by futureMV, follows by pastMV

B) medianMV is followed by pastMV, follows by futureMV

C) futureMV is followed by medianMV, follows by pastMV

D) futureMV is followed by pastMV, follows by medianMV

E) pastMV is followed by medianMV, follows by futureMV

F) pastMV is followed by futureMV, follows by medianMV

In addition, though, in step 3, carried out an iteration of directed little diamond search, can use more than one iteration as above expression.

Analog result

We provide the analog result of embodiment E-PMVFAST now.Described embodiment is embedded into H.264 reference software JM9.3[13] in, and use various QP, video sequence, resolution and hunting zone that it is simulated.Table 1 (a-c) and 2 (a-c) illustrate some typical analog results.PSNR (ratio of peak signal and noise) changes and BR (bit rate) variation is with respect to the PSNR of search (FS) fully and the variation of bit rate.Analog result illustrates, and the bit rate of the E-PMVFAST that is proposed and PSNR trend towards with search and PMVFAST are similar fully, but on large-scale video sequence and bit rate, E-PMVFAST trends towards than PMVFAST fast about 40%.The key character of E-PMVFAST is that its motion vector field trends towards level and smooth in the extreme, thereby makes that motion vector can be than other fast moving algorithm for estimating moving of indicated object more exactly.

In Fig. 5 (a) and 5 (b), the image in left side illustrates the motion vector field that (as short-term) obtained by the PMVFAST algorithm, and the imaging on right side illustrates the motion vector of the identical imaging that obtains by described embodiment.The motion vector field of the E-PMVFAST motion vector field than PMVFAST significantly is more level and smooth, particularly in the zone of being irised out.For the video motion content in perception transform coding (perceptual trans-coding), rate controlled, multiple block-sized motion estimation, the multiple standards frame estimation etc. was classified, level and smooth sports ground was very useful.

Foreman CIF
									FS			PMVFAST			E-PMVFAST
QP	PSNR	BR	PSNR changes	BR changes	Speed improves	PSNR changes	BR changes	Speed improves
									24	38.61	10457168	-0.01	1.12％	269.90	-0.01	0.73％	345.62
26	37.25	7498832	-0.02	1.13％	29.61	-0.02	1.17％	337.49
									28	35.99	5397584	-0.02	1.53％	50.27	-0.03	1.21％	329.79
30	34.64	3804232	-0.02	1.27％	41.91	-0.04	1.20％	326.41
									32	33.34	2730064	-0.02	1.19％	35.26	-0.03	1.22％	321.96
34	32.25	2044176	-0.05	1.57％	230.48	-0.05	1.38％	325.01

36	31.70	1529848	-0.07	1.20％	226.76	-0.09	1.28％	325.47
									38	29.9	1184744	-0.05	1.75％	223.29	-0.06	0.99％	328.10
40	28.89	960056	-0.09	1.85％	220.98	-0.11	1.09％	334.25
									42	27.79	794312	-0.14	1.71％	218.94	-0.16	1.74％	341.62
44	26.7	672224	-0.17	1.58％	217.42	-0.22	0.89％	351.22
									On average			-0.06	1.45％	235.89	-0.07	1.17％	333.36

The analog result of form 1a-foreman CIF sequence

Coastguard CIF
									FS			PMVFAST			E-PMVFAST
QP	PSNR	BR	PSNR changes	BR changes	Speed improves	PSNR changes	BR changes	Speed improves
									24	37.32	19901336	0	0.02％	245.39	0	0.02％	328.81
26	35.77	15169632	-0.01	-0.03％	236.12	0	0.01％	319.07
									28	34.35	11403728	-0.01	-0.07％	232.24	-0.01	-0.03％	308.17
30	32.8	8097840	0	0.07％	229.89	0	0.04％	301.18
									32	31.34	5616880	0.01	0.23％	228.38	0	-0.04％	296.55
34	30.08	3913296	-0.01	-0.08％	227.20	-0.01	-0.23％	294.13
									36	28.75	2546328	-0.02	0.03％	225.88	-0.01	-0.04％	292.18
38	27.55	1684056	-0.03	0.48％	225.13	-0.03	0.35％	293.95
									40	26.55	1172448	-0.04	-0.01％	224.78	-0.04	0.15％	299.45
42	25.52	800328	-0.06	0.65％	225.22	-0.06	1.00％	308.85
									44	24.54	572224	-0.07	-0.10％	225.47	-0.06	0.39％	321.38
On average			-0.02	0.11％	229.61	-0.02	0.15％	305.79

The analog result of form 1b-Coastguard CIF sequence

Hall CIF
									FS			PMVFAST			E-PMVFAST
QP	PSNR	BR	PSNR changes	BR changes	Speed improves	PSNR changes	BR changes	Speed improves
									24	39.63	6954760	-0.01	0.26％	424.55	-0.01	0.18％	700.53
26	38.6	4539912	-0.02	0.17％	397.68	-0.02	0.43％	704.18
									28	37.58	2993752	-0.04	0.07％	370.17	-0.04	0.06％	686.71
30	36.36	2021712	-0.04	0.01％	340.96	-0.04	0.18％	674.18
									32	35.17	1411344	-0.03	-0.86％	321.03	-0.05	-0.40％	655.08

34	33.94	957600	0.02	0.40％	301.76	-0.02	0.44％	659.53
									36	32.68	677768	-0.04	0.12％	288.55	-0.06	-0.12％	642.68
38	31.4	493168	-0.05	-0.04％	276.97	-0.05	-0.38％	641.42
									40	30.05	379240	0.05	-0.45％	263.07	0	-0.48％	625.73
42	28.67	279008	-0.03	-0.99％	251.93	-0.07	-0.84％	624.02
									44	27.38	214544	-0.02	-1.32％	241.90	-0.04	-1.03％	635.41
On average			-0.02	-0.24％	316.23	-0.04	-0.18％	659.04

The analog result of form 1c-Hall CIF sequence

Foreman QCIF
									FS			PMVFAST			E-PMVFAST
QP	PSNR	BR	PSNR changes	BR changes	Speed improves	PSNR changes	BR changes	Speed improves
									24	38.32	2816360	0	0.68％	260.10	-0.01	-0.05％	356.91
26	36.85	2122432	-0.02	0.28％	249.94	-0.02	-0.42％	343.24
									28	35.47	1615528	-0.01	0.41％	242.29	-0.01	0.09％	329.89
30	34.01	1213792	-0.01	0.44％	235.40	-0.02	0.35％	322.04
									32	32.61	918272	-0.04	0.57％	231.38	-0.01	-0.13％	316.56
34	31.39	710976	-0.04	0.65％	228.06	-0.05	0.05％	314.51
									36	30.07	536152	-0.1	0.85％	225.33	-0.07	0.60％	311.05
38	28.81	416112	-0.04	0.31％	223.95	-0.1	-0.10％	310.02
									40	27.65	328288	-0.08	0.74％	222.89	-0.08	-0.30％	311.37
42	26.5	259936	-0.12	-0.39％	221.19	-0.1	0.47％	315.72
									44	25.36	207344	-0.18	1.43％	218.38	-0.25	-0.08％	321.03
On average			-0.06	0.54％	232.63	-0.07	0.04％	322.94

The analog result of form 2a-foreman QCIF sequence

Akiyo QCIF
									FS			PMVFAST			E-PMVFAST
QP	PSNR	BR	PSNR changes	BR changes	Speed improves	PSNR changes	BR changes	Speed improves
									24	40.81	530832	0.01	0.01％	483.45	0.01	0.01％	1143.12
26	39.42	391936	0	-0.38％	428.54	0	-0.38％	1017.49
									28	38	289096	0	-0.09％	389.95	-0.01	-0.25％	879.72

30	36.57	208488	-0.01	0.40％	352.74	-0.03	0.33％	779.84
									32	35.03	153776	-0.04	0.83％	312.72	-0.02	1.01％	699.16
34	33.66	118440	0.03	-0.05％	287.94	0.02	0.58％	644.63
									36	32.27	89056	0.01	1.11％	270.48	0.03	0.88％	621.79
38	31.09	70280	-0.01	-0.01％	254.10	-0.03	-0.31％	598.06
									40	30.13	57200	0	0.92％	245.76	-0.06	1.58％	612.01
42	28.77	47648	-0.02	0.22％	241.09	-0.02	0.15％	629.43
									44	27.68	40608	0.02	0.55％	240.05	0.01	-0.91％	669.52
On average			0.00	0.32％	318.80	-0.01	0.25％	754.07

The analog result of form 2b-Akiyo QCIF sequence

Coastguard QCIF
									FS			PMVFAST			E-PMVFAST
QP	PSNR	BR	PSNR changes	BR changes	Speed improves	PSNR changes	BR changes	Speed improves
									24	37	4456208	0	0.15％	251.59	0	0.09％	346.68
26	35.4	3298352	0	0.12％	238.53	0	0.08％	330.67
									28	33.92	2386152	-0.01	0.15％	234.60	-0.01	0.24％	316.97
30	32.33	1633736	-0.02	0.06％	232.80	0	0.01％	310.88
									32	30.86	1110088	0	0.22％	231.60	0	-0.17％	309.50
34	29.62	770656	-0.01	-0.38％	230.98	-0.01	-0.32％	310.31
									36	28.3	511000	-0.02	0.28％	230.30	-0.01	0.50％	312.49
38	27.06	354208	0.02	-0.47％	229.61	0.01	-0.57％	314.41
									40	26.08	256480	-0.08	-0.72％	229.39	-0.08	-1.08％	317.67
42	25.02	179528	-0.06	2.16％	228.04	-0.07	0.45％	325.54
									44	24	131856	-0.09	-0.05％	227030	-0.09	0.57％	335.34
On average			-0.02	0.14％	233.16	-0.02	-0.02％	320.95

The analog result of form 2c-Coastguard QCIF sequence

We turn to the second embodiment of the present invention now, and it shows a second aspect of the present invention.

As mentioned above, conventional integer pixel fully allows motion vector to adopt-2.0 ,-1.0,0,1.0,2.0 or the like positional value on each direction.In second embodiment of the invention, select to approach the possible position value of fallout predictor.Near 0 positional value, we can use (replacing 1.0) another location value, such as 0.85, thereby make the positional value of permission can comprise-2.0 for, -0.85, 0, 0.85, 2.0 or the like.This advantage is that the motion vector trend approaches 0 on statistics.Therefore, by selecting more the position near 0, we can approach real motion vector more, and thereby can provide the better motion compensation that can cause high compression efficiency.Similarly, can change other positional value.For example, positional value 2.0 can be changed into 1.9, thereby make the positional value that allows to comprise -1.9, -0.85, 0,0.85,1.9 or the like.Encoding motion vector position except 1.0 should be interpreted as 0.85, and 2.0 encoding motion vector position should be interpreted as beyond 1.9, and the advantage of the change that is proposed is to use same movement vector code.

Half-pixel accuracy allows motion vector to adopt such as 0.0,0.5,1.0,1.5,2.0 or the like and so on positional value.Our these positional values of proposed amendments, particularly those approach the positional value of fallout predictor.For the positional value 0.5 that is in close proximity to 0, we advise using a different value.For example, a kind of possibility is to use 0.4 to replace 0.5.In other words, positional value will comprise 0.0, 0.4, 1.0,1.5,2.0.Similarly, other positional value can be modified.For example, positional value 1.0 can change into 0.9, thereby makes this group reposition value will comprise 0.0,0.4,0.95,1.5,2.0 or the like.Equally, this can help to improve compression efficiency.Similarly, can revise other positional value to improve compression efficiency.Yet, change the higher computational efficiency that this position can cause both places of encoder significantly.Usually, the gain of most of compression efficiency comes from positional value changed into and approaches fallout predictor.

1/4 pixel precision allows motion vector to adopt such as 0.00,0.25,0.50,0.75,1.00 or the like and so on positional value.We can the location revision value, and those approach the positional value of fallout predictor especially.For example, we can be revised as 0.00,0.20 with them, and 0.47,0.73,0.99 etc.

Notice that the method that is proposed allows us to select the positional value of any amount between each integer position value.For example, between positional value 0 and 1, half-pixel accuracy is used 1 positional value { 0.5}, 3 positional values { 0.25,0.50 of 1/4 pixel precision use, 0.75}, and 1/8 pixel precision uses 7 positional values { 0.125,0.250,0.375,0.500,0.625,0.750,0.875}.The method that proposes allows us to select any N positional value between 0 and 1.For example, we can select the N=2 value such as 0.3 and 0.6.

Uneven sub-pix estimation that is proposed and compensation are not each zones that must be applied to each frame.On the contrary, some position can be incorporated in the head, whether it is unlocked or closes for each zone (for example sheet) of described frame of video with indication.In addition, it can directly apply to existing standard under situation about changing without any grammer, because can use identical motion vector code.

Adopt H.264 JM82 software that the uneven sub-pix estimation and the compensation that are proposed are simulated, and its result is shown in the above form, wherein QP represents quantization parameter.This simulation in x and y direction use location value (...-1 ,-0.75 ,-0.5 ,-0.15,0,0.15,0.5,0.75,1..).That is, compare, only revised-0.25 and the positional value at+0.25 place with the standard method of using 1/4 pixel space positional value.Except using new positional value, the described algorithm H.264 canonical algorithm with known in other respects is identical.As shown in the table, second embodiment has reduced bit rate significantly when realizing similar PSNR.Do not need to revise grammer H.264.

Sequence		QP＝20			QP＝24
										JM82	Recommended value	Difference	JMS2	Recommended value	Difference
Akiyo	PSNR	44.11	44.13	0.02	41.73	41.74	0.01
								Bit	3332792	3140080	-5.78％	1766112	1674296	-5.20％
	News	PSNR	42.79	42.83	0.04	40.19	40.22								0.03
Bit								6908656	6665160	-3.52％	4156840	4017968	-3.34％
		weather	PSNR	43	43.01	0.01	39.75							39.58	0.01
Bit	6929368							6803496	-1.82％	4425912	4351312	-1.69％
			Sean	PSNR	42.65	42.67	0.02						39.8	39.83	0.03
Bit	5009336	4755328						-5.07％	2734592	2585480	-5.45％

Sequence		QP＝28			QP＝32
										JM82	Recommended value	Difference	JM82	Recommended value	Difference
Akiyo	PSNR	39.33	39.34	0.01	36.66	36.69	0.03
								Bit	927791	904744	-2.48％	491032	497600	1.34％
	News	PSNR	37.6	37.62	0.02	34.68	34.68								0
Bit								2497384	2441216	-2.25％	1476680	1469640	-0.48％

weather	PSNR	36.29	36.29	0	32.88	32.89	0.01
								Bit	274184	272312	-0.80％	1660648	1665368	0.28％
	Sean	PSNR	36.94	36.95	0.01	33.92	33.91								-0.01
Bit								1497232	1433544	-4.25％	793552	778816	-1.86％

Although described above only is several embodiments of the present invention, many within the scope of the present invention variations all are possible.

For example, the description of the invention that more than provides is the piece of fixed size that is used for having the P frame of a reference frame.Yet the present invention can be applied to have the piece of multiple sub-block size, and described must be not non-overlapped.More than one reference frame can be arranged, and reference frame can be the past or in the future any with respect to present frame of video sequence.

For video, a pictorial element (pixel) can have one or more components, such as luminance component, redness, and green, blue (RGB) component, YUV component, YCrCb component, infrared component, X-ray or other component.Each component of pictorial element is the symbol that can be expressed as numeral, and described numeral can be natural number, integer, real number or even plural number.Under natural situation, they can be 12,8, or any other bit resolution.Though the pixel in video is the 2 dimension samples in the rectangularly-sampled grid that has and uniform sampling cycle, described sampling grid is not that to be necessary for rectangle and described sampling period be not that to be necessary for be uniform.

Industrial applicibility

Each embodiment of the present invention is suitable for by MPEG-1, MPEG-2, MPEG-4, H.261, H.263, H.264, AVS or other can be modified with comprise above it the associated video coding standard or method fast, low postpone and the low-cost software and hardware is realized.Possible application comprises digital video broadcasting (ground, satellite, wired), digital camera, numeral camcorder (camcorder), digital VTR, set-top box, PDA(Personal Digital Assistant), can use multimedia cellular phone (2.5G, more than 3G reaches), video conferencing system, video on-demand system, WLAN devices, bluetooth applications, the webserver, video flowing server in the low or bandwidth applications, video code interpreter (from a format conversion to another form), and other television communications system etc.

List of references

Below with reference to the disclosure of document in this comprehensive introducing:

[1]Joint Video Team of ITU-T and ISO/lEC JTC 1，“Draft ITU-TRecommendation and Final Draft international Standard of Joint VideoSpecification(ITU-T Rec.H.264|ISO/IEC 14496-10AVC)”，docunmentJVT-G050rl，may 2003

[2]A.M.Tourapis，O.C.Au and M.L.Liou，“Predictive MotionVector Field Adaptive Search Technique (PMVFAST)”，ISO/IECJTCl/SC29/WG11 MPEG2000，Noordwijkerhout，ML，March’2000

[3]R.LI，B.Zeng and M.L.Liou，“A new three-step searchalgorithm for block motion estimation”，On Circuits and Systems forVideo Technology，Vol4，no4，pp438-42，Aug’94

[4] Z.L.He and M.L.Liou, " A high performance fast search algorithmfor block matching motion estimation ", IEEE Trans.on Circuits andSystems for Vld﹠amp; O 2Technology, vol.7, no5, pp826-8, Oct ' 97

[5]A.M.Tourapis，O.C.Au，and M.L. Liou，“Fast MotionEstimation using Circular Zonal Search”，Proc.of SPIE Sym.Of VisualComm.& Imagg Processin，vol2，pp.1496-1504，Jan.25-27，‘99

[6]A，M.Tourapis，O.C.Au，M.L Liou，G.Shen，and I.Ahmad，“Optimizing the Mpeg-4 Encoder-Advanced Diamond Zonal Search”，inPros.of 2000 IEEE Inter.Sym.on Circuits and Systems，Geneva，Switzerland，May，2000

[7]K.K.Ma and P.I.Hosur，“Performance Report of Motion VectorField Adaptive Search Technique(MVFAST)”，in ISO/IECJTC1/SC29/WG11 MPEG99/m81，Noordwijkerhout，NL Mar’00

[8]A.M.Tourapis，O.C.Au，and M.L.Liou，“Fast Block-MatchingMotion Estimation using Predictive Motion Vector Field Adaptive SearchTechnique(PMVFAST)”，in ISO/IEC/JTC1/SC29/WG11 MPEG2000/M5866，Noordwijkerhout，NL，Mar’00

[9]Implementation Study Group，“Experimental conditiohs forevaluating encoder motion estimation algorithms”，in ISO/IEC JTC1/SC29/WG11 MPEG99/n3141，Hawaii，USA，Dec’99

[10]“MPEG-4 Optimization Model Version 1.0”，in ISO/IECJTC1/SC29/WG11 MPEG2000/n3324，Noordwijkerhout，NL，Mar’00

[11]T，Koga，K.linuma，A.Hirano，Y.lijima，and T.Ishlguro，“Motion compensated interframe coding for video conferencing” Proc.Nat. Telecommun. Conf.，New Orleans，LA，pp. G.5.3.1-G.5.3.5，Dec’81。

[12]J.R.Jain and A.K.Jain，“Displacement measurement and itsapplication in interframe image coding”，JEEE Trans.OnCommunications，vol.COM-29，pp.1799-808，Dec’81

[13]JVT reference software JM9.2 for JVT/H.264 FRext

Claims (32)

1, a kind of is the method that each in a series of block of pixels is selected the similar block of pixels separately of second image in first image, described method comprises, according to value for the cost function separately of each position candidate, described current block for described first image, select between a plurality of position candidate in described second image, described cost function comprises:

(A) similarity measurement between the block of pixels at current pixel piece in described first image and the position candidate place in described second image, and

(B) with the corresponding candidate motion vector of described position candidate, the motion vectors relevant with described current block and in described serial piece the function of the future anticipation motion vector of at least one piece of the described image after described current block.

2, the method for claim 1, wherein, described function comprises first, its expression is to the described candidate motion vector necessary figure place of encoding, and second, if its expression will be selected as described second image similar in the block of pixels at described position candidate place, to the described future anticipation motion vector necessary figure place of encoding.

3, a kind of for selecting the method for the similar block of pixels separately of second image according in a series of block of pixels in first image of raster scan order each, described method comprises, is described current block of described first image:

(i) the current motion vectors of the described current block of definition;

(ii) in described second image definition of search region as a plurality of position candidate in described second image, each position candidate is associated with separately candidate motion vector (MV), and described candidate motion vector is the relative displacement between the position of current block in position candidate and described first image in described second image;

(iii) for each candidate motion vector and at least one the following piece of described first image after described current block in described serial piece:

(a) the future anticipation motion vector of the described following piece of definition; And

(b) calculation cost function, it comprises:

(A) similarity measurement between the piece at current block in described first image and the position candidate place in described second image, and

(B) function of described candidate motion vector, described current motion vectors and described future anticipation motion vector; And

(iv) select and corresponding similar of the position candidate with minimum institute's calculation cost function.

4, method as claimed in claim 3, wherein, described function comprises first, its expression is to the described candidate motion vector necessary figure place of encoding, and second, if its expression will be selected as described similar block of pixels at the piece at described position candidate place, to the described future anticipation motion vector necessary figure place of encoding.

5, method as claimed in claim 4, wherein, described second expression is to the needed figure place of encoding of difference between described candidate motion vector and each the described future anticipation motion vector.

6, method as claimed in claim 4, wherein, described function is described first and described second weighted sum according to the value of weighting parameters.

7, method as claimed in claim 6, it comprises the step that obtains described weighting parameters.

8, method as claimed in claim 3, wherein, the function of the motion vector of at least one piece that described current motion vectors is described first image, wherein, for described at least one piece, obtained similar in described second image in advance, and the spatially contiguous described current block of described at least one piece.

9, method as claimed in claim 3, wherein, described current motion vectors is the function of the motion vector of at least one piece of arranging in pairs or groups mutually with described current block in the frame outside described first image.

10, method as claimed in claim 3, wherein, described current motion vectors be in the frame except with described current block space on the function of motion vector of at least one piece the adjacent first frame piece.

11, method as claimed in claim 3, wherein, described following piece in (iii) is the next piece according to raster scan order, and the intermediate value that is described candidate motion vector, upper right side and right top-right motion vector of the described future anticipation motion vector in (iii) (a).

12, method as claimed in claim 11, wherein, when determining that in step first cost in the described candidate motion vector is less than first threshold in (iii), do not calculate the cost function of remaining described candidate motion vector, and generation and described first candidate motion vector have one group of second candidate motion vector of predefined spatial relationship, and step (iv) before, be each the calculation cost function in described second candidate motion vector.

13, method as claimed in claim 12, wherein, described second set of candidate motion vectors is chosen as than the more approaching described current motion vectors of described first candidate motion vector, wherein, vector between the measurement of distance define according to the needed figure place of encoding of difference between to described vector.

14, method as claimed in claim 3, wherein, step (ii) in, at least one in the described candidate vector is the motion vector according to the following piece of the described image of described raster scan order.

15, method as claimed in claim 3, wherein, step (ii) in, one in the described candidate vector be to use based on the maximized criterion of distance that makes between selected candidate vector and the described current motion vectors, and select from one group of possible candidate vector.

16, wherein, there are a plurality of second images in method as claimed in claim 3, and described a plurality of position candidate comprise at least one position candidate in each described second image.

17, method as claimed in claim 3, wherein, described is the sub-piece in the big zone of described first image, described sub-piece is according to being encoded by the coded sequence of coded number definition.

18, each in a kind of a series of block of pixels that are used in first image is selected the system of block of pixels of each self similarity of second image, described system comprises processor, described processor is configured to sequentially handle the piece of described first image, and according to cost function, be each current block in the described serial piece of described first image, select between a plurality of position candidate in described second image, described cost function comprises:

(A) similarity measurement between the block of pixels at current pixel piece in described first image and the position candidate place in described second image, and

(B) with the corresponding candidate motion vector of described position candidate, the motion vectors relevant with described current block and in described serial piece the function of the future anticipation motion vector of at least one piece of the described image after described current block.

19, a kind of software product of computer system-readable, it makes that described computer system is that in a series of block of pixels in first image each is selected the block of pixels of each self similarity of second image, this is by according to cost function, for what select between a plurality of position candidate of the current block in the described serial piece of described first image in described second image to realize, described cost function comprises:

(A) similarity measurement between the block of pixels at current pixel piece in described first image and the position candidate place in described second image, and

(B) with the corresponding candidate motion vector of described position candidate, the motion vectors relevant with described current block and in described serial piece the function of the future anticipation motion vector of at least one piece of the described image after described current block.

20, a kind of for selecting the method for the similar block of pixels separately of second image according in a series of block of pixels in first image of raster scan order each, described method is included as the current block among described of described first image, value according to the cost function separately of each position candidate between a plurality of position candidate in described second image is selected, and at least one in the described candidate vector is the motion vectors according to the following piece of the described image of raster scan order.

21, a kind of is that in a series of block of pixels in first image each is selected the method for the similar block of pixels separately of second image, described method is included as the current block among described of described first image, value according to the cost function separately of each position candidate between a plurality of position candidate in described second image is selected, wherein, in the described candidate vector at least one is to use based on the maximized criterion of distance that makes between the current motion vectors of the described current block among selected candidate vector and described, and from one group of possible candidate vector, select.

22, a kind of is that in a series of block of pixels in first image each is selected the method for the similar block of pixels separately of second image, described method is included as the current block among described of described first image, value according to the cost function separately of each position candidate between a plurality of position candidate in described second image is selected, wherein, described second set of candidate motion vectors is selected by following process, and described process comprises:

The current predicted motion of the described current block in defining described;

Define first candidate motion vector, and

Define one group of second candidate motion vector, it is than the more approaching described current motion vectors of described first candidate motion vector, wherein, vector between the measurement of distance define according to the needed figure place of encoding of difference between to described vector.

23, a kind of is that in a series of block of pixels in first image each is selected the method for the similar block of pixels separately of second image, described method comprises: be current block in the described serial piece of described first image, select between a plurality of position candidate in described second image, described a plurality of position candidate comprises by in a lateral direction the defined one group of position candidate of two positional values separately, the positional value of at least one in the described direction is for making it all to be represented by Lm, wherein, m is an integer, and L is a constant.

24, method as claimed in claim 23, wherein, each in the described position candidate group is associated with separately motion vector by the relevant position value, and described motion vector is different from motion vectors.

25, method as claimed in claim 23, wherein, each in the described position candidate group is associated with separately motion vector by the relevant position value, and described motion vector is different from (0,0) motion vector.

26, a kind of is that in a series of block of pixels in first image each is selected the method for the similar block of pixels separately of second image, described method comprises: be the current block in the described serial piece of described first image, select between a plurality of position candidate in described second image, described a plurality of described position candidate comprises that described position candidate can be write DA by two defined one group of position candidate of positional value _mM/n, m=-... wherein, 2 ,-1,0,1,2..., n are 1 or 2 power, D is a constant, and at least one the value of m A _mBe value, and it is in 0.75 to 1.0 scope for all values of m less than 1.

27, method as claimed in claim 26, wherein, each in the described position candidate group is associated with separately motion vector by the relevant position value, and described motion vector is different from motion vectors.

28, method as claimed in claim 26, wherein, each in the described position candidate group is associated with separately motion vector by the relevant position value, and described motion vector is different from (0,0) motion vector.

29, method as claimed in claim 26, wherein, for all values of m, A _mValue be at least 0.85.

30, method as claimed in claim 26, wherein, for m＜-1 and m＞1, A _mEqual 1.

31, a kind of is that in a series of block of pixels in first image each is selected the system of the similar block of pixels separately of second image, described system comprises processor, described processor is configured to, be current block in the described serial piece of described first image, select between a plurality of position candidate in described second image, described a plurality of position candidate comprises by in a lateral direction the defined one group of position candidate of two positional values separately, the positional value of at least one in the described direction is for making it all to be represented by Lm, wherein, m is an integer, and L is a constant.

32, a kind of software product of computer system-readable, it makes that described computer system is that in a series of block of pixels in first image each is selected the system of the similar block of pixels separately of second image, this is by selecting to realize between a plurality of position candidate in described second image, described a plurality of position candidate comprises by in a lateral direction the defined one group of position candidate of two positional values separately, the positional value of at least one in the described direction is for making it all to be represented by Lm, wherein, m is an integer, and L is a constant.

Images