CN1297147C

CN1297147C - Prediction and Coding of Video Object Planes for Bidirectional Prediction in Interlaced Digital Video

Info

Publication number: CN1297147C
Application number: CN 98102468
Authority: CN
Inventors: 罗伯特·O·艾弗雷格; 陈学敏; 阿贾伊·卢特拉
Original assignee: General Instrument Corp
Current assignee: General Instruments Holding Co ltd; Motorola Mobility LLC
Priority date: 1997-03-07
Filing date: 1998-03-07
Publication date: 2007-01-24
Anticipated expiration: 2018-03-07
Also published as: CN1206298A

Abstract

A system for encoding digital video images of bidirectionally predicted video object planes (B_VOPs). For a co-located B-VOP macroblock of a future fixed picture, direct prediction is performed to generate a predicted macroblock by calculating four field motion vectors. The four field motion vectors and the reference field are determined based on (1) a deviation term (MVD) of the coding vector of the current macroblock, (2) two future fixed picture field motion vectors, (3) a reference field for the two field motion vectors of the co-located future fixed macroblock, and (4) the time gap in the field period between the current B-VOP field and the fixed field.

Description

Prediction and coding to the bi-directional predicted video object plane of interlaced digital video

Technical field

The present invention relates to a kind of method and device thereof, be used for predicting such as the digital video image of bi-directional predicted video object plane (B_VOPs) and encoding, the B-VOP and/or the reference picture that particularly wherein are used for B-VOP is encoded are interleaved codes.

Background technology

The present invention is specially adapted to various multimedia equipments, and be called " MPEG-4 Video Verification Model Version 8.0 " with the name that can supply this paper reference, Stockholm, MPEG-4 Verification Model (VM) 8.0 standards (MPEG-4VM 8.0) compatibility that illustrates in the ISO/IEC/JTC1/SC29/WG11/N1796 document of July 1997.Being called the Moving Picture Experts Group-2 that illustrates in " InformationTechnology-Generic Coding of Moving Pictures and AssociatedAudio, Recommendation H.262 " the ISO/IEC 13818-2 document on March 25th, 1994 for the name of this paper reference is the predecessor of this MPEG-4 standard.

MPEG-4 is a kind of coding standard, and it is the communication of digital audio-video data, and access and operation provide a kind of structure flexibly and open coding tools.These instruments can be supported numerous characteristics.The flexibility structure of MPEG-4 can be supported the various combinations of coding tools and at computer, the relevant function that needs in telecommunications and culture (TV with the film etc.) market, database navigation for example, information extraction, and intercommunication or the like mutually.

MPEG-4 provides the core technology of standard, can store transmission and control of video data in multimedia environment effectively.MPEG-4 can realize compression effectively, but the target scalability, but the scalability of room and time, and wrong elasticity.

MPEG-4 video VM coder/decoder (codec) is a kind of hybrid coder based on block and target with motion compensation.Utilize overlapping block motion compensation texture to be encoded with 8 * 8 discrete cosine transform (DCT).Represent target shape with a kind of grapheme, and adopt a kind of Content-based Arithmetic Encoding (CAE) algorithm or the DCT encoder of remodeling to encode, the two has all adopted time prediction.This encoder can be handled the sprite that obtains from computer curve.In concrete the application, also can use other coding methods such as wavelet and sprite coding.

The texture coding of motion compensation is a kind of known method of video coding, and it can be regarded as a kind of three step programs.The first step is a signal processing, comprising motion estimation and compensation (ME/MC) and a kind of two dimension (2-D) spatial alternation.The purpose of ME/MC and spatial alternation is in order to utilize the time in the video sequence and the advantage of space coordinates, under the restriction of complexity to quantizing and the rate distortion characteristic of entropy coding is optimized.The technology that generally is used at present ME/MC is the block coupling, and the most general space map system is DCT.

Yet, if MB itself is an interleaved code, and/or used the reference picture of interleaved code material impact will occur to the ME/MC of the macroblock among the B-VOP (MB).

Particularly need to have a kind of otherwise effective technique and provide motion vector (MV) prediction for the MB among the B-VOP.Needing a kind of otherwise effective technique in addition is used for the MB that encodes in the field of B-VOP is carried out the Direct Model coding.Also need an a kind of coded system determination procedure that is used for the MB of coding B-VOP in addition, so that select to produce the reference picture of efficient coding.

The invention provides a kind of above-mentioned and systems other advantages of having.

Summary of the invention

According to the invention provides a kind of method and apparatus, be used for the digital video image of bi-directional predicted video object plane (B-VOP) is encoded, for example a kind of present image (for example macroblock), particularly present image wherein and/or to be used for to the reference picture that present image is encoded be staggered (for example) coding.

According to a first aspect of the present invention, a kind of method for example has a macroblock (MB) of field, top and field, the end for bi-directional predicted current in a series of digital video images coded image provides Direct Model motion vector (MV).Past coded reference image with field, top and field, the end is determined with the coded reference image in a future with field, top and field, the end.Following image is to predict with image in the past, makes that the MV of forward direction in the field, top of future image is MVtop with reference to above-mentioned field, top or field, the end of passing by image.Field by reference comprises a kind of optimum Match MB, and it is used as the MB in the field, top of future image.

This MV is defined as " forward direction " MV, although because it with reference in the past image (time of just passing by), just the time of forward direction has been made prediction to the image in future according to the image in past.As a kind of memory, prediction direction can be considered to the rightabout of corresponding MV.

Equally, the forward motion vector MVbot of the field, the end of future image is with reference to the field, top or the field, the end of past image.By the corresponding forward direction MV in field with future image is calibrated to determine forward direction and reverse MV, so that the field, top and/or the field, the end of prediction present image.

Forward motion vector MVf especially for field, prediction present image top, top is according to formula MVf, top=(MVtop*TRB, top)/and TRD, top+MVD determines that MVD wherein is the incremental motion vector of a search domain, TRB, the top correspondence the field, top of present image and for the time slot between the field of the past image of MVtop reference, TRD, and the top correspondence the field, top of future image and for the time slot between the field of the past image of MVtop reference.This time slot may be relevant with the frame rate that image shows.

Similarly, be used to predict the forward motion vector MVf of field at the bottom of the present image, bot is according to formula MVf, and bot=(MVbot*TRB, bot)/TRD, bot+MVD determines, MVD wherein is an incremental motion vector, TRB, and the bot correspondence the field, the end of present image and for the time slot between the field of the past image of MVbot reference, TRD, the bot correspondence the field, the end of following MB and for the time slot between the field of the past MB of MVbot reference.

When incremental motion vector MVD=0, be used to predict the counter motion vector MVb of field, current MB top, top is according to formula MVb, top=((TRB, top-TRD, top) * MVtop)/TRD, top determines, or when MVD ≠ 0, MVb, top=MVf, top-MVtop.

When incremental motion vector MVD=0, be used to predict the counter motion vector MVb of field at the bottom of the current MB, bot is according to formula MVb, bot=((TRB, bot-TRD, bot) * MVbot)/TRD, bot determines, or in MVD ≠ 0 o'clock, MVb, bot=MVf, bot-MVbot.

Also has a corresponding decoder.

Provide a kind of method according to another aspect of the present invention, be used for selecting a kind of coded system at the field coding MB that a series of digital video MB push up field and field, the end for having of current prediction.This coded system can be a reverse manner, reference MB wherein is in after the current MB in the display order in time, forward manner, and reference MB wherein is in before the current MB in time, or average (for example two-way) mode, use forward and backward mean value therein with reference to MB.

This method comprises one for determining the forward direction summation SADforward of absolute differential errors with respect to the past with reference to the current MB of MB, the step of field, and this step correspondence the forward coding mode.SADforward, field represent that current MB and past are with reference to the pixel brightness value error between the optimum Match MB among the MB.Simultaneously also will be for respect to determining the reverse summation SADbackward of absolute differential errors with reference to the current MB of MB future, field, this step correspondence the phase-reversal coding mode.SADbackward, field represent current MB and following with reference to the pixel brightness value error between the optimum Match MB among the MB.

Also to determine the mean value summation SADaverage of absolute differential errors for current MB about past and following mean value with reference to MB, field, this step correspondence average coded system.SADaverage, field represent current MB and past and future with reference to the pixel brightness value error between the mean value MB of the optimum Match MB of MB.

Coded system is to select according to the principle of SAD minimum.In the coding mode selection program, can also introduce the bias term that constitutes MB quantity required in the various coded systems.

SADforward, field, SADbackward, field and SADaverage, field determine by each component project addition of Jiang Dingchang and field, the end.

The accompanying drawing summary

Fig. 1 is the schematic diagram according to a kind of video object plane of the present invention (VOP) Code And Decode program.

Fig. 2 is the block diagram according to an encoder of the present invention.

Fig. 3 represents to be used for a kind of interpolation scheme of half pixel retrieval.

Fig. 4 represents the direct coding mode according to the field, top of a kind of interleaved code B-VOP of the present invention.

Fig. 5 represents the direct coding mode according to the field, the end of a kind of interleaved code B-VOP of the present invention.

Fig. 6 is illustrated in according to the method that rearranges the pixel line in a kind of adaptive frame of the present invention/field prediction scheme.

Fig. 7 is the block diagram according to a decoder of the present invention.

Fig. 8 represents according to a kind of grand block hierarchical structure of the present invention.

Embodiment

Therefore, use term " VOP " to represent at random or nonrandom (being rectangle) graph data shape herein.Adopt technique known to obtain a kind of section markers, and its form and ITU-R 601 brightness datas are similar.Each pixel is counted as a certain zone that belongs in the frame of video.

Frame 105 and be provided for independently encoding function from the VOP data of frame 115.Specifically, VOP117,118 and 119 pass through shape, motion and texture coding respectively in encoder 137,138 and 139.In shape coding, encode with binary system and grey chroma color shape information.In motion encoded, shape information is encoded with the motion estimation value within the frame.Carry out the partial transformation of DCT one class in texture coding, therefrom obtain a kind of conversion coefficient, the length that can change coding is so that compression.

The VOP data of using a multiplexer (MUX) 140 to come assembly coding then are so that by transmitting in the channel 145.Or can be on recording medium with storage.The coding VOP data of separating reception with a demultiplexer (DEMUX) 150, thus the VOP117-119 that separates is decoded and recovers.Frame 155,165 and 175 is represented the VOP117 through decoding and recovery respectively, 118 and 119, and can independently operate with a synthesizer 160 with video library 170 interfaces.

Synthesizer 160 for example can be a personal computer, and it is placed in user's the room, and the user can edit the data that receive with it, thereby the image that needs is provided.For example, may comprise the different VOP178 (for example circle) of VOP previously stored and reception in user's the private video library 170.The user can replace square VOP117 to form a frame 185 with circle VOP178.Such frame 185 comprises the VOP118 and 119 and at the VOP178 of this locality storage of reception.

In another example, background VOP109 can be replaced by the background that the user selects.For example, when watching TV news broadcasting, broadcasting can be encoded into and for example VOP of news broadcasting studio or the like background separation.The user can select a kind of background, for example commodity price or the channel of Weather information from video library 170 or other TV programme.The user can resemble and operate the video editing personnel.

Video library 170 can also be stored the VOP that receives by channel 145, and can be by access to netwoks VOP and other vision facilitiess of Internet one class.In a word, this video conversation comprises the VOP of single VOP or order.

The VS of Fig. 1 coding and program procedural application be in many performance, and be commercial and educate purposes, and this plays comprising personal computer, environmental imaging, picture and text user interface, video conference, Internet purposes or the like.The present invention provides stronger ability in particular for the ME/MC function of (for example field mode) VOP with interleaved code.

Fig. 2 is the block diagram according to an encoder of the present invention.This encoder is applicable to the VOP (P-VOP) of predictive coding and the VOP (B-VOP) of alternating binary coding.

P-VOP can comprise many grand blocks (MB), and they can be encoded individually with frame mode or inter-frame mode.In adopting frame during (INTRA) coding, the coding of grand block (MB) is not with reference to other MB.When adopting interframe (INTER) coding, MB carries out differential coding according to the pattern that is called forward prediction with respect to the subsequent frame of transient state.The subsequent frame of transient state is called as anchor-frame or reference frame.Anchor-frame (for example VOP) must be P-VOP or I-VOP, and can not be B-VOP.That I-VOP comprises is inner (for example in-line coding) block, it is not the coding of prediction.

In forward prediction, the MB that one in current MB and anchor-frame retrieval is distinguished compares, and therefrom determines best coupling.The motion vector (MV) that is called as the correspondence of reverse MV is represented the displacement of current MB with respect to optimum Match MB.Can also use the advanced prediction pattern to P-VOP in addition, on 8 * 8 blocks rather than the last execution of 16 * 16MB motion compensation.In addition, frame P-VOP MB interior and interframe encode can encode with frame pattern or field mode.

B-VOP can use above-mentioned forward prediction mode and backward prediction with P-VOP with interrelating, and bi-directional predicted and Direct Model, these patterns all are the technology in the frame.According to the MPEG-4 VM 8.0 that does not also have at present to change, B-VOP does not use the MB of intraframe coding.Anchor-frame (for example VOP) must be P-VOP or I-VOP, and can not be B-VOP.

When backward prediction B-VOP, the MB that one in the transient state anchor-frame of current MB and front retrieval is distinguished compares, and therefrom determines the MB of optimum Match.The MV that is called as the correspondence of forward direction MV represents the displacement of current MB with respect to optimum Match MB.For bi-directional predicted B-VOP MB, the MB in one in the transient state anchor-frame of current MB and front and the transient state anchor-frame back retrieval district is compared, therefrom determine the MB of optimum Match.Forward direction and reverse MV represent the displacement of current MB with respect to optimum Match MB.In addition, from the MB of optimum Match, obtain a average image, in the coding of current MB, use.

For the Direct Model prediction of B-VOP,, just from 8 * 8 block, derive a MV if the MB that collects later at P-VOP has adopted 8 * 8 advanced prediction patterns.8 * 8 block MV among the P-VOP are linear scaled, so as to obtain among the B-VOP block a MV, and do not need to seek the block of optimum Match by retrieval.

Comprise 210, one motion estimation functions 220 of a shape coding device with 200 encoders of representing, a kind of motion compensation function 230, and a texture encoder 240, they receive separately from the video pixel data of terminal 205 inputs.Motion estimation function 220, motion compensation function 230, texture encoder 240 and shape coding device 210 also will receive from the VOP shape information of terminal 207 inputs, and for example a kind of MPEG-4 V parameter OP's is randomly shaped.If this parameter is 0, VOP is exactly a rectangle, does not use shape coding device 210 this moment.

The fixedly VOP function 250 of reorganization provides the fixedly VOP for a kind of reorganization of

motion estimation function

220 and 230 uses of motion compensation function.From deducting current VOP through the previous VOP of motion compensation, remaining part is encoded in texture encoder 240 in a subtracter 260.Texture encoder 240 is carried out DCT, and texture information (for example conversion coefficient) is offered multiplexer (MUX) 280.Texture encoder 240 also provides information to adder 270, with the output addition of itself and motion compensator 230, and is input to the VOP function 250 of previous reorganization.

Motion estimation function 220 provides movable information (being motion vector) to MUX 280, provides the shape information of representing the VOP shape to MUX 280 by shape coding function 210 simultaneously.MUX 280 provides corresponding multiplexed data stream to a buffer 290, so that communicate continuously by a data channel.

The pixel data that is input to encoder can have the form of YUV 4:2:0.Utilize the border of rectangle to represent VOP.The top-left coordinates of square boundary is become the immediate even number of the top-left coordinates that is not more than the center rectangle.Corresponding therewith, the top-left coordinates of the square boundary in the chromatic component is half of luminance component coordinate.

Fig. 3 represents to be used for a kind of interpolation scheme of half pixel retrieval.Motion estimation and motion compensation (ME/MC) comprise a block in the retrieval district of the block (for example current block) that makes current video frame and reference frame (for example Yu Ce block or with reference to block) usually and are complementary.For prediction (P) image encoded, be a frame of front with reference to block.For bi-directional predicted (B) image encoded, can use the prediction block in former frame and back one frame.The prediction block is exactly motion vector (MV) with respect to the displacement of current block, and it has the component of level (x) and vertical (y).The MV component on the occasion of expression prediction block be in current block the right side and below.

The different blocks of motion compensation are to constitute by the pixel value that deducts to pointwise the prediction block from current block.Then different blocks is carried out texture coding.The MV of the coding of different blocks and the texture information of coding are sent to decoder.Then by the different blocks of decoder according to MV quantification of increase on the block of prediction, half approximate current block of recombinating.The block of ME/MC can be the block (grand block) of 16 * 16 frames, 8 * 8 block, or 16 * 8 block.

The precision of MV is set at pixel half.Must adopt interpolation on anchor-frame, therefore, (i+x j+y) is restricted to half of an integer to be used for the p of x or y.Interpolation is carried out according to Fig. 3.The integer pixel location represents with symbol "+", for example the A among the figure, B, C and D.Half pixel location represents with circle, for example a, b, c and d.In the drawings as seen, a=A, b=(A+B) // 2, c=(A+C) // 2, d=(A+B+CD) // 4, wherein // represent circular boundary.Can be for the above-mentioned MPEG-4 VM8.0 of this paper reference and on July 21st, 1997, the common name that proposes was called the further details of having described this interpolation method in No. 08/897847, the U.S. Patent application of " Motion Estimation and Compensation of Video ObjectPlanes for Interlaced Digital Video " by people such as Eifrig.

Fig. 6 represents according to the pixel line that rearranges in a kind of adaptive frame of the present invention/field prediction scheme.First aspect according to the advanced prediction scheme, adopt adaptive technique to determine whether 16 * 16 pixels of current grand block (MB) need to carry out as usual the ME/MC coding, or should be divided into the block of four 8 * 8 pixels, wherein each 8 * 8 blocks are to carry out the ME/MC coding separately, or should use based on motion estimation the field, MB pixel line wherein is rearranged, and forms the line of the homologous field in two 16 * 8 blocks, and each 16 * 8 block is to carry out the ME/MC coding respectively.

Represent a field mode 16 * 16 grand blocks (MB) with 600.This MB comprises the line 602,604,606,608,610,612,614 and 616 of even number, and the line 603,605,607,609,611,613,615 and 617 of odd number.Even number and odd lines are interlocked, and constitute field, top and field, the end (or first and second) respectively.

If the pixel line in the image 600 is aligned to the brightness block with the field, just formed the MB that represents with 650.Pointed out the line 602-617 that rearranges with 645 arrows of representing.For example, the first-line even lines 602 as MB600 also is first line of MB 650.Even lines 604 is rearranged into second line of MB 650.Equally, even lines 606,608,610,612,614 and 616 are aligned to the 3rd to the 8th line of MB 650 respectively.So just formed one 16 * 8 luminance area 680 with even-numbered.Equally, form 16 * 8 zone 685 with odd lines 601,603,605,607,609,611,613,615 and 617.

Be used for selecting the determination procedure of MC pattern as follows for P-VOP.For the video of frame pattern, at first obtain the absolute difference sum (SAD) of single 16 * 16 MB, promptly SAD16 (MVx, MVy); And the absolute difference sum of four 8 * 8 blocks, promptly SAD8 (MVx1, MVy1), SAD8 (MVx2, MVy2), SAD8 (MVx3, MVy3) and SAD8 (MVx4, MVy4).If

Σ_{i = 1}^{4} SAD 8 (MVxi, MVyi) <

SAD 16 (MVx, MVy) - 129,

Just select 8 * 8 predictions; Otherwise just select 16 * 16 predictions.Constant " 129 " obtains from Nb/2+1, and Nb wherein is an opaque pixel number among the MB.For terleaved video, obtain SADtop (MVx_top, MVy_top) and SADbottom (MVx_bottom, MVy_bottom), wherein (MVx_top, MVy_top) and (MVx_bottom MVy_bottom) is the motion vector (MV) of top (even number) He Dichang (odd number).From the half-court sample retrieval, (for example be SADtop and SADbottom) then and select reference field with minimum SAD.

The decision of whole predictive mode is based on from a, the minimum value of selecting among b and the c:

(a)SAD16(MVx，MVy)，

(b) Σ_{i = 1}^{4} SAD 8 (MVxi, MVyi) + 129

And (c) SADtop (MV_x top, MVy_top)+SADbottom (MVx_bottom, MVy_bottom)+65.If (a) item is minimum, just use 16 * 16 predictions.If (b) item is minimum, just use 8 * 8 motion compensation (advanced prediction pattern).If (c) minimum is just used the field based on motion estimation.Constant " 65 " obtains from Nb/4+1.

If selected 8 * 8 predictions, have four MV for four 8 * 8 brightness blocks one, promptly each 8 * 8 block has a MV.Then by these four MV are averaged and with mean value divided by 2 MV that obtain two colourity blocks.Because the MV of each 8 * 8 brightness block has the precision of half pixel, the MV of colourity block may have 16 pixel values.Following table 1 has illustrated at the conversion to half pixel value of 16 pixel values of colourity MV.For example 0 to 2/16 be rounded to 0,3/16 to 13/16 and be rounded to 1/2,14/16 and 15/16 and be rounded to 2/2=1.

Table 1

1/1 6 pixel values	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1/1 6 pixel values	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	1/2 pixel value	0	0	0	1	1	1	1	1	1	1	1	1	1	1	2	2

For field prediction, per two 16 * 8 blocks have two MV.Luma prediction is to produce like this.The even lines of MB (being line 602,604,606,608,610,612,614 and 616) is to adopt the regulation of reference to limit with a top MV.Stipulated MV in frame coordinate, therefore, the vertical displacement of full pel is equivalent to the even-integer value of vertical MV coordinate, and represents half pixel vertical displacement with the odd number even number value.When adopting half pixel vertical missing, only there is the pixel in the same reference field to be combined in together.

The MV of two colourity blocks derives from (brightness) MV, and each component divided by 2, is rounded off then.By all fractional values being transformed into one and half pixel deviations horizontal component is rounded off.Vertical MV component is an integer, and with the colourity MV vertical component rounding of gained.If by two results that remove are non integer value, just it is rounded to adjacent odd integer.Vertical interpolation between the line that this odd integer value representative is same.

The second aspect of future anticipation technology is the overlapping MC that is used for the brightness block that has described in detail in people's such as the MPEG-4 VM 8.0 that mentions hereinbefore and Eifrig the application.

The specific coding technology of B-VOP below will be described.For the VOP of the such INTER of B-VOP coding, one has four kinds of predictive modes, Direct Model just, interpolation (for example average or two-way) pattern, reverse mode, and forward mode.The three kinds of patterns in back belong to indirect pattern.Single forward direction or backward prediction is also referred to as " folk prescription is to prediction ".In various patterns, determine the prediction block of B-VOP with different modes.In addition, the block of B-VOP and fixedly block can be progressively (for example frame by frame) coding or staggered (for example by) coding.

Can have the different MB that predicts with different mode among the B-VOP.Term " B-VOP " only represents that it can comprise bi-directional predicted block, but this is not necessary.In contrast, in P-VOP and I-VOP, do not use bi-directional predicted MB.

For the MB of indirect pattern B-VOP, MV is a differential coding.For the forward direction MV of forward direction and two-way mode and the reverse MV of reverse and two-way mode, " of the same type " (for example forward direction or the reverse) MV that is right after in delegation at that MB of current MB front is used as predicted value.Be right after in the situation of MB front in this situation and the grating order identical, and normally in order of transmission.Yet, if the grating order is different with order of transmission, just should use the MV that is right after in the order of transmission in the MB front, in order to avoid in decoder, need storage and reorganization MB and corresponding M V.

Use similar MV, and the hypothesis order of transmission is identical with the grating order, and the grating order is from left to right, from top to bottom, the predicted value of the forward direction MV of left adjacent MB as the forward direction MV of current MB among the B-VOP.Similarly, with the reverse MV of the adjacent MB in a left side predicted value as the reverse MV of current MB among the B-VOP.Use this predicted value that the MV of current MB is carried out differential coding then.That is to say that predicted value and the difference that is confirmed as between the MV of current MB are sent to decoder as a motion vector difference.In decoder by recovering and with PMV with differ from MV and determine the MV of current MB in the Calais mutually.

If current MB is positioned on the edge, left side of VOP, just the predicted value with current MB is set at zero.

For the B-VOP of interleaved code, in four MV altogether, each top has two relevant motion vectors with field, the end.In order of transmission, these four prediction MV represent field, reverse top and the field, the reverse end of field at the bottom of the forward direction top of previous fixedly MB and the forward direction and next fixedly MB.Current MB and forward direction MB and/or current MB and reverse MB are separated by one or more intermediate image, and these images are not used to the ME/MC coding of current MB.B-VOP does not comprise the MB of INTRA coding, and therefore, each MB among the B-VOP is the ME/MC coding.Forward direction and oppositely fixedly MB can be from a P-VOP or I-VOP, and can be to encode frame by frame or by the field.

Indirect pattern B-VOP MB for staggered has represented possible four kinds of motion vectors (PMV) in following table 2.First tabulation of table 2 shows that forecast function, secondary series represent the type of PMV.These PMV are used to different MB predictive modes in table 3 shown below.

Table 2

Forecast function	The PMV type
Forecast function	The PMV type	The field, top, forward	0
Field, the end, forward	1	The field, top, forward	0
Field, the end, forward	1	The field, top, backward	2
Field, the end, backward	3	The field, top, backward	2

Table 3

Grand block pattern	The PMV type of using
Grand block pattern	The PMV type of using	Frame, forward direction	0
Frame, oppositely	2	Frame, forward direction	0

Frame, two-way	0，2
Frame, two-way	0，2	, forward direction	0，1
, oppositely	2，3	, forward direction	0，1
, oppositely	2，3	, two- way	0，1，2，3

For example, table 3 expression is adopted a top forward direction (" 0 ") and field, end forward direction (" 1 ") motion vector to the preceding field mode MB with forward prediction mode (i.e. " field, forward direction ").

After being used, the motion vector of previous MB just become the PMV of next MB in the order of transmission in differential coding.In the PMV zero clearing of naming a person for a particular job that rises of each row of MB, because the MV of that MB at previous row end can not be identical with the MV of MB on the previous row starting point.In Direct Model MB, do not use predicted value yet.For the MB that skips, PMV keeps last value.

B-VOP MB for the Direct Model coding does not send the poor of vector.But in decoder, directly calculate forward direction and reverse MV, and proofread and correct with the single increment MV of not prediction according to temporal next P-VOP MB.Because the MV data that send are less, this mode is very effective.

The type of having summed up in table 4 according to previous and current MB is the situation of motion vector encoder with any PMV.The matrix pmv[of a motion vectors can be provided for B-VOP], matrix be expressed as from zero to three (be pmv[0], pmv[1], pmv[2] and pmv[3]).Sign pmv[] need not send, but decoder can be determined the pmv[of use according to Mv type of coding and decoded concrete vector] sign.After a B-VOP MB was encoded, some PMV vector was updated, and made its motion vector with current MB identical.Previous, preceding two or preceding four PMV upgrade according to the MV quantity about current MB.

For example, the MB of forward field prediction has two motion vectors, its pmv[1] be exactly field, the end, the PMV of forward direction.For the MB of counter field prediction, pmv[2] be exactly the field, top, reverse PMV, pmv[3] then be field, the end, reverse PMV.For the MB of two-way field prediction, pmv[0] be the field, top, the PMV of forward direction, pmv[1] be field, the end, the PMV of forward direction, pmv[2] be the field, top, reverse PMV, and pmv[3] be field, the end, reverse PMV.Frame pattern BVOP MB for forward direction or backward prediction has only a MV, and therefore, what be used for forward direction only has a pmv[0],, be used for the reverse pmv[2 that only has].Frame pattern BVOP MB for average (promptly two-way) prediction has two MV, specifically, pmv[0] be used for forward direction MV, pmv[2] be used for reverse MV.Being expressed as the MV that delegation's representative of " upgrade pmv[] " need upgrade is one, two, and still four.

Table 4-predictive mode sign vector pmv[]

	Current grand sub-block type
	Current grand sub-block type							The type of previous grand sub-piece in the order of transmission	Direction	Forward direction, frame pattern	Oppositely, frame pattern	On average, frame pattern	Forward direction, field mode	Oppositely, field mode	On average, field mode
The pmv[that upgrades]	No	0，1	2，3	0，1，2，3	0，1	2，3	0，1，2，3		Direction	Forward direction, frame pattern	Oppositely, frame pattern	On average, frame pattern	Forward direction, field mode	Oppositely, field mode	On average, field mode
The pmv[that upgrades]	No	0，1	2，3	0，1，2，3	0，1	2，3	0，1，2，3	The pmv[that uses]	No	0	2	0，2	0，1	2，3	0，1，2，3

Obviously, table 4 only is a kind of simplified way that is used for realizing the technology of the present invention, can be used for selecting a prediction MV into current MB.Yet this scheme can also realize with various approach.

The DC adaptive prediction of sub-interblock can use among the MPEG-4 VM 8.0 identical algorithms about dct_ type numerical value.The self adaptation AC prediction of sub-interblock can be carried out according to the mode of being counted among the MPEG-4 VM 8.0, and difference only is that first row coefficient is to duplicate from top coded sub-blocks.Just can use this operation when only having current MB and top sub-piece to have identical dct_ type numerical value.If dct_ type difference just can only realize the AC prediction by duplicating from first hurdle, the sub-piece left side.If there is not the sub-piece on the left side, just in the AC prediction, use zero.

Fig. 4 sign is according to the Direct Model coding of the field, top of interleaved code B-VOP of the present invention.If having the following still image of identical relative position (promptly being total to the position) with current MB is according to (1) 16 * 16 (frame) MB, (2) MB inside, or (3) 8 * 8 (advanced prediction) MB encodes, and current grand sub-piece (MB) just adopts progressive direct coding pattern.

If the following still image Mb of position encodes according to staggered MB altogether, the Direct Model prediction is staggered.If the SAD of biasing is the minimum value in all B-VOP MB predicted values, the coding of current MB just uses Direct Model.The Direct Model of interleaved code MB is respectively the field, top of current MB and the MB that field, the end forms prediction.Through four field motion vectors of the MB of two-way motion compensation (MV (i.e. a top forward direction, field, end forward direction, push up field-reversed, and the end, is field-reversed) be directly to calculate according to the MB of correspondence in the following still image.

Because reduced required retrieval work significantly, and reduced the traffic volume of MV data, this technology is effective.In case determined MV and reference field, just current MB be used as the MB of two-way field prediction.In the bit stream of field prediction MB, has only an incremental motion vector (using) for two fields.

The foundation of the top field prediction of current MB is field, the top MV of MB of following still image (it can be P-VOP, or I-VOP, MV=0 wherein) and the past reference field of previous still image, and it is to be selected by field, the top MV that correspondence does not fix MB.That is to say, with field, the top MB of that following still image of the position of current MB corresponding (promptly altogether position) with the field, top or the field, the end of still image have best coupling in the past.Then with the fixedly MB of this optimum Match MB as field, current MB top.Determine incremental motion vector MVD by detailed retrieval, be given in a MB according to MB and go up the MV that do not fix that is total to the position.

The motion vector that is used for the field, the end of current MB equally also is that the MV with the field, the end that does not fix MB on the relevant position determines, the latter is again with reference to the field, top of still image or the optimum Match MB in the field, the end in the past.

A top motion vector is used to constitute a MB predicted value, and it is (a) pixel of obtaining from the field, top that does not fix MB on the relevant position and (b) from that the two the mean value of pixel of obtaining of fixed field in the past for field, the top MV reference that does not fix MB on the relevant position.Equally, field, end motion vector is used to constitute a MB predicted value, the pixel that the field, the end that does not fix MB that it is (a) from the relevant position obtains and (b) from that the two the mean value of pixel of obtaining of fixed field in the past for field, the end MV reference that does not fix MB on the relevant position.

As shown in Figure 4, comprise among the current B-VOP MB 420 and push up field 430 and field, the end 425 that the past fixedly comprises field 410, top and field, the end 405 among the VOP MB 400, does not comprise field 450, top and field, the end 445 among the VOP MB 440 and fix.

Motion vector MVtop is the forward motion vector that does not fix the field, top 450 of MB 440, represents in the past optimum Match MB among the fixing MB 400 with it.Although MVtop is with reference to the image of front (promptly time) in the past, it is still a forward direction MV because do not fix VOP 440 and past fixedly VOP 400 to compare be forward direction in time.In this example, MVtop is with reference to the fixing field, the end 405 of VOP 400 of past, can certainly be with reference to its top 410.MVtop is the forward direction MV of the field, top of current MB, and MVb, top is the reverse MV of the field, top of current MB.Pixel data derives out from bi-directional predicted MB in decoder, in decoder to using MVb respectively, pixel data in future that top and MVf, top represent and the past still image is averaged, the nubbin that will send originally and this average image addition then.

The motion vector computation of field, top is as follows:

If MV=0,

MVf，top＝(TRB，top*MVtop)/TRD，top+MVD；

MVb, top=((TRB, top-TRD, top) * MVtop/TRD, I; And

If MVD ≠ 0,

MVb，top＝(MVf，top-MVtop)

MVD is the increment or the deviation of motion vector.Motion vector herein is the vector of two dimension.In addition, motion vector is a kind of integer half intensity picture element movement vector.Oblique line "/" representative is near the truncation of zero division of integer.In addition, not fixing VOP is a P-VOP of direct predictive mode forever.If not fixing is an I-VOP, MV is exactly zero, and can use 16 * 16 indirect pattern.TRB, top are the time gaps in the field between the field, top 430 of in the past reference field (i.e. top or field, the end) and current B-VOP 420, and the past reference field in this example refers to such an extent that be field, the end 405.TRD, top are that the time gap between the field 450 is pushed up in the past reference field (i.e. field, top or field, the end) and following reference, and the past reference field in this example refers to such an extent that be field, the end 405.

Fig. 5 represents according to the present invention the field, the end of the B-VOP of interleaved code to be carried out the situation of Direct Model coding.Interlace video signal source wherein can have the top in preceding or field, the end form the preceding.In Figure 4 and 5, represented field, end form the preceding.Identical numbering is identical with project among Fig. 4 among the figure.Motion vector MVbot herein is the forward motion vector that does not fix the field, the end 445 of grand block (MB) 440, and block 440 has been pointed out in the past the optimum Match MB among the fixing MB 400.In this example, MVbot is with reference to the fixing field, the end 405 of MB 400 of past, can certainly be with reference to its top 410.MVf, bot and MVb, bot are respectively forward direction and counter motion vector.

As follows with field, the end motion vector computation of a top motion vector parallel computation:

If MVD=0,

MVf，bot＝(TRB，bot*MVbot)/TRD，bot+MVD；

MVb, bot=((TRB, bot-TRD, bot) * MVbot/TRD, I; And

If MVD ≠ 0,

MVb，bot＝(MVf，bot-MVbot)

TRB, bot are the time gaps between the field, the end 425 of in the past reference field (i.e. top or field, the end) and current B-VOP 420, and the past reference field in this example refers to such an extent that be field, the end 405.TRD, bot are the time gaps between the field 445 at the bottom of in the past reference field (i.e. top or field, the end) and the following reference, and the past reference field in this example refers to such an extent that be field, the end 405.

For the example of Figure 4 and 5, TRB, top, TRD, top, TRB, bot and TRD, the calculating of bot is not only depended on and is worked as the front court, reference field and frame reference time, also to depend on current vision signal be top preceding or field, the end preceding.Specifically

TRD, top or TRD, bot=2* (TR future-TR goes over)+δ; And

TRB, top or TRB, bot=2* (TR current-TR go over)+δ;

In TR future wherein, the current and TR of TR is respectively future in the display order in the past, the frame number of current and past frame, and δ is the additive correction value by time gap between field given in the following mark 5.The unit of δ is the cycle of field.

For example, the following fixed field of " 1 " in first row last column expression is a top, and reference field be one at the bottom of the field.Situation just as shown in Figure 4.Field at the bottom of the following fixed field of " 1 " expression in secondary series last column is one, and reference field also is field, the end.Situation as shown in Figure 5.

Table 5-time adjustment, δ

Reference field		Field, the end is preceding		The field, top is preceding
Reference field		Field, the end is preceding		The field, top is preceding		Do not fix=push up	Do not fix=end	A top δ	Field, end δ	A top δ	Field, end δ
The top	The top	0	-1	0	1	Do not fix=push up	Do not fix=end	A top δ	Field, end δ	A top δ	Field, end δ
The top	The top	0	-1	0	1	The top	The end	0	0	0	0
The end	The top	1	-1	-1	1	The top	The end	0	0	0	0
The end	The top	1	-1	-1	1	The end	The end	1	0	-1	0

About coefficient coding, need a suitable coding mode determination program.As mentioned above, for B-VOP, the coding of MB can adopt (1) direct coding, (2) 16 * 16 motion compensation (comprising forward direction, reverse and average mode), or (3) motion compensation (comprising forward direction, reverse and average mode).When correspondence do not fix MB be frame or direct coding the time, to current MB adopt respectively frame or direct coding.

For the field motion compensation MB among the B-VOP, decide according to minimum brightness half pixel SAD and use forward direction with respect to the coding still image, reverse or average mode is encoded to MB.Specifically, seven biasing SAD projects are following calculating:

(1) SAD directly+b1, (2) SAD forward direction+b2, (3) SAD oppositely+b2,

(4) SAD average+b3, (5) SAD forward direction, field+b3,

(6) SAD is reverse, field+b3, and (7) SAD is average, field+b4,

Subscript is wherein represented Direct Model, propulsion prediction, counter motion prediction, average (for example interpolation or two-way) motion prediction, frame pattern (for example advance in the part) and field mode (for example local staggered).Above-mentioned field SAD (SAD forward direction for example,, SAD is reverse,, and SAD is average) be top and field, end SAD sum, have reference field and the motion vector of oneself separately.Specifically, SAD forward direction, field=SAD forward direction, top field+SAD forward direction, field, the end; SAD is reverse, and field=SAD is reverse, and field+SAD is reverse on the top, field, the end; And SAD is average, and field=SAD is average, and field+SAD is average on the top, field, the end.

SAD directly is best Direct Model prediction, the SAD forward direction is the best 16 * 16 predictions according to forward direction (past) reference, SAD oppositely is the best 16 * 16 predictions according to oppositely (future) reference, SAD on average is the best 16 * 16 predictions that the pixel one by one by average best forward direction and optimal reverse reference constitutes, the SAD forward direction, the field is the best field prediction according to forward direction (past) reference, SAD is reverse, the field is the best field prediction according to reverse (future) reference, and SAD is average, is the best field prediction that the pixel one by one by average best forward direction and optimal reverse reference constitutes.

Bi ' is the bias defined in the following table 6, is used for calculating the predictive mode of vector of need more doing more physical exercises.The fewer pattern of Direct Model and MV preferably.

Table 6

Pattern	The quantity of motion vector	bi	Biasing	Value
Pattern	The quantity of motion vector	bi	Biasing	Value	Directly	1	b1	-(Nb/2+1)	-129
Frame, forward direction	1	b2	0	0	Directly	1	b1	-(Nb/2+1)	-129
Frame, forward direction	1	b2	0	0	Frame, oppositely	1	b2	0	0
Frame, average	2	b3	(Nb/4+1)	65	Frame, oppositely	1	b2	0	0
Frame, average	2	b3	(Nb/4+1)	65	, forward direction	2	b3	(Nb/4+1)	65
, oppositely	2	b3	(Nb/4+1)	65	, forward direction	2	b3	(Nb/4+1)	65
, oppositely	2	b3	(Nb/4+1)	65	, average	4	b4	(Nb/2+1)	129

The negative bias of Direct Model is to be used for being used for step by step with existing that the MPEG-4 VM of vision signal is consistent, and may have many MB like this and be left in the basket.

Fig. 7 is the block diagram according to a decoder of the present invention.Unified these decoders with 700 signs can be used for receiving Fig. 2 the encoder transmission encoded data signal and decode.The motion vector of the vedio data of received code and differential coding (MV) data on terminal 740, and offer a demultiplexer (DEMUX) 742.The dct transform coefficient that the vedio data of coding normally is used as predictive error signal (just remaining) carries out differential coding.

When VOP had at random shape, decoded shape function 744 was recovered shape information by signal processing, and this information is provided for motion compensation function 750 and VOP recombination function 752.746 pairs of conversion coefficients of texture decoder function are carried out reverse DCT, therefrom recover residual, information.For the grand block (MB) of INTRA coding, pixel information is directly to recover, and offers VOP recombination function 752.

For the block and the MB of the INTER among B-VOP coding, the pixel information that offers VOP recombination function 752 from texture coding function 746 is represented the remnants between current MB and the reference picture.Reference picture can come free forward direction or reverse one of the MV appointment fixing pixel data of MB.Otherwise for the MB of interpolation (promptly average), reference picture is two mean values with reference to MB, just fixedly MB and mean value that does not fix MB of past.In this case, before the current MB pixel data of recovery, decoder must calculate average pixel data according to forward direction and reverse MV.

For the block and the MB of INTER coding, motion decoding function 748 needs to handle the MV data of coding, therefrom recovers the MV of difference, and provides it to the motion vector store 749 of motion compensation function 750 and RAM formation.Motion compensation function 750 receives this difference MV data, and determines reference motion vector (be predictor motion vector, or PMV) according to mode of the present invention.PMV determines according to coding mode (being forward direction, oppositely two-way, or directly).

In case motion compensation function 750 has been determined whole with reference to after MV and the difference MV addition with itself and current MB, the whole M V of current MB just can use.Therefore, motion compensation function 750 just can be recovered the anchor-frame best match data at this moment from the VOP memory 754 that RAM constitutes, calculate the average image as required, and the anchor-frame pixel data is offered the VOP recombination function of the current MB that is used for recombinating.

Recover or the best match data that calculates is added on the pixel remnants of VOP recombination function 752 again, therefrom obtain the current MB or the block of decoding.The block of reorganization is output as video output signals, and is provided for VOP memory 754, so that obtain new anchor-frame data.It should be noted that may the suitable video data buffer capacity of needs according to the order of the transmission of frame and image because the anchor-frame of B-VOP MB in picture order may be temporal future frame or.

Fig. 8 represents according to a MB pack arrangement of the present invention.This structure is applicable to B-VOP, and represents the data format that decoder receives.For the ease of expression, this bag one total four lines among the figure.This bag is actually and begins from top line that serial sends, and is from left to right in delegation.First row 810 comprises a first_shape_code, MVD_shape, CR, ST, and BAC.Second row 830 comprises a MODB and MBTYPE.The third line 850 comprises a CBPB, DQUANT, Interlaced_Information, MVDf, MVDb, and MVDB.Fourth line comprises a CODA, CBPBA, Alpha Block Data and Block Data.Above-mentioned each all limits according to MPEG-4 VM 8.0.

First_shape_code represents whether a MB is in the bounding box of a VOP.CR represents the transfer ratio of Binary Alpha Block.ST represents level or vertical scanning order.BAC represents a kind of code word of binary arithmetic operation.

Represent the MODB of MB pattern to be used to (not having uncared-for) MB of each coding among the B-VOP.If the indication of MODB, will produce poor motion vector (MBDf, MVDb, or MVDB) and CBPB.MBTYPE is used to refer to grand block type, and it can also be represented motion vector pattern (MVD) and quantize (DQUANT).For interleaving mode, each MB can have nearly four MV.MBTYPE is used to refer to type of coding, and just forward direction is oppositely two-way or direct.CBPB is the Coded Block Pattern that is used for the grand block of B-type.Definition and the CBPB of CBPBA are similar, and difference only is to have only four bits at most.DQUANT defines the variation of numerical value in the quantizer.

Field Interlaced_Information in the third line 850 represents that MB is the MB of an interleaved code, and the field MV reference data of coding mode or the block of current MB is provided to decoder.Decoder uses this information to calculate the MV of current MB.The field of Interlaced_Information can be stored in the MV memory 749 or other memories in the decoder, uses afterwards as required.

The InterlaceD_Information field can also comprise a kind of sign dct_type, and example is situation described with reference to Figure 6, and whether the field, top among the MB of its expression coding is different with staggered order with field, end pixel line.

Under the situation of VOP_prediction_type=10, use illustrated MB schichtenaufbau.If COD indicates the I-of up-to-date decoding or P-VOP are skipped (COD=" 1 ") MB, just skip that MB of position identical (promptly being total to the position) among the B-VOP.Therefore, in bit stream, do not comprise information.

MVDf be among the B-VOP with respect to the time the preceding with reference to the motion vector of the MB of VOP (I-or P-VOP).It comprises a code word that is used for the variable-length of horizontal component, and the back is a code word that is used for the variable-length of vertical component.For a field prediction value is the staggered B-VOP MB of " 1 " and the forward direction or the MBTYPE of interpolation, MVDf representative a pair of with the past fixedly VOP be the field motion vector (field, the end is in a back, field, top) of reference.

MVDb be among the B-VOP with respect to the time after the motion vector of a MB of reference VOP (I-or P-VOP).It comprises a code word that is used for the variable-length of horizontal component, and the back is a code word that is used for the variable-length of vertical component.For a field prediction value is the staggered B-VOP MB of " 1 " and the forward direction or the MBTYPE of interpolation, and the MVDb representative is a pair of not fix the field motion vector that VOP is reference (field, the end is in a back, top).

MVDB only just appears among the B-VOP when MODB and MBTYPE indicate Direct Model, and it comprises the variable length codewords that is used for horizontal component in each vector, and the back is the variable length codewords that is used for vertical component.The MVDB representative is used for proofreading and correct the incremental vector of B-VOP MB motion vector, and the latter is by calibration obtains to P-VOP MB motion vector.

CODA represents grey chroma color shape coding.

The form of Fig. 8 only is an example, and those skilled in the art can also be various other communication formats of the employing for information about of decoder on this basis.

The bitstream syntax and the MB stratificational grammar that use have in the present invention been described in people's such as MPEG-4 VM 8.0 and Eifrig above-mentioned application.

Therefore as can be seen, the present invention has been for the current MB among the B-VOP provides a kind of encoding scheme, be specially adapted to current MB and be a coding and/or fixedly MB be the situation of a coded system.This is the scheme that a kind of MB to the field coding carries out direct coding, also has a kind of coding determination procedure, and the minimum value of its use absolute difference project sum is selected a kind of pattern of the best.Also provide a kind of motion vectors (PMV) in addition, can also comprise forward direction and reverse PMV therein as required, and also can be used for the MB of frame coding for field, top and the field, the end of the current MB of field coding.

Although above be to describe of the present inventionly, apparent with reference to various specific embodiments, those skilled in the art still can realize various modifications and changes under the condition of the spirit and scope of the present invention that do not break away from claims and limited.

Claims

1, a kind of method of Direct Model motion vector of the current bi-directional predicted field image encoded with top and field, the end that is used for calculating a digital video image sequence may further comprise the steps:

Determine a reference picture and a reference picture with the following field coding that pushes up field and field, the end with past field coding of field, top and field, the end;

Wherein, use in the past image to predict future image, make the forward motion vector MV of field, top of future image _TopWith reference to the field, top in the image in the past and field, the end one of them, and make the forward motion vector MV of field, the end of future image _BotWith reference to the field, top of described image in the past and field, the end one of them; And

By forward motion vector calibration to the respective fields of future image, be identified for predicting wherein one of at least forward direction and counter motion vector of the field, top of present image and field, the end, wherein:

(A) according to formula (MV _Top* TR _{B, top})/TR _{D, top}+ MV _D, be identified for predicting the forward motion vector MV of the field, top of present image _{F, top}And

TR wherein _{B, top}Corresponding to the field, top of present image with for MV _TopTime slot between the field of the past image of reference, TR _{D, top}Corresponding to the field, top of future image with for MV _TopTime slot between the field of the past image of reference, and MV _DIt is an incremental motion vector;

(B) according to formula (MV _Bot* TR _{B, bot})/TR _{D, bot}+ MV _D, be identified for predicting the forward motion vector MV of the field, the end of present image _{F, bot}And

TR wherein _{B, bot}Corresponding to the field, the end of present image with for MV _BotTime slot between the field of the past image of reference, TR _{D, bot}Corresponding to the field, the end of future image with for MV _BotTime slot between the field of the past image of reference, and MV _DIt is an incremental motion vector;

(C) according to formula (a) MV _{B, top}=((TR _{B, top}-TR _{D, top}) * MV _Top)/TR _{D, top}(b) MV _{B, top}=MV _{F, top}-MV _TopOne of, be identified for predicting the counter motion vector MV of field, present image top _{B, top}

TR _{B, top}Corresponding to the field, top of present image with for MV _TopTime slot between the field of the past image of reference, TR _{D, top}Corresponding to the field, top of future image with for MV _TopTime slot between the field of the past image of reference, and MV _{F, top}It is the forward motion vector that is used to predict field, present image top; And

(D) according to formula (a) MV _{B, bot}=((TR _{B, bot}-TR _{D, bot}) * MV _Bot)/TR _{D, bot}(b) MV _{B, bot}=MV _{F, bot}-MV _BotOne of, be identified for predicting the counter motion vector MV of the field, the end of present image _{B, bot}And

TR wherein _{B, bot}Corresponding to the field, the end of present image with for MV _BotTime slot between the field of the past image of reference, TR _{D, bot}Corresponding to the field, the end of future image with for MV _BotTime slot between the field of the past image of reference, and MV _{F, bot}It is the forward motion vector that is used to predict field at the bottom of the present image.

2, according to the process of claim 1 wherein

Determine MV with division of integer with to zero truncation _{F, top}And

MV _TopAnd MV _BotIt all is integer half intensity picture element movement vector.

3, according to the method for claim 1 or 2, wherein:

TR _{B, top}And TR _{D, top}Comprise that the time proofreaies and correct, its explanation described when the front court image encoded be top preceding or field, the end preceding.

4, according to claim 1 method, wherein:

Determine MV with division of integer with to zero truncation _{F, bot}And

5, according to the process of claim 1 wherein:

TR _{B, bot}And TR _{D, bot}Comprise that the time proofreaies and correct, its explanation described when the front court coded image be top preceding or field, the end preceding.

6, according to the process of claim 1 wherein:

In (C), at incremental motion vector MV _DSelected described formula (a) at=0 o'clock, at MV _DSelected described formula (b) at ≠ 0 o'clock.

7, according to the process of claim 1 wherein:

In (D), at incremental motion vector MV _DSelected described formula (a) at=0 o'clock, at MV _DSelected described formula (b) at ≠ 0 o'clock.

8, a kind of decoder, be used for a macro block from the bit stream recovery one digital video macroblock sequence that receives with current field coding of field, top and field, the end, described current macro is to adopt the reference macroblock of the past field coding with top and field, the end and have top and the reference macroblock encoded in the following field of field, the end carries out bi-directional predictedly, and this decoder comprises:

Be used for recovering MV _TopAnd MV _BotDevice, MV _TopBe forward motion vector with reference to the field, top of the following macro block of one of the field, top of macro block in the past and field, the end, and MV _BotIt is forward motion vector with reference to the field, the end of the following macro block of one of the field, top of described macro block in the past and field, the end; And

By the forward motion vector of the field of corresponding future image is calibrated the field, top that is identified for predicting current macro and field, the end wherein one of at least forward direction and the device of counter motion vector, described device comprises:

(A) according to formula (MV _Top* TR _{B, top})/TR _{D, top}+ MV _DBe identified for predicting the forward motion vector MV of field, current macro top _{F, top}Device; And

TR wherein _{B, top}Corresponding to the field, top of preceding macro block with for MV _TopTime slot between the field of the past macro block of reference, TR _{D, top}Corresponding to the field, top of following macro block with for MV _TopTime slot between the field of the past macro block of reference, and MV _DIt is an incremental motion vector;

(B) according to formula (MV _Bot* TR _{B, bot})/TR _{D, bot}+ MV _DBe identified for predicting the forward motion vector MV of field at the bottom of the current macro _{F, bot}Device; And

TR wherein _{B, bot}Corresponding to the field, the end of current macro with for MV _BotTime slot between the field of the past macro block of reference, TR _{D, bot}Corresponding to the field, the end of following macro block with for MV _BotTime slot between the field of the past macro block of reference, and MV _DIt is an incremental motion vector;

(C) according to formula (a) MV _{B, top}=((TR _{B, top}-TR _{D, top}) * MV _Top)/TR _{D, top}(b) MV _{B, top}=MV _{F, top}-MV _TopOne of be identified for predicting the counter motion vector MV on current macro top _{B, top}Device; And

TR wherein _{B, top}Corresponding to the field, top of current macro with for MV _TopTime slot between the field of the past macro block of reference, TR _{D, top}Corresponding to the field, top of following macro block with for MV _TopTime slot between the field of the past macro block of reference, and MV _{F, top}It is the forward motion vector that is used to predict field, current macro top; And

(D) according to formula (a) MV _{B, bot}=((TR _{B, bot}-TR _{D, bot}) * MV _Bot)/TR _{D, bot}(b) MV _{B, bot}=MV _{F, bot}-MV _BotOne of be identified for predicting the counter motion vector MV on current macro top _{B, bot}Device;

TR wherein _{B, bot}Corresponding to the field, top of current macro with for MV _BotTime slot between the field of the past macro block of reference, TR _{D, bot}Corresponding to the field, top of following macro block with for MV _BotTime slot between the field of the past macro block of reference, and MV _{F, bot}It is the forward motion vector that is used to predict field, current macro top.

9, according to the decoder of claim 8, wherein:

Determine MV with division of integer with to zero truncation _{F, top}And

10, according to the decoder of claim 8, wherein:

TR _{B, top}And TR _{D, top}Comprise that the time proofreaies and correct, its explanation described when the front court coded image be top preceding or field, the end preceding.

11, according to the decoder of claim 8, wherein:

Determine MV with division of integer with to zero truncation _{F, bot}And

12, according to the decoder of claim 8, wherein:

13, according to the decoder of claim 8, it further comprises:

At incremental motion vector MV _DBe used to select the device of described formula (a) at=0 o'clock; And

At MV _DBe used to select the device of described formula (b) at ≠ 0 o'clock.

14, according to the decoder of claim 8, it further comprises:

At incremental motion vector MV _DBe used to select the device of described formula (a) at=0 o'clock; And at MV _DBe used to select the device of described formula (b) at ≠ 0 o'clock.