CN100544442C

CN100544442C - A kind of method that is applicable to the sheet stored of estimation

Info

Publication number: CN100544442C
Application number: CN 200710044624
Authority: CN
Inventors: 田野
Original assignee: Huaya Microelectronics Shanghai Inc
Current assignee: Zhangjiagang Kangdexin Optronics Material Co Ltd
Priority date: 2007-08-07
Filing date: 2007-08-07
Publication date: 2009-09-23
Anticipated expiration: 2027-08-07
Also published as: CN101102493A

Abstract

The invention discloses a kind of method that is applicable to the sheet stored of estimation, by space in the central column sheet, realize the reusing of the looking ahead of data, data and space, estimation and motion compensation, the girth of described annulated column and height all are to be unit with the pixel, described girth is EXT+2*SRX+16+2*SRX+EXT+H*16, and height is EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16; Wherein SRX is the hunting zone of directions X, and SRY is the hunting zone of Y direction, and EXT is the whole pixel count of the whole pixel upper and lower, left and right of optimum of 1/4 picture element interpolation needs.Adopt said method,, can reduce the speed of the number of times of mis-cache, can reduce memory bandwidth simultaneously, and reduce the memory space in the sheet, outside the sheet with the minimizing power consumption with the quickening data processing by with the central column space of data pre-fetching to above-mentioned structure.

Description

A kind of method that is applicable to the sheet stored of estimation

Technical field

The invention belongs to the video coding and decoding technology field, can pick up speed during particularly a kind of estimation, reduce bandwidth, save the method for the sheet stored of storage.

Background technology

Along with the development of information technology, the application of video more and more widely.Video monitoring, video conference, video request program, Digital Television, mobile TV, IPTV, video telephone, videodisc etc. have constituted a huge market.Coding and decoding video algorithm and standard be also in continuous progress, and AVS wherein and H.264 be domestic at present, the international two big standards of representing second generation coding and decoding video advanced technology has good market prospects.Because AVS and H.264 all adopted technology such as multi-reference frame, many sized blocks, 1/4 subpixel accuracy, hang down code check, high-quality while obtaining, also greatly increased the computation complexity and the memory bandwidth of estimation in the encoder.When estimation was realized by large scale integrated circuit, to the video that SD/the high definition motion amplitude is big, memory bandwidth became the main bottleneck of estimation.

Take exercises when estimating, what have can not be become and can be stored in the outer internal memory of sheet by 16 images of dividing exactly high or wide by the 16 images expansion limits of dividing exactly; The limit is expanded to whole pixel in the hunting zone according to estimation that has, and is stored in the outer internal memory of sheet; That have even sub-pixel that difference is come out and expand the limit also is stored in the outer internal memory of sheet; Direct internal memory reading of data outside the sheet of storage of reference frames during coupling.Operation like this, memory bandwidth is huge, real-time coding almost is impossible.The present method of existing multiple reduction memory bandwidth, its common ground all are that the pixel with reference frame is taken to internal memory in the sheet in advance by the outer internal memory of sheet.The search window to each macro block that has is all looked ahead once, although reduced the number of times of mistake buffering (mis-cache), has accelerated the speed of data processing, is a kind of huge waste to memory bandwidth.What have once reads to the macro block of several rows in the sheet, and this mobile video to low resolution is all right, then is impracticable to high-resolution SD/high-definition image, because the sheet stored that it occupies is too big.The technology of a lot of relevant memory bandwidth is only considered whole pixel motion estimation, to sub-pixel motion estimation, motion compensation is otherwise processed, this just need occupy extra memory bandwidth and sheet stored, and more than the sheet stored that sheet stored that the sub-pixel motion estimation that has occupies even whole pixel motion are estimated to occupy.

Summary of the invention

The object of the present invention is to provide a kind of method that is applicable to the sheet stored of estimation, can carry out corresponding configuration to different frameworks, to whole pixel motion estimation, sub-pixel motion estimation, motion compensation is unified handles, and can accelerate data processing speed better, reduce memory bandwidth, reduce the sheet stored.

In order to achieve the above object, technical scheme of the present invention is as follows:

A kind of method that is applicable to the sheet stored of estimation, by space in the central column sheet, realize the reusing of the looking ahead of data, data and space, estimation and motion compensation, the girth of described annulated column and height all are to be unit with the pixel, described girth is EXT+2*SRX+16+2*SRX+EXT+H*16, and height is EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16; Wherein SRX is the hunting zone of directions X, and SRY is the hunting zone of Y direction, and EXT is the whole pixel count of the whole pixel upper and lower, left and right of optimum of 1/4 picture element interpolation needs.

Adopt said method,, can reduce the speed of the number of times of mis-cache, can reduce memory bandwidth simultaneously, and reduce the memory space in the sheet, outside the sheet with the minimizing power consumption with the quickening data processing by with the central column space of data pre-fetching to above-mentioned structure.

Description of drawings

Fig. 1 has shown and has been recycled space in the central column sheet of reusing, and the vertical one by one macro block bar of at every turn looking ahead in the space.

Fig. 2 shows the rectangle that annulated column is launched into, the search window that whole pixel motion is estimated, whole pixel motion is estimated the relation between initial point, future position, the match point in the window, the window exterior pixel EXT that sub-pixel motion estimation and motion compensation are required, the H that the right side is reserved vertical macro block bar, bottom side are used for doing simultaneously the required V-1 of the capable macroblock coding of V laterally macro block bar.

When Fig. 3 A-3C shows delegation or two row macro block bars is encoded simultaneously, the kind situation of data pre-fetching when first macro block is mbx=0.Wherein Fig. 3 A shows the situation of initial several capable macro blocks; Fig. 3 B shows the situation of most macro-block line broad in the middle; Fig. 3 C shows the situation of last several capable macro-block line.

Fig. 4 A1-4C3 shows delegation or two row during macroblock codings, mbx! The situation of=0 o'clock data pre-fetching.Wherein Fig. 4 A1-4A3 shows the situation of initial several capable macro blocks; Fig. 4 B1-4B3 shows the situation of most macro-block line broad in the middle; Fig. 4 C1-4C3 shows the situation of last several capable macro-block line.When wherein Fig. 4 A1,4B1,4C1 show the most of macroblock coding of every interline, the situation that the macro block bar laterally is full of, when Fig. 4 A2,4B2,4C2 show a certain macroblock coding in every capable rear portion, the situation that the macro block bar laterally is not full of, when Fig. 4 A3,4B3,4C3 showed the most last several macroblock coding of every row, the macro block bar did not have the situation of prefetch data.

Fig. 5 A-5B shows the adjacent macroblocks coding and takes exercises when estimating, reuses the situation of data in the sheet, and Fig. 5 A is reusing of horizontal direction adjacent macroblocks, and Fig. 5 B is reusing of vertical direction adjacent macroblocks.

When Fig. 6 A-6B shows whole pixel motion and estimates coupling, the reusing of data between neighbor, Fig. 6 A is the reusing of data between the horizontal direction neighbor, Fig. 6 B is the reusing of data between the adjacent picture of vertical direction.

Fig. 7 shows the position of this device in video coding system, space in the central column sheet.Reference frame (reference frames) is stored in the outer internal memory of sheet, at every turn will required pixel prefetching to the central column sheet space (cache), utilize the data realization estimation (ME) in the space and the algorithm of motion compensation (MC).

Embodiment

According to Fig. 1 to Fig. 7, provide preferred embodiment of the present invention, and described in detail below, enable to understand better function of the present invention, characteristics.

The present invention has designed space in the central column sheet, realizing the reusing of the reusing of the looking ahead of data, data, space, whole pixel motion estimation, the estimation of sub-pixel, motion compensation, thereby reach pick up speed, reduce bandwidth, save the purpose of storage.

The space is such in the central column sheet: the girth=EXT+2*SRX+16+2*SRX+EXT+H*16 of annulated column, the height=EXT+2*SRY+16+2*SRY+EXT+ of annulated column (V-1) * 16,16th, macro block wide and high.SRX is the hunting zone of directions X, and SRY is the hunting zone of Y direction.EXT is the whole pixel whole pixel count up and down of optimum that 1/4 picture element interpolation needs, to AVS and h.264, and EXT=3.H＝1、2、3，V＝1、2。Referring to accompanying drawing 1: space, accompanying drawing 2 in the central column sheet: the rectangle that annulated column is launched into.

Variable H, the V in central column space, EXT, SRX, SRY, to different frameworks, even Tong standard can not done corresponding configuration.No matter to which type of framework, PC, DSP or VLSI, if mby increases 1 at every turn, promptly to delegation's macroblock coding, V=1 is if mby increases 2 at every turn, promptly simultaneously to two row macroblock codings, V=2.To SD V=1 or 2, to high definition, V=2, the meaning of V=2 is more effectively to reduce memory bandwidth.During V=1, to PC, DSP framework, H=1 shows that the data pre-fetching of the estimation of current macro and next macro block can executed in parallel; To the pipelined architecture of VLSI, H=2, purpose is that the pixel of the sub-pixel motion estimation of current macro and motion compensation is not override by the pixel that the whole pixel motion of next macro block is estimated, and whole pixel motion when estimating data prefetched good.During V=2, to PC, DSP framework, H=2, initial 2 macro blocks of the first row macro block are wanted continuous programming code in the two row macro blocks, and the estimation and the data pre-fetching executed in parallel of macro block subsequently; To the VLSI pipelined architecture, H=3, initial 3 macro blocks of the first row macro block are wanted continuous programming code in the two row macro blocks, and data will be looked ahead before the estimation.EXT disposes according to the difference of sub-pixel, to AVS and H.264, and EXT=3; To MPEG2 and MPEG4-2, EXT=1.SRX, SRY are according to the motion amplitude size of image and the rational distribution of resources of encoder, as SRX=SRY=16 or SRX=32, SRY=16.

Looking ahead of data is such: the wide and height of establishing piece image is respectively width and height, (mbx, mby) be an images be the coordinate of unit with the macro block, 0＜=mbx＜ceil (width/16) so, 0＜=mby＜ceil (height/16).When mbx=0, referring to accompanying drawing 3: the mbx=0 that looks ahead of data, divide A:0＜=mby＜ceil ((2*SRY+EXT)/16), B:floor ((2*SRY+EXT)/16)＜mby＜ceil ((height-(16+2*SRY+EXT+ (V-1) * 16))/16), C:mby〉three kinds of situations of floor ((height-(16+2*SRY+EXT+ (V-1) * 16))/16); Finish prefetching process by following three steps:

The first step, with the pixel prefetching of reference frame relevant position to the central column space, grey color part among the figure, this be a horizontal direction 16+2*SRX+EXT pixel, vertical direction EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16-(up_dy〉0? up_dy:bot_dy〉0? bot_dy:0) rectangle of individual pixel, up_dy=EXT+2*SRY-mby*16 wherein, bot_dy=mby*16+16+2*SRY+EXT-height.To situation A, up_dy〉0, what draw among the figure is the situation of mby=0; To case B, up_dy＜=0 and bot_dy＜=0; To clear shape C, bot_dy〉0, what draw among the figure is that mby=height/16-1 and height are by 16 situations that divide exactly.

Second step, each pixel with the first from left row of the gray area of looking ahead, EXT+2*SRX pixel write in parallel left expansion, this will form a horizontal direction EXT+2*SRX pixel, vertical direction EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16-(up_dy〉0? up_dy:bot_dy〉0? bot_dy:0) rectangle of individual pixel, up_dy, bot_dy are the same.

In the 3rd step, expand the top that is written to the central column space vertically upward with the lastrow pixel of looking ahead (comprising the pixel that expands left), or be written to the bottom in central column space with next line pixel (comprising the pixel that the expands left) expansion vertically downward of looking ahead.To situation A, form the rectangle of a horizontal direction EXT+2*SRX+16+2*SRX+EXT pixel, a vertical direction EXT+2*SRY-mby*16 pixel; To case B, the gray area upper and lower does not all have white space, and meet the demands this moment; To situation C, form the rectangle of a horizontal direction EXT+2*SRX+16+2*SRX+EXT pixel, a vertical direction mby*16+16+2*SRY+EXT-height pixel.

When mbx!=0 o'clock, referring to accompanying drawing 4: data look ahead mbx!=0, divide A1, A2, A3, B1, B2, B3, C1, C2, nine kinds of situations of C3, the condition of A, B, C is the same during with mbx=0, and 1,2,3 condition is respectively 1:0＜mbx＜ceil ((width-(2*SRX+EXT+16))/16); 2:floor ((width-(2*SRX+EXT+16))/16)＜mbx＜ceil ((width-(2*SRX+EXT))/16), 3:mbx〉floor ((width-(2*SRX+EXT))/16); Finish the prefetching process of data by following two steps:

The first step, relevant position from the reference frame prefetch data to the central column space, grey color part among the figure, this is that horizontal direction x pixel is wide, 0＜=x＜=16, vertical direction EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16-(up_dy〉0? up_dy:bot_dy〉0? bot_dy:0) the high pixel bars of individual pixel, up_dy, bot_dy are the

same.To situation

1,16 pixels of the every capable horizontal direction of pixel bars all are taken in the sheet outward by sheet in advance; To situation 2, the every capable horizontal direction of pixel bars only is prefetched to x0,0＜x0＜16 pixel, and then the pixel that every row is looked ahead is at last write x1 pixel by right expansion of horizontal direction, makes x0+x1=16; To situation 3, outside sheet, be not prefetched to data, then the rightest row pixel level of left side macro block bar is expanded to the right and be written to 16 pixels.After this step finishes, to every kind of situation will obtain all that one 16 pixel is wide, EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16-(up_dy〉0? up_dy:bot_dy〉0? bot_dy:0) the macro block bar that pixel is high.

Second step, 16 pixels of using the lastrow of macro block bar that a step obtains expand the top that is written to annulated column vertically upward, or are written to the bottom of annulated column with 16 pixels of next line expansion downwards of macro block bar.To situation A, upwards expand and write the macro block bar that 16 pixels of a horizontal direction of formation are wide, a vertical direction EXT+2*SRY-mby*16 pixel is high; To case B, do not need to expand and write; To situation C, expand downwards and write the macro block bar that 16 pixels of a horizontal direction of formation are wide, a vertical direction mby*16+16+2*SRY+EXT-height pixel is high.After this step finishes, just obtained that 16 pixels are wide, the macro block bar of EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16 pixel height (central column height).

How the process of top data pre-fetching has been described the pixel prefetching of sheet External Reference frame space in the central column sheet.No matter width, height can be divided exactly by 16, do not need to do any processing, all lie in concrete realization and have suffered, also do not put in order pixel according to the hunting zone and expand the limit, what look ahead only is whole pixel, and sub-pixel is not stored, this had both reduced storage, had also reduced the bandwidth of prefetch data.Reducing of memory bandwidth, the reusing of data, reusing of space also embodied in this process.Reducing of memory bandwidth mainly is to realize by reusing of data.

Reusing of data comprises two levels: the one, be taken to reusing of data in the sheet outward in advance by sheet, and the 2nd, by reusing of the data that are sent to processing unit in the sheet.The data reusing of first level, referring to Fig. 5: between adjacent macroblocks data reuse A: horizontal direction, B: vertical direction.Reusing of horizontal direction data, when occurring in left and right sides adjacent macroblocks and encoding successively, the scope of single macro block is: high is EXT+2*SRY+16+2*SRY+EXT, wide is the rectangle of EXT+2*SRX+16+2*SRX+EXT, just the grey color part among the figure adds one the 16 wide vertical macro block bar (white portion) of pixel, the data of ash color part (lap) do not need to look ahead once more, and this has just reduced memory bandwidth widely; Reusing of vertical orientation data, when occurring in V=2, the scope of single macro block also is the high EXT+2*SRY+16+2*SRY+EXT that is, wide is the rectangle of EXT+2*SRX+16+2*SRX+EXT, just the grey color part among the figure adds one the 16 high horizontal macro block bar (white portion) of pixel, the data of ash color part (lap) do not need to look ahead once more, and this has just more effectively reduced memory bandwidth; EXT in the space does the sub-pixel difference and uses, and individual processing does not repeat to look ahead, and is reused yet, and has further reduced bandwidth.

The data reusing of second level, referring to accompanying drawing six: between neighbor data reuse the grey part A: horizontal direction, B: vertical direction.The data reusing of this level is when occurring in data with space in the sheet and being sent to processing unit processes.When once mating, 256 reference pixels are in processing unit, if the pixel of next coupling is the neighbor of horizontal direction, then only need 16 pixels (the elongated informal voucher among the figure) of next-door neighbour reference macroblock right side (or left side) vertical direction are sent to processing unit; If the pixel of next coupling is the neighbor of vertical direction, then only need 16 pixels (the elongated informal voucher among the figure) of next-door neighbour's reference macroblock downside (or upside) parallel direction are sent to processing unit.So each coupling has 15 x 16 or 15 pixels of 16 x (the grey color part among the figure) to be reused, and this will reduce the bandwidth of the transmission data between interior internal memory of sheet and the processing unit, and accelerates the speed of data processing.

To the saving in space in the sheet, mainly be the realization of reusing by the space.Reusing of space is such: during mbx=H, the data of looking ahead are filled up whole central column space, the every increase by 1 of mbx, and the data of looking ahead will cover a vertical pixel bar to the right, and reusing like this of space realized.Referring to Fig. 1: space in the central column sheet, dotted line is separated out vertical pixel bar one by one.Illustrate: because the girth=EXT+2*SRX+16+2*SRX+EXT+H*16 of annulated column, figure image width=EXT+2*SRX+width+2*SRX+EXT in logic is to SD D1, width=720, height=576, if SRX=16, H=V=1, EXT=3, figure image width/annulated column girth=(3+2*16+720+2*16+3)/(3+2*16+16+2*16+3+16)=790/102=7.7 so, that is to say, every pair of delegation's macroblock coding, the central column space is reused 7 times at least; 576/16=36,36*7=252 that is to say a width of cloth D1 reference frame, takes exercises when estimating, the central column space is just reused 252 times.

Whole pixel motion estimates, referring to Fig. 2: the rectangle that annulated column is launched into, the search window among the figure, initial point (0,0), future position MVP (pmvx, pmvy), (mvx mvy) has reflected that whole pixel motion estimates the relation with the central column space to match point MV.Whole pixel motion estimate with MVP (pmvx, pmvy) ,-SRX＜=pmvx＜=SRX ,-SRY＜=pmvy＜=SRY are central point, with-SRX＜=mvx-pmvx＜SRX ,-SRY＜=mvy-pmvy＜SRY is search in the zone of hunting zone.The zone of search is 1/4th of a search window, can do full search, is suitable for the realization of large scale integrated circuit.The algorithm that whole pixel motion is estimated is realized in the central column space, and the data of region of search are sent to the Data Matching processing unit, is fit to the data reusing of second level.

Sub-pixel motion estimation around optimum whole pixel, will be carried out 1/2 pixel value difference earlier, obtains optimum 1/2 pixel; Carry out 1/4 pixel value difference around optimum 1/2 pixel again, obtain optimum 1/4 pixel.When the whole pixel of optimum (bmvx, bmvy) with the boundary point of search window (emvx, emvy) satisfy | bmvx-emvx|＜EXT or | during bmvy-emby|＜EXT, will use accompanying drawing 2: the data of EXT in the rectangle that annulated column is launched into.For saving sheet stored space, sub-pixel is not do too much storage, with using with inserting, uses promptly and throws away, if with inserting again.

Motion compensation, (mvx is when putting in order pixel, not need to do the sub-pixel difference mvy), and motion compensation can directly be obtained from search window as optimal motion vector MV; (mvx when mvy) being 1/2 pixel, does not need to do 1/4 sub-pixel difference, rationally inserts out 1/2 pixel value as motion compensation as MV; (mvx when mvy) being 1/4 pixel, does 1/2 pixel value difference earlier, does 1/4 pixel value difference again, inserts out the result of motion compensation as MV.The difference of sub-pixel is the same with the difference rule of sub-pixel motion estimation in the motion compensation.

Space in whole pixel motion estimation, sub-pixel motion estimation, the shared central column sheet of motion compensation is provided with grey module among the figure referring to Fig. 7 module.By looking ahead of data, processing unit is in time got data, has accelerated processing speed; By reusing of data, reduced memory bandwidth; By reusing of space, reduced the sheet stored.

The front provides the description to preferred embodiment, so that any technical staff in this area can use or utilize the present invention.To this preferred embodiment, those skilled in the art can make various modifications or conversion on the basis that does not break away from the principle of the invention.Should be appreciated that these modifications or conversion do not break away from protection scope of the present invention.

Claims

1, a kind of method that is applicable to the sheet stored of estimation, by space in the central column sheet, realize the reusing of the looking ahead of data, data and space, estimation and motion compensation, the girth of described annulated column and height all are to be unit with the pixel, described girth is EXT+2*SRX+16+2*SRX+EXT+H*16, and height is EXT+2*SRY+16+2*SRY+EXT+ (V-1) * 16; Wherein SRX is the hunting zone of directions X, SRY is the hunting zone of Y direction, EXT is the whole pixel count of the whole pixel upper and lower, left and right of optimum of 1/4 picture element interpolation needs, H represents the number of the vertical macro block bar of the rectangular right side reservation that annulated column is launched into, V represents that rectangular bottom side that annulated column is launched into is used for doing simultaneously the line number of macroblock coding, 16 expression macro blocks wide and high.

2, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 1 is characterized in that, for the AVS and the coded system of standard h.264, EXT=3, H=1 or 2 or 3, V=1 or 2.

3, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 1 or 2 is characterized in that, when piece image be the abscissa of unit when equalling zero with the macro block, it is as follows to finish the step of looking ahead:

Relevant position with the pixel prefetching of reference frame space to the central column sheet;

EXT+2*SRX pixel write in the parallel left expansion of pixel with the first from left row of looking ahead;

Expand the top that is written to space in the central column sheet vertically upward or expand the bottom that is written to space in the central column sheet vertically downward with the lastrow pixel of looking ahead with the next line pixel of looking ahead.

4, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 3, it is characterized in that, when piece image be the abscissa of unit when equalling zero with the macro block, be divided into 0＜=mby＜ceil ((2*SRY+EXT)/16), floor ((2*SRY+EXT)/16)＜mby＜ceil ((height-(16+2*SRY+EXT+ (V-1) * 16))/16), mby〉three kinds of situations of floor ((height-(16+2*SRY+EXT+ (V-1) * 16))/16); Wherein mby be piece image be the ordinate of unit with the macro block, height is the height of image.

5, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 1 or 2 is characterized in that, when piece image be the abscissa of unit when being not equal to zero with the macro block, it is as follows to finish the step of looking ahead:

The wide vertical macroblocks bar of corresponding 16 pixels in space in from the reference frame prefetch data to the central column sheet, the pixel level of then this row being looked ahead at last less than 16 pixels expands to the right and is written to 16 pixels if every row is looked ahead, and is written to 16 pixels if pixel of every row is all looked ahead less than then the rightest row pixel level of left side macro block bar being expanded;

Expand the top that is written to annulated column vertically upward or expand the bottom that is written to annulated column downwards with 16 pixels of the lastrow of the macro block bar of looking ahead with 16 pixels of next line of the macro block bar of looking ahead.

6, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 5, it is characterized in that, when piece image be the abscissa of unit when being not equal to zero with the macro block, be divided into 0＜=mby＜ceil ((2*SRY+EXT)/16) and 0＜mbx＜ceil ((width-(2*SRX+EXT+16))/16), 0＜=mby＜ceil ((2*SRY+EXT)/16) and floor ((width-(2*SRX+EXT+16))/16)＜mbx＜ceil ((width-(2*SRX+EXT))/16), 0＜=mby＜ceil ((2*SRY+EXT)/16) and mbx〉floor ((width-(2*SRX+EXT))/16), floor ((2*SRY+EXT)/16)＜mby＜ceil ((height-(16+2*SRY+EXT+ (V-1) * 16))/16) and 0＜mbx＜ceil ((width-(2*SRX+EXT+16))/16), floor ((2*SRY+EXT)/16)＜mby＜ceil ((height-(16+2*SRY+EXT+ (V-1) * 16))/16) and floor ((width-(2*SRX+EXT+16))/16)＜mbx＜ceil ((width-(2*SRX+EXT))/16), floor ((2*SRY+EXT)/16)＜mby＜ceil ((height-(16+2*SRY+EXT+ (V-1) * 16))/16) and mbx〉floor ((width-(2*SRX+EXT))/16), mby〉floor ((height-(16+2*SRY+EXT+ (V-1) * 16))/16) and 0＜mbx＜ceil ((width-(2*SRX+EXT+16))/16), mby〉floor ((height-(16+2*SRY+EXT+ (V-1) * 16))/16) and floor ((width-(2*SRX+EXT+16))/16)＜mbx＜ceil ((width-(2*SRX+EXT))/16), mby〉floor ((height-(16+2*SRY+EXT+ (V-1) * 16))/16) and mbx〉nine kinds of situations of floor ((width-(2*SRX+EXT))/16); Wherein mby be piece image be the ordinate of unit with the macro block, mbx be piece image be the abscissa of unit with the macro block, height is the height of image, width is the wide of image.

7, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 1 or 2 is characterized in that, the reusing of data comprises by sheet and be taken to reusing of data and reusing by the data that are sent to processing unit in the sheet in the sheet outward in advance.

8, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 7 is characterized in that, when adjacent macroblocks is encoded, takes place to be taken to reusing of data in the sheet outward in advance by sheet; Left and right sides adjacent macroblocks coding has the individual pixel of (EXT+2*SRX+16+2*SRX+EXT-16) x (EXT+2*SRY+16+2*SRY+EXT) to reuse; During neighbouring macroblock coding, there is the individual pixel of (EXT+2*SRX+16+2*SRX+EXT6) x (EXT+2*SRY+16+2*SRY+EXT-16) to reuse.

9, as claim 7 or 8 described a kind of methods that are applicable to the sheet stored of estimation, it is characterized in that, when the data in space are sent to processing unit processes in sheet, reusing by the data that are sent to processing unit in the sheet taken place; When once mating, 256 reference pixels in processing unit, if the pixel of next coupling is the neighbor of horizontal direction, then only need will next-door neighbour reference macroblock right side or 16 pixels of left vertical direction be sent to processing unit; If the pixel of next coupling is the neighbor of vertical direction, then only need 16 pixels of next-door neighbour's reference macroblock downside or upside parallel direction are sent to processing unit.

10, a kind of method that is applicable to the sheet stored of estimation as claimed in claim 1 or 2 is characterized in that, whole pixel motion estimation, sub-pixel motion estimation and motion compensation are all finished in the space in the central column sheet; Whole pixel motion estimate with MVP (pmvx, pmvy) ,-SRX＜=pmvx＜=SRX ,-SRY＜=pmvy＜=SRY are central point, with-SRX＜=mvx-pmvx＜SRX ,-SRY＜=mvy-pmvy＜SRY is search in the zone of hunting zone; Sub-pixel motion estimation is around optimum whole pixel, will carry out 1/2 pixel value difference earlier and obtain optimum 1/2 pixel, carries out 1/4 pixel value difference and obtains optimum 1/4 pixel around optimum 1/2 pixel again; The step of motion compensation is: (mvx is when putting in order pixel, not need to do the sub-pixel difference mvy) as optimal motion vector MV; When MV (mvx, when mvy) being 1/2 pixel, 1/2 pixel value difference only; (mvx when mvy) being 1/4 pixel, does 1/2 pixel value difference earlier, does 1/4 pixel value difference again as MV; Mvx wherein represents the position of optimal motion vector MV at directions X, and mvy represents the position of optimal motion vector MV in the Y direction; (wherein pmvx represents the position of future position MVP at directions X to MVP for pmvx, pmvy) expression future position, and pmvy represents the position of future position MVP in the Y direction.