CN101621685B - Coder and coding method - Google Patents
Coder and coding method Download PDFInfo
- Publication number
- CN101621685B CN101621685B CN 200810130316 CN200810130316A CN101621685B CN 101621685 B CN101621685 B CN 101621685B CN 200810130316 CN200810130316 CN 200810130316 CN 200810130316 A CN200810130316 A CN 200810130316A CN 101621685 B CN101621685 B CN 101621685B
- Authority
- CN
- China
- Prior art keywords
- border
- image
- zone
- frame
- frame refreshing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention relates to a video coder and coding method. The video coder comprises a frame storage, an intra-frame refresh area controller, an intra-frame coder, an inter frame coder and a grammar coder. The intra-frame refresh area controller receives information of images adjacent in coding sequence, the images adjacent in coding sequence comprises a current image and a next image arranged next to the current image; each macro block of the next image searches a motion vector by taking the current image as a reference image; for each predetermined candidate intra-frame refresh area boundary, a matching block crossing from the inside of a candidate intra-frame refresh area boundary to the outside thereof is found for the motion vector; the intra-frame refresh area boundary of the current image is selected in the candidate intra-frame refresh area boundary according to the information of the matching block; and the boundary position information is output. The video coder of the invention can realize coding with short delay and high efficiency.
Description
Technical field
The present invention relates to video coding technique, particularly intraframe coding and prediction interframe encoding mode coexist as the frame refreshing video coding technique in the piece image.
Background technology
The frame refreshing video coding technique is a kind of mild technology of video code flow that makes.It has the effect that improves the video compression coding real-time simultaneously.
Intraframe coding and interframe encode are two kinds of methods of carrying out compressed encoding at video sequence.The coding of video sequence is unit with the image.Intraframe coding is meant when a certain width of cloth image of coding, the information that does not need other images as a reference, therefore, during decoding, decoder can only be finished the decoding of this image with the compressed information of piece image.Because image adjacent in the video sequence is very similar usually, promptly the life period redundancy can come video sequence is adopted the interframe encode method by estimation and motion compensation mechanism.Estimation is the movable information of obtaining between image sequence, i.e. the process of motion vector.Motion vector is attached to by unified size or regular divided image piece, and this image block is called macro block or macro block is divided.When adopting interframe encode (being called inter prediction encoding again) technology, encoder is at first obtained present image with the motion vector between reference picture, go out predicted picture by reference picture and motion-vector prediction again, send the difference between motion vector and present image and predicted picture to decoder.The reference picture that decoder obtains motion vector and decoded, the generation forecast image again with the difference addition, just can utilize the reference picture of having decoded to decode present image.All reference pictures show that constantly all the inter prediction before present image is called forward prediction, and a part of reference picture is engraved in before the present image when showing, the inter prediction that another part reference picture is engraved in when showing after the present image is called bi-directional predicted.The interframe encode method of motion compensation principle that Here it is.Above-described intraframe coding and interframe encode method are the existing dynamic image encoding methods that the most extensively adopts, and are applied in the series standard that ISO/IEC and ITU-T issue.In normal circumstances, in the frame and interframe encode be that unit divides according to image, Figure 11 has shown the video sequence of being made up of the inter prediction image of infra-frame prediction image (underlined I image) and forward prediction (underlined P image).
Because interframe encode has made full use of reference image information, eliminated coding burden, the interior and inter-frame encoding obvious difference on code efficiency of frame to time redundancy.Figure 12 has shown that the corresponding typical code check of the every width of cloth image of Figure 11 video sequence distributes.Usually the code check of I image reaches ten times of the P image or more.This structure of sequence is brought two main problems: the first, and the code check shake behind the coding is violent.Code check difference great disparity causes transmission band occupation inequality in time between image, the waste transfer resource, and for the strategy that level and smooth code check must adopt buffering and repack, this has increased processing burden and delay.The second, the big delay.Decoder can decode the prerequisite of any inter prediction image be received in advance the data of front one width of cloth in-frame encoding picture and successfully decoding generate reference picture.When decoder receives the video data of compression at random, sizable probability is arranged since a width of cloth inter coded images, this moment, decoder did not have any reference picture of this width of cloth image.Therefore all images before receiving next in-frame encoding picture data all can not be decoded.The size of average retardation when the distance between continuous two in-frame encoding pictures has determined random access.Generally, this delay at half second between the several seconds.And can significantly increase code check for the interval that reduces to postpone to shorten in-frame encoding picture.
The frame refreshing technology is in order to overcome the above problems.In the frame refreshing video coding, in same width of cloth image, mix and use intraframe coding and interframe encode, promptly intraframe coding is used in the subregion of same image and to using interframe encode in another part zone, intraframe coding constantly refreshes zones of different in continuous some images simultaneously.Behind the process image of some, a regional inswept complete image of intraframe coding.In any piece image, the typical case in zone divides as shown in figure 13.Image is divided into three parts, and in the fixing cycle, the intraframe coding banded zone begins to streak all areas of image until the other end from an end of image.Figure 14 shows that intra-coded regions streaks all surfaces of image from top to bottom in the cycle of 4 width of cloth images, and the position non-overlapping copies, t express time among the figure.The main benefit of frame refreshing coding is level and smooth and low delay of code check.Figure 15 has shown that the video sequence identical with Figure 12 starts anew to distribute with the code check behind the frame refreshing technology for encoding.As shown in figure 15, except first width of cloth image has the very big code check, the code check of other images is all very approaching, though be higher than the code check of the code check of P image among Figure 12 much smaller than the I image.
When decoder receives the packed data of any piece image at random, necessarily comprise the intra-coding data of subregion, can be with this regional decoding, other zones are owing to be interframe encode, can't decode under the situation of reference picture not having.Along with the continuous inswept imaging surface of intra-coded regions, the part that can decode also can be increasing, can be also increasing as the zone of reference picture, and therefore the zone of increasing interframe encode can be decoded.
Recommended to adopt the frame refreshing regional structure to realize the low delayed video coding of conformance with standard standard in the mode of standard option in the video coding of international standard MPEG-2 (ISO/IEC 13818) part (be ITU-T accordingly H.262).
Low delayed video coding is applied to fields such as the video transmission of two-way interactives such as face-to-face communication, video conference and monitoring in real time, for the delay in the switching channels in the reduction digital television broadcasting direct improvement effect is arranged also.
Yet under normal conditions, the every width of cloth image of this division is that the technique decodes effect in intra-coded regions and interframe encode zone is also incomplete and desirable.Reason is the problem that does not have locality protection as the reference picture in interframe encode zone.Particularly, if the estimation in interframe encode zone, the searching zone that is motion vector is under no restraint, when reference image block pointed comprises regional beyond the last width of cloth image intra refresh areas territory, will produce the failure of motion compensation, the decoding error that this image block occurs, as shown in figure 17.Like this, as long as constantly there is the motion vector of this sensing, video sequence just can not be correctly decoded all the time.
The U.S. 5,057, No. 916 patents have proposed to solve by the locality protection method that limits the estimation scope technology of the problem of completeness of decoding.Locality protection is the method at coding side constrained motion vector scope.From the start position of a certain agreement, guarantee that the reference picture of all interframe encodes all is limited in the inswept zone, frame refreshing zone.In the inswept from top to bottom coding example of intraframe coding band as shown in figure 14, agreement refreshes the top that start position is an image, the frame refreshing zone begins to the picture number that arrives at the top next time from the top be exactly the cycle of frame refreshing, and the cycle is 4 among Figure 14.The area dividing of image has characteristic shown in Figure 13, promptly by frame refreshing regional inswept " clean district " with not by inswept " dirty district "." clean district " and " dirty district " all are meant the interframe encode zone.In each cycle, when " the clean district " of any piece image of encoding, estimation all is limited to the hunting zone of motion vector in " clean district " scope and intra-coded regions of reference picture, i.e. in the locality protection scope, as the a-quadrant among Figure 16; And when " the dirty district " of any piece image of encoding, the hunting zone of motion vector does not add particular determination, and for example the hunting zone is as the B zone among Figure 16.First frame refreshing from the off is in the cycle, and " clean district " can correctly carry out the decoding of interframe encode, and the interframe encode decoding in " dirty district " is then done one's best, and can not guarantee complete.In case finish the decoding in first cycle, " clean district " and " dirty district " just can be correctly decoded.
Because the existence of locality protection, the restriction in motion-vector search zone make that motion vector is not optimum, promptly exists the estimation failure.The interframe encode efficient step-down of " clean district ", under the poorest situation can near in addition be lower than intraframe coding efficient.This mainly occurs in the fringe region that closes on the frame refreshing zone in " clean district ", and its size is by kinetic characteristic decision between frame of video.
At this problem, people such as R.M.Schreier are at IEEE Transactions on ConsumerElectronics, Vol.52, No.1, in the paper on the Feb.2006 " Motion Adaptive IntraRefresh for the is Video Coding Standard H.264 ", proposed by selecting a frame refreshing path to make the technology that binary encoding efficient improves in the refresh cycle.At first, be image division plurality of fixed size boxed area; Then, these boxed area successively become intra-coded regions in a refresh cycle, because numerous paths that can travel through these boxed area are arranged, so need be to every possible path calculation code efficient.This code efficiency is mainly influenced by incorrect estimation.The degree of estimation failure can be crossed over the summation on " clean district " border with motion vector and be weighed, and chooses the frame refreshing strategy of a minimum path of this value as optimum.There are following three shortcomings in this method.(1) high latency, because need the initial data of all images of a complete refresh cycle to determine refresh path, promptly all images need be stored in a refresh cycle, postpones the whole cycle, and because path number has the factorial order of magnitude of boxed area number, operand is huge.(2) cut apart bad adaptability for horizontal or vertical band, cut apart for more normal adopted horizontal or vertical band, almost do not have boxed area optional, choosing of jump becomes unique possibility, and it is extremely low to adapt to the statistical probability of motion in this path.(3) exist the code efficiency of fixed boundary to lose, because the shape in every frame refreshing zone is fixed, the intrinsic problem of bringing from the motion vector of " clean district " sensings " dirty district " can only be alleviated by selecting the less zone of refreshing of the code check extent of damage, and can not eliminate the loss of this code check.
Summary of the invention
The present invention is directed to above-mentioned because the code efficiency that frame refreshing brings reduces problem, proposed dynamically to revise the method on border, frame refreshing zone, effectively avoided above-mentioned by selecting three shortcomings of frame refreshing path method according to motion vector.
Particularly, the invention provides a kind of video coding apparatus, it comprises frame memory, frame refreshing zone controller, intra encoder, interframe encode device, grammatical tagging device.Frame memory receives and stores image to be encoded, and keeps at least two cycle of images.The frame refreshing zone controller is received in the information of image adjacent on the coded sequence from frame memory, adjacent image comprises present image and is positioned at this current image back piece image afterwards on this coded sequence, is the reference picture Vector search of taking exercises to each macro block of this back piece image with this current image, each predetermined border, candidate's intra refresh areas territory for this current image, find out the match block that motion vector is crossed in the border, candidate's intra refresh areas territory outside border, candidate's intra refresh areas territory, and from border, candidate's intra refresh areas territory, selects the frame refreshing border, zone of this current image and export the positional information on the regional border of this frame refreshing according to the information of this match block.Intra encoder receives the positional information on border, intra refresh areas territory, and receive the information of present image from frame memory, generate in this current image coded data with intra-frame encoding mode, and export this coded data by the frame refreshing zone of border, frame refreshing zone defined.The interframe encode device receives the positional information on border, intra refresh areas territory, and the information that receives present image and be positioned at the preceding piece image before this current image from frame memory, generate the coded data in the zone except that the frame refreshing zone in this current image with the inter motion compensation pattern, and export this coded data.The grammatical tagging device receives from the coded data of intra encoder and the output of interframe encode device, adds grammatical organization information according to the video encoding standard form, thereby generates video flowing and export this video flowing.
Video coding apparatus of the present invention is determined the border in frame refreshing zone by the frame refreshing zone controller, thereby has realized postponing coding little and that code efficiency is high.
Specifically in the frame refreshing zone controller of video coding apparatus of the present invention, the determining of the border in frame refreshing zone goes out the area sum of these match block outside border, candidate's intra refresh areas territory according to the information calculations of match block, with the border, candidate's intra refresh areas territory of area sum minimum value correspondence as the frame refreshing border, zone of present image and export the positional information on this border, frame refreshing zone.
The definite of the border in frame refreshing zone also can be, in the frame refreshing zone controller, go out the summation of difference of the ordinate on the ordinate of these match block and border, candidate's intra refresh areas territory according to the information calculations of match block, with the border, candidate's intra refresh areas territory of the summation minimum value correspondence of the difference of ordinate as the frame refreshing border, zone of present image and export the positional information on this border, frame refreshing zone.
In addition, in the video coding apparatus of the present invention, the interframe encode device can adopt the interframe encode device of forward prediction mode.
And, can also comprise the interframe encode device of bi-directional predicted mode in the video coding apparatus of the present invention.
The video coding apparatus of the invention described above all can be realized the coding that delay of the present invention is little, code efficiency is high.
The invention still further relates to a kind of method for video coding, it is characterized in that, comprising: frame storing step, frame refreshing zone controlled step, intraframe coding step, interframe encode step, grammatical tagging step.In the frame storing step, receive and store image to be encoded, and keep at least two cycle of images.In the controlled step of frame refreshing zone, be received in the information of image adjacent on the coded sequence, adjacent image comprises present image and is positioned at this current image back piece image afterwards on this coded sequence, is the reference picture Vector search of taking exercises to each macro block of this back piece image with this current image, each predetermined border, candidate's intra refresh areas territory for this current image, find out the match block that motion vector is crossed in the border, candidate's intra refresh areas territory outside border, candidate's intra refresh areas territory, and from border, candidate's intra refresh areas territory, selects the frame refreshing border, zone of present image and export the positional information on the regional border of this frame refreshing according to the information of this match block.In the intraframe coding step, receive the positional information on border, intra refresh areas territory and the information that receives present image, generate in this current image coded data with intra-frame encoding mode, and export this coded data by the frame refreshing zone of border, frame refreshing zone defined.In the interframe encode step, the information that receives positional information, the present image on border, intra refresh areas territory and be positioned at this current image preceding piece image before, generate the coded data in the zone except that the frame refreshing zone in this current image with the inter motion compensation pattern, and export this coded data.In the grammatical tagging step, receive from the coded data of intraframe coding step and the output of interframe encode step, add grammatical organization information according to the video encoding standard form, thereby generate video flowing and export this video flowing.
Method for video coding of the present invention is determined the border in frame refreshing zone by frame refreshing zone controlled step, thereby has realized postponing coding little and that code efficiency is high.
Specifically in the controlled step of the frame refreshing zone of method for video coding of the present invention, information calculations according to match block goes out the area sum of these match block outside border, candidate's intra refresh areas territory, with the border, candidate's intra refresh areas territory of area sum minimum value correspondence as the frame refreshing border of present image and export the positional information on this border, frame refreshing zone.
In the controlled step of frame refreshing zone, also can go out the summation of difference of the ordinate on the ordinate of these match block and border, candidate's intra refresh areas territory according to the information calculations of match block, with the border, candidate's intra refresh areas territory of the summation minimum value correspondence of the difference of 7 ordinates as the frame refreshing border of present image and export the positional information on this border, frame refreshing zone.
In addition, in the method for video coding of the present invention, the interframe encode step can adopt the inter-frame encoding of forward prediction.
And, can also comprise the coding step that carries out interframe encode in bi-directional predicted mode in the method for video coding of the present invention.
The method for video coding of the invention described above all can be realized the coding that delay of the present invention is little, code efficiency is high.
Description of drawings
Fig. 1 represents an embodiment of video coding apparatus of the present invention.
Fig. 2 is illustrated in the flow process of determining border, frame refreshing zone in the frame refreshing zone controller.
Fig. 3 represents to add up the process of border, candidate's intra refresh areas territory motion vector distribution on every side.
The position coordinates of the match block of Fig. 4 motion search.
Fig. 5 represents that the method according to this invention determines the schematic diagram in frame refreshing zone.
Fig. 6 is illustrated in the schematic diagram that adopts many reference pictures in the prediction interframe encoding mode.
Fig. 7 represents the schematic diagram of third embodiment of the invention.
Fig. 8 represents the schematic diagram of four embodiment of the invention.
Fig. 9 represents the schematic diagram of fifth embodiment of the invention.
The schematic diagram in dynamic refresh cycle when Figure 10 represents to adopt coding method of the present invention.
Figure 11 represents in the prior art video sequence with intraframe coding method and inter-frame encoding coding.
Figure 12 represents that the typical code check in the video sequence shown in Figure 11 distributes.
Figure 13 represents with the area dividing in the frame refreshing coded system image encoded.
Figure 14 is illustrated in residing position, frame refreshing zone in the video sequence.
Figure 15 represents that the video sequence identical with Figure 12 starts anew to distribute with the code check behind the frame refreshing technology for encoding.
Figure 16 represents the position in frame refreshing zone in present image and the preceding piece image.
Figure 17 represents the situation on motion vector leap present image intra refresh areas territory and " dirty district " border.
Embodiment
Below in conjunction with accompanying drawing the specific embodiment of the present invention is done specific description.But these execution modes only are to give an example to of the present invention, are not limitation of the invention.
Fig. 1 represents video coding apparatus 100 of the present invention, and it comprises frame memory 101, frame refreshing zone controller 102, intra encoder 103, interframe encode device 104, grammatical tagging device 105.Frame storing step in the method for video coding of the present invention, frame refreshing zone controlled step, intraframe coding step, interframe encode step and grammer edit step are carried out in frame memory 101, frame refreshing zone controller 102, intra encoder 103, interframe encode device 104 and grammatical tagging device 105 respectively.
Frame memory 101 receives and stores image to be encoded, and at least two cycle of images of maintenance, because the frame refreshing zone controller need present image and with the information of this current image adjacent back piece image on coded sequence, and interframe encode device when the coding present image (and when adopting forward prediction mode) needs present image and the information of adjacent preceding piece image on coded sequence.
Frame refreshing zone controller 102 from frame memory receive present image and coded sequence be positioned at deserve before the information of back piece image (in the following description this being positioned at present image back piece image abbreviation " back piece image " afterwards) after the image, is the reference picture Vector search of taking exercises to each macro block of this back piece image with this current image, each predetermined border, candidate's intra refresh areas territory for this current image, find out the match block that motion vector is crossed in the border, candidate's intra refresh areas territory outside border, candidate's intra refresh areas territory, and from border, candidate's intra refresh areas territory, selects the frame refreshing border, zone of this current image and export the positional information on the regional border of this frame refreshing according to the information of this match block.
Intra encoder 103 receives the positional information on border, intra refresh areas territory, and receive the information of present image from frame memory, generate in this current image coded data with intra-frame encoding mode, and export this coded data by the frame refreshing zone of border, frame refreshing zone defined.Therefore, following to explanation of the present invention in, for this frame refreshing zone, also be called " intra-coded regions " by border, frame refreshing zone defined.
Interframe encode device 104 receives the positional information on border, intra refresh areas territory, and receive present image and be positioned at preceding piece image before this current image information of (the preceding piece image that in the following description this is positioned at before the present image is called for short " preceding piece image ") from frame memory, generate the coded data in the zone except that the frame refreshing zone in this current image with the inter motion compensation pattern, and export this coded data.
Grammatical tagging device 105 receives from the coded data of intra encoder 103 and 104 outputs of interframe encode device, adds grammatical organization information according to the video encoding standard form, thereby generates video flowing and export this video flowing.
In frame refreshing zone controller 102, when determining the border, frame refreshing zone of present image according to the information of above-mentioned match block, can be in different ways, will provide below and specify.
Fig. 2 is illustrated in the flow process of determining border, frame refreshing zone in the frame refreshing zone controller.As shown in Figure 2, at first read in present image and buffer memory; Then, read in back piece image and buffer memory; Then, to each macro block in the whole image of back piece image is the reference picture Vector search of taking exercises with the present image, with macro block and the position thereof of finding coupling, thereby obtain the motion vector of this match block, the starting point of motion vector is the position of this macro block in the piece image of back, terminal point be in the present image with the position of the macro block of this macroblock match.Do not limit the scope of motion-vector search terminal point at this.Above motion vector has been described and has been present in present image and the back movable information between the piece image with macro block or the macro block unit of being divided into.At last, add up border, candidate intra refresh areas territory motion vector distribution on every side, and determine the position (just finishing the prediction on border, frame refreshing zone) on the border, frame refreshing zone of present image according to the extreme value of the motion vector distribution around the border, candidate's intra refresh areas territory.
Specify the process of statistics border, candidate intra refresh areas territory motion vector distribution on every side below in conjunction with Fig. 3.At this, suppose that each macro block MB has a motion vector MV, do not consider the situation that macro block is divided here.
At first, select all motion vectors, form the motion vector collection by these motion vectors from " clean district " sensing " dirty district " direction.
In image, so that (i, j) the expression macro block position is a unit with the macro block number, is defined in position (i, j) the macro block MB in the piece image of back
IjThe motion vector MV that search obtains in present image
IjFor,
(Δx
ij，Δy
ij)＝(x
ij1-x
ij0，y
ij1-y
ij0) (1.1)
Wherein, Δ x
IjExpression macro block MB
IjThe cross stream component of corresponding motion vector, Δ y
IjExpression macro block MB
IjThe longitudinal component of corresponding motion vector, x
Ij1Expression macro block MB
IjThe abscissa value of the terminal point of corresponding motion vector, y
Ij1Expression macro block MB
IjThe ordinate value of the terminal point of corresponding motion vector, x
Ij0Expression macro block MB
IjThe abscissa value of the starting point of corresponding motion vector, y
Ij0Expression macro block MB
IjThe ordinate value of the starting point of corresponding motion vector.Choose all Δ y
Ij=y
Ij1-y
Ij0＞0 motion vector.In addition, about the span of i, j, decide according to the pixel of each width of cloth image.For example for the macroblock size of high definition (1920 * 1080) picture and (16 * 16), the i span is 0 to 199, and the j span is 0 to 67.Because the frame refreshing zone is inswept to low side from the top usually, promptly the edge ordinate in " clean district " is every through all constantly increasing behind the piece image; Δ y
IjThe sensing of＞0 motion vector is from " clean district " sensing " dirty district " direction.
Secondly, set the initial position of boundary line, candidate intra refresh areas territory, i.e. the ordinate y value on border, candidate's intra refresh areas territory.Stipulate that this y coordinate is the ordinate of coboundary of delegation's macro block at edge, frame refreshing zone, unit is a pixel.Initial value is chosen the ordinate of the coboundary of the adjacent delegation's macro block of the clean district lower boundary of present image, and promptly candidate's intra refresh areas territory is made of delegation's macro block.The initial point of the upper left corner positioning image of image is that the horizontal ordinate of unit all is a minimum value with the pixel, to the lower right to, horizontal ordinate value constantly increases.
Once more, satisfy Δ y from above-mentioned
Ij=y
Ij1-y
Ij0The motion vector of＞0 condition is concentrated, and selects the motion vector of motion vector leap to border, candidate's intra refresh areas territory outside in the border, candidate's intra refresh areas territory, promptly satisfies y
Ij1＞y＞y
Ij0The motion vector of condition, these motion vectors piece pointed is exactly above-mentioned match block.The match block of crossing over the motion vector correspondence on border, candidate intra refresh areas territory has covered frame refreshing zone and " clean district " image-region in addition, the view data that has promptly comprised " dirty district " by the reference picture that match block constituted of these motion vector correspondences, therefore the failure that will bring estimation, as shown in figure 17.Cause the difference of the match block of macroblock to be encoded image and prediction to increase, the effect of motion compensation is lowered.
Therefore,, calculate the area sum of the part of match block beyond, and write down the position on this value and corresponding border, candidate's intra refresh areas territory on border, candidate's intra refresh areas territory for the influence that the failure of estimating to move brings.The area sum of these match block is proportional to the ordinate terminal point y of match block
Ij1With the summation of the difference of the ordinate of candidate's intra refresh areas territory border y, promptly
The value of formula.Wherein, h is that mod is a modulo operation in the height of a macro block of pixel, question mark (?) whether the value of the expression formula that constituted with colon (:) part before according to question mark be that zero is not chosen value before and after the colon as the value of this expression formula.Particularly, as shown in Figure 4, for crossing over, according to the distance (y that crosses over apart from the leap that is no more than a macro block height h
Ij1-y) modh counts sum formula; Leap for the height that surpasses a macro block counts sum formula according to a macro block height h.
The reference picture that " clean district " motion search of interior macro block of piece image obtains after when making present image become the reference picture of back piece image remains in " distinguishing only " as far as possible, motion vector as much as possible final position need be encompassed in new " the clean district " that present image coding back produces, make the minimum that influences that above estimation failure causes with this.
Therefore border, candidate's intra refresh areas territory is moved down, amount of movement is a macro block height h, till the image bottom is arrived on border, candidate's intra refresh areas territory.At last, obtain beyond the border, candidate's intra refresh areas territory of record the match block area and minimum, find corresponding border, candidate's intra refresh areas territory, i.e. the border in Zui Jia frame refreshing zone.In other words, the border in Zui Jia frame refreshing zone is by with y
OptBe the macro-block line formation of coboundary ordinate, y
OptComputing formula be shown below.
Above-mentioned formula 1.3 represented implications are exactly, each predetermined border, candidate's intra refresh areas territory for this current image, obtain the summation of difference of the ordinate on the ordinate of match block and this border, candidate's intra refresh areas territory, with the border, candidate's intra refresh areas territory of the summation minimum value correspondence of the difference of ordinate border y as the frame refreshing zone of present image
Opt
Illustrate said process in conjunction with Fig. 5, the position of mark among the present image n " X " is for to meet y for different y value (promptly for different borders, candidate's intra refresh areas territory)
Ij1＞y＞y
Ij0On the position of same " X ", can there be a plurality of motion vector terminal points in the final position of the motion vector of condition.y
pBe predeterminated position, promptly use the lower boundary in the frame refreshing zone of conventional intra refresh method according to the frame refreshing zone of the inswept image of constant speed.The definition picture altitude is H, and the refresh cycle is P, and the predetermined width W that refreshes band is H/P, and W is the integral multiple of macro block height h, y
pBe the coboundary ordinate of delegation's macro block of this bottom, default frame refreshing zone, owing to drop on y in the final position of above motion vector as can be seen from motion vector terminal point distribution (block diagram on the left side Fig. 5)
pAnd y
pThe probability in the zone the between+W is more relatively, if the default border that conventional method is determined is chosen on indication frame refreshing border, has more motion-vector prediction failure and takes place.And distributing (block diagram on the left side Fig. 5) as can be seen from the motion vector terminal point, the minimum value of criterion appears at y in the formula (1.2)
OptTo y
OptZone between the+h, y here
OptIt is the integral multiple of macro block height.Therefore, select y
Opt+ h is as the boundary pixel position in adjusted frame refreshing zone, the lower boundary in the frame refreshing zone of expansion as shown in Figure 5 just.The macro-block line position at place, border is y
Opt(ordinate in the macro block upper left corner).
Discussed above is the situation that motion vector is attached to macro block, and the situation principle that is attached to the division of macro block for motion vector is identical, only need substitute macro block with the son division at concrete video code model and can adopt said method.
Below explanation the present invention is applied to the adaptability in the existing video encoding standard.
Mark in the code stream that encoder-side generates, need be carried out to the boundary position in frame refreshing zone in the zone of the frame refreshing that the border in code device of the present invention and coding method generation frame refreshing zone is variable therefore.Frame refreshing zone with strips is divided into example, and positional information is the pixel ordinate value of strip edge, can add that the number of macroblocks in strip width or the band describes by band starting point ordinate value.This is to support the current method for expressing in the video encoding standard of multi-ribbon band described, appears at usually in the syntactic structure of slice header.For example, define slice_vertical_position (back 8 bits of slice_start_code) and slice_vertical_position_extension (for the image that surpasses 2800 row) in the MPEG-2 video (ISO/IEC 13818-2), indicated with the macro block upright position of the bar tape leader macro block that is unit.Therefore, when the present invention is applied to support the video encoding standard of multi-ribbon, can not change the basic coding form, need in output code flow, not add special marking, therefore adopt the code stream after method of the present invention is encoded still to meet original coding standard.
Present embodiment can also adopt the technology of many reference pictures in the inter prediction zone.As shown in Figure 6, among the image n+2 during coding " clean district ", " clean district " data (A shown in Figure 6 and B zone) of reference picture n+1 and image n simultaneously; Coding is when " dirty district " among the image n+2, simultaneously the full width view data of reference picture n+1 and image n.
This present invention's code device and coding method have the following advantages:
First, the single reference picture of situation have only forward prediction and to(for) interframe encode, the present invention only utilizes the information of piece image more than the simplest inter prediction in coding present image process, promptly only need present image, back piece image and the last width of cloth three width of cloth images altogether.Therefore, the coding delay of this invention is extremely low, and image sequence (as the index of the definition among CCIR601 or the ITU-R BT.709) that frame frequency is 25fps for example is because the coding delay that piece image causes increasing behind the buffer memory has only 1/25 second.
The second, accommodation is very extensive.Code device of the present invention and coding method can be applied to all frame refreshing codings that adopt band or block to divide.
The 3rd, the present invention dynamically changes the size in frame refreshing zone, under the situation of the coding structure that keeps frame refreshing, has farthest optimized code efficiency according to the movable information between image.
In order further to reduce in the controlled step of frame refreshing zone, the calculating strength of the motion search between back piece image and the present image has been taked the series of optimum technology in the following execution mode.
Below the method for second execution mode of the present invention is elaborated.
In second execution mode, limit the scope of the macro block that participates in estimation in the piece image of back and in present image, carry out the scope of motion search.This qualification can realize in many ways, and the cycle request that refreshes according to video is set the scope of the macro block that participates in estimation in the piece image of back and carry out the scope of motion search in present image in the present embodiment.
As shown in Figure 7, be H for the definition picture altitude, the refresh cycle is P, the predetermined width W that refreshes the zone is H/P.Among Fig. 7, L1 represents that the expansion of hunting zone is limited to one times of default frame refreshing band scope; L2 represents the optimization frame refreshing zone expanded; L3 represents to participate in the scope at the macro block place of estimation.That is to say,, carry out motion search and move to fail influencing estimation for the macro block that all are distinguished only and expansion is twice in the default frame refreshing band scope corresponding to the expansion of hunting zone.
In order to keep a stable refresh cycle, do not increase or reduce the number in frame refreshing zone, realized with the maximum changing range in the every width of cloth image intra refresh areas of restriction territory.The k that one-period has begun (decide according to refresh cycle P by the span of this k value, the k span is 1 to P, for example, at 4 width of cloth images is that the span of k under the situation of one-period is 1 to 4) in the width of cloth image, the excursion of the optimization boundary position in frame refreshing zone is limited in the frame refreshing zone lower boundary of k-1 width of cloth image between (k+1) * (H/P).Promptly limit the span of the ordinate candidate value y on " clean district " border in formula (1.2) and (1.3).For example, rule of thumb value is default, and promptly condition is by original y
Ij1＞y＞y
Ij0Strengthen and be y
Ij1＞y＞y
Ij0And (k+1) * (H/P)＞y＞Y
1Wherein, Y
1It is the border in the frame refreshing zone of k-1 width of cloth image.The hunting zone that reduces helps to reduce computational complexity.
Below the method for the 3rd execution mode of the present invention is described.
In the 3rd execution mode, adopt the judgment criterion on the border, adjustment frame refreshing zone of raising the efficiency.Step is as follows, at first, calculates the motion vector vertical component Δ y that the back piece image is obtaining with present image image for referencial use by band (the frame refreshing zone is shaped as the situation of band) for unit
Ij=y
Ij1-y
Ij0For on the occasion of this component Δ y
IjSum; Then, choose this component Δ y
IjThe lower boundary of a band of sum minimum is as the lower boundary in frame refreshing zone.So just can draw the information on required border, frame refreshing zone through once calculating.
That is, each the y value in the formula (1.3) satisfies y
Ij1＞y＞y
Ij0The set of motion vector all different, therefore,, all to recomputate formula (1.2) for each this y value
Value.In order to reduce this calculating strength, use following approximate formula.
Y is chosen in formula (1.4) expression
Ij1The ordinate of that band at place refreshes the ordinate y (y of band boundaries as the candidate
Ij1∈ Slice
yAnd obtain y represented implication),
Ij1-y
Ij0The Δ y of＞0 motion vector
Ij=y
Ij1-y
Ij0Sum is chosen Δ y
Ij=y
Ij1-y
Ij0The corresponding y of sum minimum value is as the ordinate y on the border in preferred frame refreshing zone
Opt
The precondition that can be similar to like this is: the texture laws of motion have the space similitude between image, and promptly close more texture has more close mobile trend between adjacent image probability is big more.As one of component, the vertical component of motion vector also has similar space similitude.Therefore, the possibility on " clean district " border of piece image is also more little after the leap that motion vector brought that Δ y is generally more little.Under normal conditions, this prerequisite all can satisfy.Violent and when presenting very scrappy shape, the result of formula (1.4) will have more different with formula (1.3) when running into image texture motion.
Below in conjunction with Fig. 8 the method for the 4th execution mode of the present invention is described.
The present invention not only can be applied to have only usually the low delayed video coding of the interframe encode of intraframe coding and forward direction inter prediction, also can be used for having the situation that bi-directional predicted interframe encode and frame refreshing are encoded and mixed.As shown in Figure 8, bidirectional predictive picture (with " B " expression) is inserted between the frame refreshing image encoded, and with the frame refreshing coded image as the reference image, the arrow of pointing out from the B image among Fig. 8 is exactly a direction of pointing to reference picture.T express time among Fig. 8.
The benefit that adds the B image is farthest to keep high coding efficiency when making code check level and smooth.
Below the method for the 5th execution mode of the present invention is described.
The match block of calculating for aforementioned each candidate's intra refresh areas territory boundary position the border outer area and size, the reduction of the code efficiency that causes with the estimation failure has the relation of intimate direct ratio.Therefore, can make it near desired value by adjusting the Position Control encoder bit rate on border.
Fig. 9 is for making video frequency coding rate reach the schematic diagram of the method for target bit rate scope by changing border, frame refreshing zone.
At first, read in and preset coding target bit rate and fluctuation tolerance band, and set the width in frame refreshing zone according to predetermined width.
Secondly, the boundary position for each candidate's intra refresh areas territory calculates the area sum of match block outside the border.
Then, to the present image precoding.Final encoding code stream is not exported in precoding, and just in order to calculate the code check size after present image is encoded.
At last, judge that above-mentioned code check is whether near the fluctuation tolerance band target bit rate, if code check surpasses the fluctuation tolerance band, then become the situation of the code check that still diminishes greatly as required and select the moving direction on different borders, frame refreshing zone, in code check access admittance scope, selected border, final candidate's intra refresh areas territory is for optimizing the border.For example, if code check is lower than target bit rate, then select to make aforementioned match block border outer area and bigger candidate's intra refresh areas territory boundary position.
Because the present invention dynamically changes the size in frame refreshing zone, therefore effect be exactly the level and smooth purpose of code check under, when a fixing frame refreshing cycle does not finish, entire image that the frame refreshing zone is inswept.Figure 10 has shown this effect.In the cataloged procedure that comprises the refresh cycle in two whole frame shown in Figure 10, first cycle comprises the image of four width of cloth frame refreshings and interframe encode mixing; Second period only comprises three similar width of cloth images.Figure 10 has also comprised first width of cloth coded image in the 3rd cycle in addition.
The benefit of this technology is, eliminated the restriction that fixing refresh cycle smoothly brings to code check, can control code check more accurately, makes it smooth to a higher degree of stability.
According to code device of the present invention and coding method, because can dynamically revise border, frame refreshing zone according to motion vector, delay height, bad adaptability, the code efficiency that therefore can effectively avoid existing in the existing frame refreshing coding techniques are lost big defective.That is to say, according to code device of the present invention and coding method, because the frame refreshing zone is according to motion vector and dynamic change, can when keeping that the frame refreshing technology is original and making the level and smooth advantage of code check, can also realize postponing coding little, that adaptability is high, code efficiency is high.
Claims (10)
1. a video coding apparatus is characterized in that, comprising:
Frame memory receives and stores image to be encoded, and keeps at least two cycle of images;
The frame refreshing zone controller, be received in the information of image adjacent on the coded sequence from described frame memory, adjacent image comprises present image and is positioned at this current image back piece image afterwards on this coded sequence, is the reference picture Vector search of taking exercises to each macro block of this back piece image with this current image, each predetermined border, candidate's intra refresh areas territory for this current image, find out the match block that motion vector is crossed in the border, described candidate's intra refresh areas territory outside border, described candidate's intra refresh areas territory, and from border, described candidate's intra refresh areas territory, selects the frame refreshing border, zone of this current image and export the positional information on the regional border of this frame refreshing according to the information of this match block;
Intra encoder, receive the positional information on border, described frame refreshing zone, and receive the information of described present image from described frame memory, generate in this current image coded data with intra-frame encoding mode, and export this coded data by the frame refreshing zone of border, described frame refreshing zone defined;
The interframe encode device, receive the positional information on border, described frame refreshing zone, and the information that receives described present image and be positioned at the preceding piece image before this current image from described frame memory, generate the coded data in the zone except that described frame refreshing zone in this current image with the inter motion compensation pattern, and export this coded data;
The grammatical tagging device receives from the coded data of described intra encoder and the output of described interframe encode device, adds grammatical organization information according to the video encoding standard form, thereby generates video flowing and export this video flowing.
2. video coding apparatus as claimed in claim 1 is characterized in that,
In described frame refreshing zone controller, information calculations according to described match block goes out the area sum of these match block outside border, described candidate's intra refresh areas territory, with the border, candidate's intra refresh areas territory of described area sum minimum value correspondence as the frame refreshing border, zone of described present image and export the positional information on this border, frame refreshing zone.
3. video coding apparatus as claimed in claim 1 is characterized in that,
In described frame refreshing zone controller, go out the summation of difference of the ordinate on the ordinate of these match block and border, described candidate's intra refresh areas territory according to the information calculations of described match block, with the border, candidate's intra refresh areas territory of the summation minimum value correspondence of the difference of described ordinate as the frame refreshing border, zone of present image and export the positional information on this border, frame refreshing zone.
4. as any described video coding apparatus in the claim 1～3, it is characterized in that,
The interframe encode device that described interframe encode device is the forward prediction mode.
5. as any described video coding apparatus in the claim 1～3, it is characterized in that,
The interframe encode device that also comprises bi-directional predicted mode.
6. a method for video coding is characterized in that, comprising:
The frame storing step receives and stores image to be encoded, and keeps at least two cycle of images;
Frame refreshing zone controlled step, be received in the information of image adjacent on the coded sequence, adjacent image comprises present image and is positioned at this current image back piece image afterwards on this coded sequence, is the reference picture Vector search of taking exercises to each macro block of this back piece image with this current image, each predetermined border, candidate's intra refresh areas territory for this current image, find out the match block that motion vector is crossed in the border, described candidate's intra refresh areas territory outside border, described candidate's intra refresh areas territory, and from border, described candidate's intra refresh areas territory, selects the frame refreshing border, zone of this current image and export the positional information on the regional border of this frame refreshing according to the information of this match block;
The intraframe coding step, receive the positional information on border, described frame refreshing zone and the information that receives described present image, generate in this current image coded data with intra-frame encoding mode, and export this coded data by the frame refreshing zone of border, described frame refreshing zone defined;
The interframe encode step, receive positional information, the described present image on border, described frame refreshing zone and the information that is positioned at this current image preceding piece image before, generate the coded data in the zone except that described frame refreshing zone in this current image with the inter motion compensation pattern, and export this coded data;
The grammatical tagging step receives from the coded data of described intraframe coding step and the output of described interframe encode step, adds grammatical organization information according to the video encoding standard form, thereby generates video flowing and export this video flowing.
7. method for video coding as claimed in claim 6 is characterized in that,
In the controlled step of described frame refreshing zone, information calculations according to described match block goes out the area sum of these match block outside border, described candidate's intra refresh areas territory, with the border, candidate's intra refresh areas territory of described area sum minimum value correspondence as the frame refreshing border of described present image and export the positional information on this border, frame refreshing zone.
8. method for video coding as claimed in claim 6 is characterized in that,
In the controlled step of described frame refreshing zone, go out the summation of difference of the ordinate on the ordinate of these match block and border, described candidate's intra refresh areas territory according to the information calculations of described match block, with the border, candidate's intra refresh areas territory of the summation minimum value correspondence of the difference of described ordinate as the frame refreshing border, zone of present image and export the positional information on this border, frame refreshing zone.
9. as any described method for video coding in the claim 6～8, it is characterized in that,
Described interframe encode step adopts the inter-frame encoding of forward prediction.
10. as any described method for video coding in the claim 6～8, it is characterized in that,
After described frame storing step and before described grammer edit step, also comprise the coding step that carries out interframe encode in bi-directional predicted mode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200810130316 CN101621685B (en) | 2008-07-04 | 2008-07-04 | Coder and coding method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200810130316 CN101621685B (en) | 2008-07-04 | 2008-07-04 | Coder and coding method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101621685A CN101621685A (en) | 2010-01-06 |
| CN101621685B true CN101621685B (en) | 2011-06-15 |
Family
ID=41514677
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200810130316 Expired - Fee Related CN101621685B (en) | 2008-07-04 | 2008-07-04 | Coder and coding method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101621685B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3565246A1 (en) * | 2018-05-01 | 2019-11-06 | Agora Lab, Inc. | Progressive i-slice reference for packet loss resilient video coding |
| RU2815483C1 (en) * | 2021-03-19 | 2024-03-19 | Ханчжоу Хиквижн Диджитал Текнолоджи Ко., Лтд. | Decoding and encoding based on adaptive internal update mechanism |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102271250B (en) * | 2010-06-01 | 2013-04-17 | 杭州华三通信技术有限公司 | Intra-macro block refreshing method and apparatus thereof |
| KR101456499B1 (en) * | 2010-07-09 | 2014-11-03 | 삼성전자주식회사 | Method and apparatus for encoding and decoding motion vector |
| CN108540813B (en) | 2012-09-28 | 2021-03-16 | 杜比国际公司 | Image decoding device |
| CN103841427B (en) * | 2012-11-27 | 2016-12-21 | 炬芯（珠海）科技有限公司 | The sliding method of a kind of search window and device thereof |
| CN103873872B (en) * | 2012-12-13 | 2017-07-07 | 联发科技（新加坡）私人有限公司 | Reference pictures management method and device |
| CN105989175B (en) * | 2015-03-05 | 2019-11-22 | 深圳市腾讯计算机系统有限公司 | Image display method and apparatus |
| US10652572B2 (en) * | 2016-04-29 | 2020-05-12 | Ati Technologies Ulc | Motion-adaptive intra-refresh for high-efficiency, low-delay video coding |
| CN111355959B (en) * | 2018-12-22 | 2024-04-09 | 华为技术有限公司 | Image block dividing method and device |
| WO2020237466A1 (en) * | 2019-05-27 | 2020-12-03 | 深圳市大疆创新科技有限公司 | Video transmission method and apparatus, and aircraft, playback device, and storage medium |
| EP3809700B1 (en) * | 2019-10-16 | 2022-02-16 | Axis AB | Periodic intra refresh pattern for video encoding |
| CN113873236B (en) * | 2021-02-23 | 2022-12-02 | 杭州海康威视数字技术股份有限公司 | Decoding and encoding method based on adaptive intra-frame refreshing mechanism and related equipment |
| CN114630122B (en) * | 2021-03-19 | 2023-04-28 | 杭州海康威视数字技术股份有限公司 | Decoding and encoding method based on self-adaptive intra-frame refreshing mechanism and related equipment |
| CN115695889A (en) * | 2022-09-30 | 2023-02-03 | 聚好看科技股份有限公司 | Display device and floating window display method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1471321A (en) * | 2003-07-14 | 2004-01-28 | 大唐微电子技术有限公司 | Method for coding and recoding ripple video frequency based on motion estimation |
| CN1684518A (en) * | 2004-04-15 | 2005-10-19 | 微软公司 | High-fidelity transcoding |
| CN1882082A (en) * | 2006-05-11 | 2006-12-20 | 北京经天科技有限公司 | Selective encryption algorithm aiming at network video |
| CN1964487A (en) * | 2005-11-11 | 2007-05-16 | 北京微视讯通数字技术有限公司 | Quick refreshing method and device for video code processing |
-
2008
- 2008-07-04 CN CN 200810130316 patent/CN101621685B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1471321A (en) * | 2003-07-14 | 2004-01-28 | 大唐微电子技术有限公司 | Method for coding and recoding ripple video frequency based on motion estimation |
| CN1684518A (en) * | 2004-04-15 | 2005-10-19 | 微软公司 | High-fidelity transcoding |
| CN1964487A (en) * | 2005-11-11 | 2007-05-16 | 北京微视讯通数字技术有限公司 | Quick refreshing method and device for video code processing |
| CN1882082A (en) * | 2006-05-11 | 2006-12-20 | 北京经天科技有限公司 | Selective encryption algorithm aiming at network video |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3565246A1 (en) * | 2018-05-01 | 2019-11-06 | Agora Lab, Inc. | Progressive i-slice reference for packet loss resilient video coding |
| RU2815483C1 (en) * | 2021-03-19 | 2024-03-19 | Ханчжоу Хиквижн Диджитал Текнолоджи Ко., Лтд. | Decoding and encoding based on adaptive internal update mechanism |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101621685A (en) | 2010-01-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101621685B (en) | Coder and coding method | |
| CN101610417B (en) | Image filling method, device and equipment | |
| CN101513071B (en) | Method and apparatus for determining expected distortion in decoded video blocks | |
| CN100425077C (en) | Video compression code ratio control method using uneven distribution of frame target bit number | |
| CN102595135B (en) | Method and device for scalable video coding | |
| CN102752595B (en) | Hybrid skip mode used for depth map encoding and decoding | |
| CN101345877B (en) | Moving-picture coding device and moving-picture coding method | |
| CN102833532B (en) | For to the method for digital coding video data and digital video encoder system | |
| CN101232615A (en) | Method for random access and gradual picture refresh in video coding | |
| CN102037730A (en) | Content adaptive video encoder and coding method | |
| CN104243982A (en) | Video coding and decoding processing method and device | |
| CN103338370B (en) | A kind of multi-view depth video fast encoding method | |
| CN101867813B (en) | Multi-view video coding method oriented for interactive application | |
| CN101094411A (en) | Code rate control method of video code | |
| CN101278564A (en) | Method and apparatus for motion projection error concealment in block-based video | |
| CN104378643A (en) | Intra-frame prediction mode selection method and system of 3D (3-dimension) video plus depth image | |
| CN103634606A (en) | Video encoding method and apparatus | |
| CN101355707B (en) | Error control method for demixing and encoding stereo video | |
| CN101715128A (en) | Method and apparatus of encoding image sequence with a plurality of images | |
| CN103533356A (en) | Video data transmission method and device | |
| CN106060539A (en) | Video encoding method with low transmission bandwidth | |
| CN101401436B (en) | Intra-forecast mode selecting method, moving picture coding method, and device | |
| CN101888561B (en) | Multi-view video transmission error control method for rate distortion optimization dynamic regulation | |
| CN102946533B (en) | Video coding | |
| CN103024384A (en) | Method and device for encoding and decoding videos |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110615 Termination date: 20190704 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |