CN1882096A

CN1882096A - Moving picture signal coding method, decoding method, coding apparatus, and decoding apparatus

Info

Publication number: CN1882096A
Application number: CN 200610099957
Authority: CN
Inventors: 铃木芳典
Original assignee: Hitachi Ltd
Current assignee: Hitachi Ltd
Priority date: 2002-01-24
Filing date: 2003-01-24
Publication date: 2006-12-20
Anticipated expiration: 2023-01-24
Also published as: CN1882097A; CN100508615C; CN100508616C; CN100546385C; CN1856106A; CN1882098A; CN1882099A; CN1882095A; CN100493200C; CN100508614C; CN1882097B; CN1882100A; CN100493201C

Abstract

Moving picture motion compensation coding/decoding predicts a motion vector in a coding block from a motion vector of the adjacent position and codes a differential motion vector calculated from the motion vector to be coded and its prediction motion vector. The coding is performed by switching the coding accuracy between a coding accuracy of the differential motion vector (1) and a coding accuracy of the prediction motion vector and the coding motion vector (2) for each block. Moreover, when decoding the motion vector from the differential motion vector coded by the aforementioned coding, a reversed processing with respect to the coding is performed.

Description

The coding method of moving image, coding/decoding method, code device and decoding device

The application is that application number is 200610084483.3, and the applying date is on January 24th, 2003, and denomination of invention is divided an application for " coding method of moving image, coding/decoding method, code device and decoding device ".

Technical field

The present invention relates to, the coding method of moving image, coding/decoding method, code device and decoding device particularly relate to the coding and the decoding technique of motion vector.

Background technology

The summary that the coding and decoding of moving image is handled describes.The coding of moving image generally is that the code device by Figure 13 carries out.Become 1 frame of the moving image of coded object, as shown in Figure 3, by 1 luminance signal (Y-signal: 61) and 2 color difference signals (Cr:62, the Cb signal: 63) constitute, the image size of color difference signal all is 1/2 of luminance signal in length and breadth.In general video standard, each frame of moving image is divided into fritter, be that unit carries out reproduction processes with the piece that is called as macro block.

Above-mentioned macro block, as shown in Figure 5, by 1 Y-signal piece 30 of 16 * 16 pixels, spatially the Cr block 31 of 8 * 8 pixels consistent with it and Cb block 31 constitute.The coding of picture signal is that unit handles with above-mentioned macro block.In coding, roughly be divided into intraframe coding (mode intra coding in the frame) and predictive coding (interframe mode inter coding).

In the code device of the picture signal of Figure 13, under the situation of intraframe coding, the input macroblock image 201 of coded object is divided into the DCT encoding block and directly carries out DCT203 by dct transform device 203, is transformed into the DCT coefficient.Each conversion coefficient is quantized by quantizer 204, is encoded by demultiplexing device 206.

The DCT coded block size is generally 8 * 8 pixels in coded system in the past, but has also discussed the dct transform that utilizes 4 * 4 pixel sizes recently in MPEG-4 Part 10 (Advanced Video Coding) etc.

As shown in Figure 6, each macro block is divided into 24 4 * 4 block of pixels, is transformed into 16 DCT coefficients respectively by dct transform device 203.Its part of DCT coefficient that is quantized is decoded into the input macro block by the inverse DCT 207 and the anti-DCT device 208 of local decoder 220, and is synthesized in the frame memory 210.Local decoder 220 is made the decoded picture same with decoding device described later.Be stored in image in the frame memory 210 and be used to the prediction of time orientation described later.This intraframe coding is configured in the macro block (also comprising the 1st coded frame) of not similar to previous frame part, or desire is eliminated the part of the storage arithmetic eror of following DCT etc.

Under the situation of interframe (predictive coding) coding, input macroblock image 201 and be stored in motion compensation process between the decoded picture (reference frame) of the coded frame in the frame memory 210 is undertaken by motion compensator 211.Motion compensation is meant from reference frame and retrieves the part similar to the content of target macroblock (generally being absolute value and or square error and the less part of selecting the predictive error signal in the piece), and the compress technique of the time orientation that its amount of exercise (motion vector) is encoded.

The processing structure of motion compensation shown in Fig. 4.Fig. 4 is expression about with the luminance signal piece 72 of the present frame 71 of thick frame encirclement, the prediction piece 75 on the previous frame 73 (reference frame) and the figure of motion vector 76.Motion vector 76 expression spatially is equivalent to amount of movement (the motion vector length that color difference signal is used in prediction piece 75 zones of piece (dotted line) 74 to the present frame on the previous frame of same position from the thick frame piece for present frame, be made as half of luminance signal, do not encode).Detected motion vector, and the motion vectors (predicting) that obtains from MV fallout predictor 215 from the motion vector of adjacent block between carried out encoding by demultiplexing device 206 after the difference processing.The predicted macroblock image 213 that extracts from previous frame by motion compensation, and carry out difference processing by difference engine 202 between the input macroblock image 201 of present frame, and generate the difference macroblock image.

The difference macroblock image as shown in Figure 6, is divided into 24 4 * 4 block of pixels, is transformed into 16 DCT coefficients respectively by dct transform device 203.Each DCT coefficient is quantized by quantizer 204, is encoded by demultiplexing device 206.Under the situation of interframe encode also be by the inverse DCT 207 of local decoder 220 and anti-DCT device 208 with quantization DCT coefficient, be decoded into the difference macroblock image, undertaken being synthesized in the frame memory 210 after the addition by adder 209 and predicted macroblock image.In addition, here the number of the motion vector that will encode in 1 macro block is made as 1, but generally is that macro block further is divided into fritter, each fritter is calculated motion vector carry out Methods for Coding.For example, in MPEG-4 Part10 (Advanced Video Coding), the macroblock partition type in the time of will carrying out motion compensation (brightness composition) is made as Fig. 7 and considers.Basically be 4 kinds of type 51～type 54.Type 54 is respectively to 48 * 8 54-0～54-3, further selects from 5 kinds of type 54a, 54b, 54c, 54d and intraframe codings.

And then as the motion compensation process of predictive coding, the information that is divided into the frame that use passes by is in time carried out the forward predictive coded of motion compensation and is used and pass by in time and bi-directional predictive coding that the information of the frame in future is carried out motion compensation.In the motion compensation of above-mentioned forward predictive coded, only carry out forward prediction, but in the motion compensation of alternating binary coding, except forward prediction, also handle back forecast, bi-directional predicted and utilize motion vector information in following in time frame from the two-way direct prediction of predicting.

The encoding precision of motion vector is generally handled integer pixel or the precision below it.For example, in the Advanced of MPEG-4 Part 2 Simple Profile, can select 1/2 pixel precision and 1/4 pixel precision to each bit data.Computational methods about integer-pel precision or the predicted value below it, save explanation, in 1/2 pixel precision, adapt to the linear filter that ground uses 2 taps, in 1/4 pixel precision, adapt to ground and use 8 tap filters and 2 tap filters according to location of pixels according to location of pixels.In addition, in MPEG-4 Part 10, also discussed and come the expressive movement vector, used 6 tap filters and 2 tap filter calculations to go out predicted value with 1/4 pixel precision.Like this, by the filter that uses long tap number estimated performance is improved.

To be suitable for the frame that intraframe coding obtains to whole macro blocks and be called I-Picture, will be called P-Picture, will be called B-Picture by alternating binary coding or frame that intraframe coding constituted by forward predictive coded or frame that intraframe coding constituted.In P-Picture or B-Picture,, judge each macro block is carried out with which of predictive coding and intraframe coding, and encode by 206 pairs of result of determination 218 of demultiplexing device by the INTRA/INTER determinant 214 of Figure 13.

The decoding of the code signal of moving image is carried out with the step opposite with coding.Generally be that decoding device by Figure 14 carries out.

Code is deciphered portion 501, and the coded data of being imported is analyzed, and motion vector related information and macro block type information are given MV fallout predictor 508, and quantization DCT coefficient information is given inverse DCT 502.At macro block (mb) type is under the situation of intraframe coding, in inverse DCT 502 and anti-DCT device 503, decoded quantization DCT coefficient information is carried out the anti-DCT of inverse quantization to each 4 * 4 block of pixels handle, and reproduces macroblock image.Prediction mode at macro block is under the situation of predictive coding, calculates decoding motion vectors by MV fallout predictor 508 from decoding differential motion vector and motion vectors, and is input to motion compensator 504.In motion compensator 504, extract the predicted macroblock image from the frame memory 507 of the decoded picture of storing previous frame according to amount of exercise.

Then, in inverse DCT 502 and anti-DCT device 503,, each 4 * 4 block of pixels is carried out the anti-DCT of inverse quantization handle, reproduce the difference macroblock image the coded data relevant with predictive error signal.Then, carry out addition by 505 pairs of predicted macroblock images of adder and difference macroblock image and handle, reproduce macroblock image.Reproduced macroblock image synthesizes the decoded frame image by synthesizer 506.In addition, the decoded frame image is that the prediction of next frame is with being stored in the frame memory 507.

As the document related, patent disclosure communique (the flat 11-46364 communique of Japan Patent Publication Laid-Open) is arranged with the present invention.Recording in this document, being that unit carries out preparing several motion vector precision in the coding-decoding method of the moving image that coding and decoding handles with the piece, is the means of the encoding precision of unit motion switch vector with a plurality of pieces that bind together.

In the motion compensation encoding of known moving image in the past, to have discussed and prepared the multiple motion vector number in the piece that is included in, the encoding precision of motion vector is 1/8 a pixel precision high coded system like that.But, because the relation of the size of code of motion vector even comprise in piece under the situation of a plurality of motions, also has selection to carry out the tendency of Methods for Coding with few motion vector number.In addition, generate at the filter that uses long tap number under the situation of integer-pel precision or the predict pixel below it, just produce for generate 1 prediction piece used become the problem of big such storage access with reference to the pixel count in the image.This problem particularly just becomes problem when carrying out motion compensation.

Thereby purpose of the present invention just provides, and can cut down the size of code of motion vector and can cut down coding method, coding/decoding method, code device and the decoding device that is increased the motion image signal of the prediction error information that causes by the piece number in the encoding block.

Other purposes of the present invention are exactly, in code device and decoding device, and storage access scope when reducing motion compensation.

Summary of the invention

The invention provides a kind of coding method of moving image, have following steps,

Image segmentation is become a plurality of encoding blocks, come the motion vector of predictive coding piece by motion vector from adjacent block, wherein with the motion vector of this adjacent block as motion vectors, with the piece is that unit encodes to the differential motion vector that is calculated by the motion vector that should encode and motion vectors thereof, wherein add that by motion vector differential motion vector is used as the motion vector of encoding block in this adjacent block

It is characterized in that: this method also has following steps,

Prepare the precision of several differential motion vector,, and the information of selected precision and the differential motion vector of this precision encoded the precision of the selected differential motion vector of each encoding block.

It is characterized in that: this method also has following steps,

Prepare the precision of several motion vectors, to the precision of the selected motion vector of each encoding block, and encode to the information of above-mentioned selected precision and by the differential motion vector that the motion vector computation of motion vectors that is modified to above-mentioned selected precision and selected precision goes out.

The invention provides a kind of coding/decoding method of moving image, have following steps,

Image segmentation is become a plurality of decoding blocks, comes the motion vector of prediction decoding piece, calculate decoding motion vectors by decoding differential motion vector and motion vectors from the decoding motion vectors of adjacent block,

It is characterized in that: this method also has following steps,

Prepare the precision of several differential motion vector, to the precision information of each decoding block decoding differential motion vector and the differential motion vector of this precision, after the precision of the differential motion vector of will decoding is modified to the precision identical with motion vectors, the motion vector that calculates decoding block from the decoding differential motion vector of carrying out the precision correction and motion vectors.

It is characterized in that: also have following steps:

Prepare the precision of several motion vectors, to the precision information of each decoding block decoding motion vectors and the differential motion vector of this precision, after the precision with motion vectors is modified to the precision identical with decode precision information, calculate the motion vector of decoding block from decoding differential motion vector and the motion vectors of carrying out the precision correction.

In order to achieve the above object, the invention provides a kind of coding method of moving image, comprise image segmentation is become a plurality of encoding blocks, come motion vector in the predictive coding piece from the motion vector of adjoining position, with the piece is the step that unit encodes to the differential motion vector that is calculated by the motion vector that should encode and motion vectors thereof, motion vector is encoded, have: (1) prepares the precision of several differential motion vector, to the precision of the selected differential motion vector of each encoding block, and the step that the differential motion vector of the information of selected precision and this precision is encoded; Perhaps (2) prepare the precision of several motion vectors, precision to the selected motion vector of each encoding block, and to the information of selected precision, and the step of encoding by the differential motion vector that the motion vector computation of motion vectors that is modified to selected precision and selected precision goes out.

Here, in the situation of above-mentioned (1) and (2), the decision and the encoding process that comprise the accuracy value (below abbreviate the mv_shift value as) of the coding of following differential motion vector, except in whole macro blocks, carrying out, the type of also utilizing macro block (for example, the size of the number of encoding motion vector) or situation on every side (for example), omit the situation of coding of the mv_shift value of specific macroblock in abutting connection with the situation of macro block.

The situation of above-mentioned (2) also can be distributed different precision on the horizontal composition of motion vector and vertical composition, for each composition the differential vector that the motion vector computation from the precision of being distributed goes out is encoded.

The code utilize moving image encoding of the present invention is used in the decoding of moving image of the present invention, by with the opposite step of step of coding, carry out the reconstruction of image of encoding block.Promptly, prepare the precision of a plurality of motion vectors, distribute different precision on the horizontal composition of motion vector and vertical composition, each composition is modified to this precision that is assigned with, and the differential motion vector that the motion vector computation from this corrections motion vectors and the precision of being distributed goes out is encoded.

And then, under the situation of decoding, the setting of mv_shift value, the processing procedure that the code of corresponding decoding is encoded is carried out.For example, when the code that the mv_shift value is not encoded under the type of macro block or the situation on every side in coding was reproduced, the mv_shift value is made decision, handled to the type of decoded macroblock or situation on every side.

The explanation of the embodiment of the structure of purpose of the present invention, feature, code device and decoding device by following the following drawings will further obtain clearly.

Description of drawings

Fig. 1 is the figure of example of the data syntax of expression forward predicted frame.

Fig. 2 is the figure of example of the data syntax of expression bi-directional predicted frames.

Fig. 3 is the figure of explanation macroblock partition.

Fig. 4 is the figure of the principle of account for motion compensation.

Fig. 5 is illustrated in being included in the block structured figure that is utilized when encoding that has or not of meaningful DCT coefficient in the macro block.

Fig. 6 illustrates expression to carry out that DCT handles and the figure of the macroblock structure of the block unit of encoding process.

Fig. 7 illustrates the figure of structure of luminance block that the block unit of motion compensation is carried out in expression.

Fig. 8 is the figure of the generation of explanation motion vectors.

Fig. 9 is the figure of example of the code table of expression motion vector precision.

Figure 10 is the figure of the Forecasting Methodology of account for motion vector accuracy.

Figure 11 represents that the code table and 8 * 8 of the macro block (mb) type among the P-Picture according to an embodiment of the invention divides the figure of the code table of types.

Figure 12 represents that the code table and 8 * 8 of the macro block (mb) type among the B-Picture according to an embodiment of the invention divides the figure of the code table of types.

Figure 13 is the block diagram of the general structure of expression dynamic image encoding device.

Figure 14 is the block diagram of the general structure of this moving image decoding apparatus of expression.

Figure 15 is the figure of example of the code table of the macro block (mb) type of explanation in the previous methods.

Figure 16 is the figure of the configuration example of expression versatility code table.

Figure 17 is the figure of configuration example of the macro block of the frame structure in the representative domain signal.

Figure 18 is the flow chart of processing that comes the encoding precision of motion switch vector by block size.

Figure 19 is by the assign to flow chart of processing of encoding precision of motion switch vector of the one-tenth of the motion vector in the embodiments of the invention.

Figure 20 is the flow chart of processing that comes the encoding precision of motion switch vector by the composition of the motion vector in the embodiments of the invention and macroblock structure.

Figure 21 is the flow chart of processing that comes the encoding precision of motion switch vector by the composition of the block size in the embodiments of the invention, motion vector and macroblock structure.

Figure 22 is other example of flow chart of processing that comes the encoding precision of motion switch vector by the composition of the block size in the embodiments of the invention, motion vector and macroblock structure.

Figure 23 is the example by the flow chart of the processing of the encoding precision of the motion switch vector that becomes to assign to of block size in the embodiments of the invention and motion vector.

Figure 24 is other example by the flow chart of the processing of the encoding precision of the motion switch vector that becomes to assign to of block size in the embodiments of the invention and motion vector.

Figure 25 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 26 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 27 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 28 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 29 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 30 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 31 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 32 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 33 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 34 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 35 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 36 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 37 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 38 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 39 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 40 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 41 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 42 is the structure chart according to the embodiment of the lsb decoder of the motion vector of moving image decoding apparatus of the present invention.

Figure 43 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 44 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 45 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 46 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Figure 47 is according to the encoding section of the motion vector of dynamic image encoding device of the present invention

The structure chart of embodiment.

Figure 48 is according to the lsb decoder of the motion vector of moving image decoding apparatus of the present invention

The structure chart of embodiment.

Embodiment

Before the explanation embodiments of the invention,, the data syntax of the code data created by the demultiplexing device of code device is described for the understanding that makes the following description is easy.

Fig. 1 represents the part of the data syntax of P-Picture, and Fig. 2 represents the part (owing to identical Therefore, omited with Fig. 1 after the CBP15) of the data syntax of B-Picture.In addition, establish data (motion vector precision information) 13 here, do not appear in the data syntax and describe.In addition, about I-Picture, owing to be not included in Therefore, omited explanation in the feature of the present invention.

In picture header10, comprise the data relevant such as image size, the reproduction moment with the information reproduction of frame.Picture type11 is a frame type information, and the kind of I-Picture, P-Picture, B-Picture is encoded at this.From MB type12 to Tcoeff chroma17 to each macro block data, Nei macroblock number is carried out repeatedly frame by frame.In MB type12, from the table 95 (P-Picture) of Figure 15 or the macro block mode shown in the table 96 (B-Picture), select a mode to encode to each macro block.

The SKIP mode means, isolates the macroblock image that spatially is equivalent to same position from the decoded picture of previous frame and copies.Block sizes when M, the N of the IntraM * N in the table represents to carry out spatial prediction.Little block unit when M * N represents to carry out motion compensation, the mode 1～mode 4 of corresponding diagram 7 respectively.

Direct means direct prediction.Blcok1 and Block2, the mode 2 of identification Fig. 7 or 2 fritters in the mode 3, which of Forward (forward prediction), Backward (back forecast), Bi-directional (bi-directional predicted) be the prediction direction of representing each fritter be.About selected the macro block of 8 * 8 (split) in MB type12, then 8 * 8Partition18 produces.In macro block mode in addition, then 8 * 8Partition18 does not produce.In 8 * 8Partition18, from 8 * 8 dividing mode shown in the table 94 (B-Picture) of the table 92 (P-Picture) of Figure 11 or Figure 12 each 8 * 8 is divided and select a mode, encode.Intra representation space prediction in the table.Little block unit when M * N represents to carry out motion compensation, the division 1～division 4 of corresponding diagram 7 respectively.Direct means direct prediction.

Prediction represents which of Forward (forward prediction), Backward (back forecast), Bi-directional (bi-directional predicted) be the prediction direction of each fritter be.MVD14 represents the coded data of the differential motion vector among the P-Picture.Thereby, be under the situation of IntraM * N at MB type12, then these data do not produce.At MB type12 is under the situation of M * N, and MVD14 encodes repeatedly by the vector number shown in MB type and the 8 * 8Partition.For example, under 16 * 8 situation, the differential motion vector number that should encode is 2, and each motion vector is made of level, 2 vertical compositions, and 4 MVD are carried out coding continuously.MVDFW23 and MVDBW24 represent the coded data of the differential motion vector among the B-Picture.MVDFW23 produces under the situation of the macro block that comprises the motion vector of following the forward prediction shown in MB type12 and the 8 * 8Partition 18 (also comprising bi-directional predicted).Equally, MVDBW24 produces under the situation of the macro block that comprises the motion vector of following the back forecast shown in MB type12 and the 8 * 8Partition 18 (also comprising bi-directional predicted).Thereby, be under the situation of IntraM * N or Direct at MB type12, then these data do not produce.At MB type12 is M * N, follow under the situation that the motion vector of forward prediction (also comprising bi-directional predicted) exists, MVDFW23 is by following the vector number of (also comprising bi-directional predicted) of the forward prediction shown in MB type and the 8 * 8Partition (differential motion vector number * 2, horizontal vertical power) to encode repeatedly.Equally, at MB type12 is M * N, follow under the situation that the motion vector of back forecast (also comprising bi-directional predicted) exists, MVDBW24 is by following the vector number of (also comprising bi-directional predicted) of the forward prediction shown in MB type and the 8 * 8Partition (differential motion vector number * 2, horizontal vertical power) to encode repeatedly.

CBP15 is expression, and about 24 DCT pieces shown in Figure 6, whether the quantization DCT coefficient (meaningful coefficient) beyond 0 is included in 16 coded datas in the coefficient.But, because if about 24 each DCT pieces respectively to meaningful coefficient have or not information to encode then additional information will increase, so here, be that unit encoded to having or not of meaningful coefficient to have gathered 8 * 8 of 4 DCT pieces as shown in Figure 6.

Tcoeff luma16 represents the coded data of the quantization DCT coefficient in the luminance block 40.Coding is from 4 * 4 40-0 to 4 * 4 40-15, respectively 16 quantization parameters is encoded by the little order of numbering.But about represented not have 8 * 8 of meaningful coefficient by CBP, encoding process just is omitted.

Tcoeff chroma17 represents the coded data of the quantization DCT coefficient in the luminance block 40.Coding is respectively 16 quantization parameters to be encoded by the order of 4 * 4 41-0,41-2,41-3,42-0,42-1,42-3.But about represented not have the signal component of meaningful coefficient by CBP15, encoding process just is omitted.

Coding method about the coded data of macroblock level illustrated in figures 1 and 2, consider only to use the Universal VLC (UVLC) of a kind of variable-length codes table, the coding method of having made up fixed-length code (FLC) and variable length code (preparing the code table different) and arithmetic coding (for example, with reference to non-patent literature 1) etc. to each coding elements.The structure of the table 83 expression UVLC of Figure 16, the value of Xn is " 0 " or " 1 ".The example of the variable-length codes table of expression reality in table 84.As the concrete grammar of arithmetic coding, consider to replace the connotation of each code with the binary data of number bit, everybody is carried out Methods for Coding according to the occurrence Probability Model of each generation probability of 0 and 1 of expression.The method is called as CABAC (Context-basedAdaptive Binary Arithmetic Coding) (Witten et al., " ArithmeticCoding for Data Compression ", Comm.of the ACM, 30 (6), 1987, pp.520-541).

The present invention particularly has feature on the encoding section/lsb decoder of motion vector.That is, on by the structure of the MV portion 508 of the decoding device of MV prediction section 215, dynamic compensating unit 211 and Figure 14 of the code device of the processing of the data M VD grammer MVD precision13 portion of Fig. 1 and Fig. 2, Figure 13, dynamic compensating unit 504 and information processing method, has feature.The structure of other parts and signal processing method are substantially identical with in the past device, method, and the embodiment to characteristic of the present invention describes below.

＜embodiment 1 〉

Present embodiment is the encoding section at motion vector, image segmentation is become a plurality of encoding blocks, predict the picture signal of above-mentioned encoding block and the motion vector in the above-mentioned encoding block from the motion vector of adjoining position, be in unit the moving image encoding that the differential motion vector that is calculated by the motion vector that should encode and motion vectors thereof the is encoded motion compensation encoding of moving image (below abbreviate as) with the piece, prepare the precision of several differential motion vector, to the precision of the selected differential motion vector of each decoding block, and the information of selected precision and the differential motion vector of this precision information carried out Methods for Coding.

At first, use Fig. 8 that the generation of above-mentioned motion vectors is described.Here, the fritter of establishing the object that motion vector is encoded is 50.In fritter 50, respectively to the horizontal vertical composition of motion vector, 3 the motion vector that will be positioned at adjoining position A, B, C calculates its median as candidate, and the motion vector of median is made as motion vectors.But, in the relation of coded sequence and macro block position the piece of position C before for coding situation or be positioned under the situation outside the image, the motion vector that the position of substitution C will be positioned at the piece of position D uses as one of candidate motion vector.In addition, do not have under the situation of motion vector, its motion vector is made as 0 vector carries out prediction processing at the piece of position A, B, C, D.At this moment, under the situation that 2 do not have a motion vector among 3 candidate pieces, a remaining candidate motion vector is made as motion vectors.Above-mentioned Forecasting Methodology, piece 51-0, fritter 54a-0,54b-0,54b-1,54c-0,54c-2 and the 54d-0～54d-3 of mode 4 (54) to the mode 1 (51) of Fig. 7 use identical Forecasting Methodology.

About 2 fritters (52-0,52-1) of mode 2 (52), 2 fritters (53-0,53-1) of mode 3 (53), the motion vector that will be positioned at the piece at arrow root place shown in Figure 8 is made as predicted value.In addition, in arbitrary mode, the motion vector that the aberration composition is used is not carried out coding, and the motion vector of brightness composition is used divided by 2 grades.

In present embodiment 1, the candidate of at first establishing the precision of differential motion vector is 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision.Be set as value after 4 times if establish MV for the motion vector composition that 1/4 pixel precision is expressed, if PWV is set as value after 4 times for the motion vectors composition that 1/4 pixel precision is expressed, then in this expresses, as following, limit for the value of the MV of each precision candidate.

1 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 4 multiple.

1/2 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 2 multiple.

1/4 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 1 multiple.

Thereby, for the effect that makes the manner is reflected in coding characteristic in motion vector is inferred, prepare the absolute value of predictive error signal or square error and on added the impact assessment function of the size of code of motion vector.At this moment, each precision of (MV-PMV) is calculated the optimum evaluation of estimate of each fritter, the motion vector of the whole fritters in macro block infer handle be through with after, by the addition value to the mean value of each accuracy computation fritter, the precision decision that the summation of mean value is less is the precision of differential motion vector.The differential motion vector MVD that encodes calculates with formula (1).

MVD＝(MV-PMV)＞＞mv_shift……(1)

Mv_shift in the formula (1) is the value shown in the table 81 of Fig. 9.Formula (1) expression will deduct value behind the motion vectors composition PMV from motion vector composition MV, and the value of having carried out to the right behind the arithmetic shift position by the number shown in the mv_shift is differential motion vector composition MVD.Here, deduct value that motion vectors composition PMV obtains from motion vector composition MV and be 1＜＜multiple of mv_shift, carried out the value that differential motion vector composition MVD that the right bit shift of arithmetic handles just becomes 1 multiple.

At Picture type is under the situation of P-Picture, and mv_shift is as the coded data key element shown in the data M VD precision13 of Fig. 1, and MVD is carried out coding as the coding elements shown in the data M VD14.At Picture type is under the situation of B-Picture, mv_shift is as the coded data key element shown in the data M VD precision13 of Fig. 2, and MVD will usually encode as the coded data shown in data M VDFW23 (being under the situation of differential motion vector of forward prediction at MVD) or the data M VDBW24 (being under the situation of differential motion vector of back forecast at MVD).In addition, then be omitted in the macro block that the coded data that is coded in differential motion vector of MVD precision (MVD14 or MVDFW23 and MVDBW24) does not produce.Specifically, be exactly P-Picture:MB type be IntraM * N or, MB type is that 8 * 8partition of 4 fritters of 8 * 8 (split) all is the situation of Intra, B-Picture:MB type is the situation of IntraM * N or Direct, and then MB type is that 8 * 8partition of 4 fritters of 8 * 8 (split) all is the situation of Intra or Direct.

Shown in example in the past, like that, adopt UVLC or arithmetic coding etc. as coding method.Mode shown in the table 81 of Fig. 9 is to have used the example of the coded system under the UVLC situation.Under the situation of arithmetic coding, the generation value of the motion vectors of the MB that for example encodes (7C of Figure 10) is encoded according to the urgent MB of keeping left (7A of Figure 10) with near the generation probability tables that the mv_shift value of last MB (7B of Figure 10) calculates.

In the decoding side, from decoding mv_shift value and decoding MVD and PMV, according to following formula,

MV＝(MVD＜＜mv?shift)+PMV……(2)

Carry out the decoding of each motion vector composition, promptly the MV value is reproduced (decoding).Formula (2) is illustrated in differential motion vector composition MVD has been carried out left adding that the value behind the motion vectors composition PMV becomes decoding motion vectors composition MV on the value behind the arithmetic shift position by the number shown in the mv_shift.Handle by this arithmetic left dislocation position, decoded differential motion vector composition MVD just from 1 multiple become 1＜＜multiple (corresponding to the MV-PMV of the side of encoding) of mv_shift.By adding that on this value PMV decoding motion vectors composition MV just is recovered to the value of original precision.

Then, be that the situation that 1/8 pixel precision, 1/4 pixel precision, 1/2 pixel precision, 1 pixel precision are 4 kinds is narrated to the candidate of the precision of differential motion vector.Be set as value after 8 times if establish MV for the motion vector composition that 1/8 pixel precision is expressed, if PWV is set as value after 8 times for the motion vectors composition that 1/8 pixel precision is expressed, then in this expresses, as following, limit for the value of the MV of each precision candidate.

1 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 8 multiple.

1/2 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 4 multiple.

1/4 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 2 multiple.

1/8 pixel precision: the value horizontal vertical composition of (MV-PMV) in the whole fritters in the macro block all is the MV of 1 multiple.

The method of inferring about motion, since identical with the situation of 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision, the Therefore, omited explanation.Differential motion vector MVD calculates with formula (1).Here, the value that deducts motion vectors composition PMV from motion vector composition MV is 1＜＜multiple of mv_shift, carried out the value that differential motion vector composition MVD that the right bit shift of arithmetic handles just becomes 1 multiple, calculate with above-mentioned formula (1).Mv_shift in the formula (1) is the value shown in the table 82 of Fig. 9.

Coding method about mv_shift and MVD, because with the candidate of the precision of differential motion vector is that the situation of 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision is identical, Therefore, omited explanation has used code under the UVLC situation just according to table 82 rather than table 81.

The processing of decoding side is that the situation of 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision is identical with the candidate of the precision of differential motion vector also, from decoding mv_shift value and decoding MVD and PMV, according to above-mentioned formula (2), the decoding MV value of each motion vector composition is reproduced.Formula (2) is illustrated in differential motion vector composition MVD has been carried out left by the number shown in the mv_shift adding that the value of motion vectors composition PMV becomes decoding motion vectors composition MV on the value of arithmetic shift position.Handle by this arithmetic left dislocation position, decoded differential motion vector composition MVD just from 1 multiple become 1＜＜multiple (corresponding to the MV-PMV of the side of encoding) of mv_shift.By adding that on this value PMV decoding motion vectors composition MV just is recovered to the value of original precision.

＜embodiment 2 〉

Present embodiment 2 is in the motion compensation encoding of moving image, prepare the precision of multiple motion vector, precision to the selected motion vector of each macro block, to selected precision information, and by the motion vectors that is modified to this precision of having selected be modified to the differential motion vector that the encoding motion vector of this precision of having selected calculates and carry out Methods for Coding.

At first, the candidate of establishing the precision of motion vectors and encoding motion vector is 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision.Be set as value after 4 times if establish MV for the motion vector composition that 1/4 pixel precision is expressed, if PWV is set as value after 4 times for the motion vectors composition that 1/4 pixel precision is expressed, then in this expresses, as following, limit for the value of the MV of each precision candidate.

1 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 4 multiple.

1/2 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 2 multiple.

1/4 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 1 multiple.

Thereby, for the effect that makes the manner is reflected in coding characteristic in motion vector is inferred, prepare the absolute value of predictive error signal or square error and on added the impact assessment function of the size of code of motion vector.At this moment, each precision of MV is calculated the optimum evaluation of estimate of each fritter, the motion vector of the whole fritters in macro block infer handle be through with after, by the addition value to the mean value of each accuracy computation fritter, precision decision that just can the summation of mean value is less is the precision of differential motion vector.The differential motion vector MVD that encodes calculates with formula (3).

MVD＝(MV-((PMV＞＞mv_shift)＜＜mv_shift))＞＞mv_shift……(3)

Mv_shift in the formula (3) is the value shown in the table 81 of Fig. 9.Formula (3) expression is to deducting from motion vector composition MV, motion vectors composition PMV has been carried out value behind the arithmetic shift position to the right behind the arithmetic shift position at once left by the number shown in the mv_shift, and further the value of having implemented the right bit shift of the 2nd arithmetic by the number shown in the mv_shift becomes differential motion vector composition MVD.Here, handle and handle the left dislocation position for the continuous right bit shift of motion vectors composition, play that value with motion vectors composition PMV is identical with MV to be transformed into 1＜＜effect of the value of the multiple of mv_shift.In addition, handle by the right bit shift of the 2nd arithmetic, differential motion vector composition MVD just becomes the value of 1 multiple.

Like this, in the method for the precision of switching encoding motion vector, when motion is inferred, just do not need to consider the precision of PMV, but when the generation of MVD is handled, need carry out the precise decreasing of PMV motion vectors is arrived the processing of the precision of mv_shift.But formula (3) just can be cut down the computing step number by simplifying as following.When the realization of circuit design etc. just effectively.In this formula, implement difference processing after motion vector composition MV and motion vectors composition PMV being transformed into the value of 1 multiple." will be transformed into predictive vector composition behind 1 the multiple be set as 1＜＜mv_shift value doubly " with following formula in " be transformed into 1＜＜the motion vectors composition of mv_shift value doubly " be equal to, in addition since motion vector composition MV be 1＜＜multiple of mv_shift, so the disappearance of the information that following formula caused of use just can not take place.

MVD＝(MV＞＞mv_shift)-(PMV＞＞mv_shift)……(4)

About distribution and the coding method of the data syntax of mv_shift and MVD to Fig. 1 and Fig. 2 since with the method identical Therefore, omited explanation of the precision of switched differential motion vector.

MV＝(MVD＜＜mv_shift)+((PMV＞＞mv_shift)＜＜mv_shift)……(5)

Decoding MV value to each motion vector composition is reproduced.This formula (5) expression, on the back value of differential motion vector composition MVD having been carried out the arithmetic shift position by the number shown in the mv_shift left, add, motion vectors composition PMV is become decoding motion vectors composition MV by the value that the number shown in the mv_shift has carried out behind the arithmetic shift position behind the arithmetic shift position at once to the right left.Here, handle and the processing of left dislocation position for the continuous right bit shift of motion vectors composition, with differential motion vector composition MVD is carried out the processing of arithmetic shift position left by the number shown in the mv_shift, play that each composition is transformed into 1＜＜effect of the value of the multiple of mv_shift.Thereby that decoding motion vectors composition MV just becomes is original 1＜＜multiple of mv_shift.In addition, formula (5) just can be cut down the computing step number by simplifying as following.When the realization of circuit design etc. just effectively.In formula (5), expression is to the value of the decoded differential motion vector DMV of addition on the motion vectors composition PMV of the value that is transformed into 1 multiple, and the value of having implemented left arithmetic shift position by the number shown in the mv_shift becomes decoding motion vectors composition MV.In " the predictive vector composition that will be transformed into 1 multiple be set as 1＜＜mv_shift value doubly " and the following formula " be transformed into 1＜＜the motion vectors composition of mv_shift multiple " be equal to, in addition because differential motion vector components D MV is 1 multiple, so the disappearance of the information that the following formula of use (6) is caused just can not take place.

MV＝(MVD+(PMV＞＞mv_shift))＜＜mv_shift……(6)

1 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 8 multiple.

1/2 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 4 multiple.

1/4 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 2 multiple.

1/8 pixel precision: the value horizontal vertical composition of the MV in the whole fritters in the macro block all is the MV of 1 multiple.

The method of inferring about motion, since identical with the situation of 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision, the Therefore, omited explanation.The differential motion vector MVD that encodes calculates with following formula.

MVD＝(MV-((PMV＞＞mv_shift)＜＜mv_shift))＞＞mv_shift……(3)

Mv_shift in the formula is the value shown in the table 82 of Fig. 9.This formulate is to deducting from motion vector composition MV, motion vectors composition PMV has been carried out value behind the arithmetic shift position to the right behind the arithmetic shift position at once left by the number shown in the mv_shift, and further the value of having implemented behind the right bit shift of the 2nd arithmetic by the number shown in the mv_shift becomes differential motion vector composition MVD.Here, handle and handle the left dislocation position for the continuous right bit shift of motion vectors composition, play that value with motion vectors composition PMV is identical with MV to be transformed into 1＜＜effect of the value of the multiple of mv_shift.

In addition, handle by the right bit shift of the 2nd arithmetic, differential motion vector composition MVD just becomes the value of 1 multiple.

Coding method about mv_shift and MVD, because with the candidate of the precision of differential motion vector is that the situation of 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision is identical, Therefore, omited explanation has used code under the UVLC situation just according to table 82 rather than table 81.But following formula just can be cut down the computing step number by simplifying as following.When the realization of circuit design etc. just effectively.In this formula, implement difference processing after motion vector composition MV and motion vectors composition PMV being transformed into the value of 1 multiple.In " the predictive vector composition that will be transformed into 1 multiple be set as 1＜＜mv_shift value doubly " and the following formula " be transformed into 1＜＜the motion vectors composition of the multiple of mv_shift " be equal to, in addition since motion vector composition MV be 1＜＜multiple of mv_shift, so the disappearance of the information that following formula caused of use just can not take place.

MVD＝(MV＞＞mv_shift)-(PMV＞＞mv_shift)……(4)

The processing of decoding side is that the situation of 3 kinds of 1/4 pixel precisions, 1/2 pixel precision, 1 pixel precision is identical with the candidate of the precision of differential motion vector also, from decoding mv_shift value and decode MVD and PMV, and according to following formula,

MV＝(MVD＜＜mv_shift)+((PMV＞＞mv_shift)＜＜mv_shift)……(5)

The decoding MV value of each motion vector composition is reproduced.This formulate, add on the value after differential motion vector composition MVD has been carried out the arithmetic shift position left by the number shown in the mv_shift, motion vectors composition PMV is become decoding motion vectors composition MV by the value that the number shown in the mv_shift has carried out behind the arithmetic shift position behind the arithmetic shift position at once to the right left.Here, handle and the processing of left dislocation position for the continuous right bit shift of motion vectors composition, with differential motion vector composition MVD is carried out the processing of arithmetic shift position left by the number shown in the mv_shift, play that each composition is transformed into 1＜＜effect of the value of the multiple of mv_shift.Thereby that decoding motion vectors composition MV just becomes is original 1＜＜multiple of mv_shift.In addition, this formula just can be cut down the computing step number by simplifying as following.When the realization of circuit design etc. just effectively.In this formula, expression is to the value behind the decoded differential motion vector DMV of addition on the motion vectors composition PMV of the value that is transformed into 1 multiple, and the value of having implemented behind the left arithmetic shift position by the number shown in the mv_shift becomes decoding motion vectors composition MV." being transformed into the motion vectors composition of mv_shift multiple " in " the predictive vector composition that will be transformed into 1 multiple be set as 1＜＜mv_shift value doubly " and the following formula is equal to, in addition because differential motion vector components D MV is 1 multiple, so the disappearance of the information that following formula caused of use just can not take place.

MV＝(MVD+(PMV＞＞mv_shift))＜＜mv_shift……(6)

If the foregoing

description

1 and 2 relatively, then because embodiment 1 is actually the precision of adjusting the differential motion vector of encoding, so the computing of comparing with embodiment 2 is simple.With respect to this,, just do not rely on predictive vector so supposed the exploration pixel under the situation of each precision because embodiment 2 is precision of adjusting the motion vector that should encode.Thereby embodiment 1 is effective under the situation of only pursuing coding efficiency, but in storage access cutting method described later, embodiment 2 is effective.

＜embodiment 3 〉

Present embodiment 3 is the codings of not doing one's utmost to carry out the mv_shift value, and according to coding method and coding/decoding method that block mode is predetermined, prevents in the high zone of the selection rate of macro block, the method that the size of code of mv_shift value increases.That is, in the few mode of the motion vector number of following macro block, carry out the coding and the decoding processing of differential motion vector with the default motion vector precision of defined in advance.Handle by this, just omit the coding and the decoding processing of motion vector precision information.

For example, set in coding side and decoding side so that in MB type is 16 * 16 macro block, encode and decoding processing, in the macro block of these modes, just omit the coding and decoding of precision information and handle with 1/4 pixel precision or 1/8 pixel precision.Here, it is 16 * 16 macro block that the mode of handling as the coding and decoding that omits precision information is only set MB type, but can further expand its object.For example, set in coding side and decoding side and make that be that 16 * 16,16 * 8 and 8 * 16 macro block is encoded and decoding processing also is effective with 1/4 pixel precision or 1/8 pixel precision to MB type.In addition, in P-Picture, it is 16 * 16 macro block that the mode of handling as the coding and decoding that omits precision information is only set MB type, in the many B-Picture of the motion vector number of encoding, can be used as also that mode that the coding and decoding that omits precision information handles sets until MB type is 16 * 16,16 * 8 and 8 * 16 macro block.

In addition, the default value of above-mentioned motion vector precision can be set in the scope of 0-3 (at 1/4MC type time be 0-2) equally with mv_shift, switches with bit stream unit or frame unit.At this moment, establish the code of default value, under the situation of bit stream unit, be included in the initial head portion of data, under the situation of frame unit, be included in the picture header data.In addition, under the situation of bit stream unit,, then do not need to be included in the coded data if determine in advance in coding side and decoding side.

And then, by replacing table 95 and the table 96 of Figure 15, prepare to have stipulated the macro block (mb) type of the mv_shift value shown in the table 93 of the table 91 of Figure 11 and Figure 12, the additional information that just can cut down the coding of following the mv_shift value.In addition, when the mv_shift value is carried out arithmetic coding,, adapt to ground and switch the used probabilistic model of coding, code efficiency is improved by based on the motion vector number that in macro block, is comprised.In addition, here, the switching unit of mv_shift is made as macro block unit, but also considers per 8 * 8 situations about switching.Such processing is by being provided with per 8 * 8 modes of handling and per 16 * 16 modes of handling just can realize in MB type.In addition, by per 8 * 8 modes of switching are set in table 81 or table 82, and then per 8 * 8 pairs of 4 mv_shift values are carried out Methods for Coding and also can be realized.

In addition further, switch unit, also can in all frames, be set as fixing or in total sequence, be set as fixing as the selection of differential motion vector or motion vector precision.Unit is switched in the selection of fixed precision, will simplify coding/decoding and handle.Thereby, realize in the simple coding/decoding mode of cost having considered, be effective with the switching unit of frame unit or sequence unit fixed precision.

＜embodiment 4 〉

This form of implementation 4 is to judge the method for shift value uniquely from situation on every side.That is,, cut down the size of code of mv_shft value by come the mv_shift value of predictive coding target macroblock from the adjacency macro block.For example, the urgent macro block that keeps left (7A), the mv_shift value of coming predictive coding target macroblock (the macro block 7C of Figure 10) near the mv_shift value of last macro block (7B).Have from the addition value of the mv_shift value of 7A and 7B as Forecasting Methodology and to carry out forecast method.As the concrete means of utilizing, having at addition value is 0, the MB type of 7C is under 16 * 16 the situation, omit the coding of mv_shift value (MVD precison) and use 1/8 pixel precision or 1/4 pixel precision as setting the both sides of coding and decoding side, under the situation of the combination of in addition addition value and MB type, carry out the control device of the coding of mv_shift value.In addition, this concrete example is an example, and the combination that the situation the when kind of the kind in abutting connection with macro block that forecasting institute is used, Forecasting Methodology, the MB type that controls, control is distinguished kind etc. has multiple.For example, expand to differential motion vector below 2 or 2 by the MB type that will control, perhaps the control point with addition value expands to below 2 or 2, and the reduction of the size of code of mv_shift value just increases.In addition, also have, its difference value is carried out Methods for Coding etc. by the mv_shift value of the such method of the prediction of motion vector from the current MB of mv_shift value prediction of adjacency MB.

Such distortion below in the encoding precision switch means of the motion vector of the foregoing description, also comprising.

Although in the present embodiment, the candidate of establishing the encoding precision of motion vector is 1/8 pixel precision, 1/4 pixel precision, 1/2 pixel precision, 1 pixel precision, is not limited thereto.For example situation about all comprising up to 1/16 pixel precision or, be made as the situation of 2 kinds of 1/2 pixel precisions, 1 pixel precision.

In addition, although in the above-described embodiments, be the coding that unit carries out the encoding precision information of motion vector with the macro block, the method to the encoding precision information of each motion switch vector of a plurality of macro block groups is also arranged.

And then, although in the above-described embodiments, the partitioning scheme of the motion compensation block in the macro block is made as shown in Figure 74 kinds, but because the motion vector encoder method of this specification, even the kind of the partitioning scheme of motion compensation block and number are changed also and can be implemented, so the encoding block structure of motion vector does not limit.

In addition,, UVLC and arithmetic coding have been described as the coded system of coded data although in the above-described embodiments, the motion vector encoder method of this specification, the kind that can be not limited to entropy coding is implemented.In addition, the structural approach of the structure of the code table of UVLC and arithmetic encoding/decoding also is not limited to the table of embodiment.In addition and then, although in the present embodiment,, utilized the default value of mv_shift value about 16 * 16,16 * 8 and 8 * 16 modes, this default value can be applicable to the Partition mode of the part of 8 * 8 (split).In addition, although in an embodiment, that establishes the default value is set at bit stream unit or frame unit, also can distribute different values about each macro block unit.For example have in 16 * 16 modes the default value is set at 1, in 16 * 8 and 8 * 16 modes, the default value is set at 0 method.If briefly say, then about macro block mode and 8 * 8Partition mode, stipulate the encoding precision of the differential motion vector (perhaps motion vector) of each self application with frame unit or sequence unit, encode to transmit and get final product (under the situation of sequence unit, if determine encoding precision in each mode in coding side and decoding side in advance, then do not need to carry out the coding of precision information).As described above, unit is switched in the selection of regular coding precision, will simplify coding/decoding and handle.By increasing the mode of regulation default value, realize that cost just further reduces.In the coding/decoding mode of the simpler structure of needs, for more mode, the switching unit of regular coding precision is effective.

Above to being illustrated according to the coding method of moving image of the present invention and the embodiment of coding/decoding method.Code device and decoding device according to moving image of the present invention, code device about method Figure 13 of embodiment, at first in motion compensator 211, append, according to the combination of MB type and 8 * 8Partition come the coding of ON/OFF mv_shift value determination processing, considered that the motion of mv_shift value infers processing, follows MVD, the MVDFW of mv_shift, the computing of MVDBW and the processing of the mv_shift value being passed to demultiplexing device 206.In addition, in demultiplexing device 206, append the processing that the mv_shift value is encoded as MVDprecision.Decoding device about Figure 14, decipher in the device 501 to be provided with at code, append the handling part that comes the determination processing of the decoding of ON/OFF mv_shift value, MVD precision is carried out process of decoding and the mv_shift value passed to the processing of MV fallout predictor 508 as the mv_shift value according to the decode value of MB type and 8 * 8Partition.In addition, in MV fallout predictor 508, be provided with, will follow MVD, the MVDFW of mv_shift, the handling part that MVDBW is rendered as MV.

Figure 25, Figure 27, Figure 29, Figure 31, Figure 33, Figure 35, Figure 37, Figure 39 and Figure 41 are the block diagram of expression according to major part (encoding section of the motion vector) structure of the embodiment of the code device of moving image of the present invention.Figure 26, Figure 28, Figure 30, Figure 32, Figure 34, Figure 36, Figure 38, Figure 40 and Figure 42 represent to correspond respectively to Figure 25, Figure 27, Figure 29, Figure 31, Figure 33, Figure 35, Figure 37, Figure 39 and Figure 41 code device, according to the block diagram of major part (lsb decoder of the motion vector) structure of the embodiment of the code device of moving image of the present invention.

In each figure, to the additional same numbering of identical function piece in fact.The structure of the device of each figure, action, the explanation by the foregoing description 1 to embodiment 4 can easily realize, so corresponding with the various embodiments described above only is shown, detailed explanation is omitted.

The device of Figure 25 and Figure 26 is made of the formula (1) of carrying out embodiment 1 and the circuit of (2) respectively, and the decision of (mv_shift) value that is shifted is only imported (obtaining when the motion vector detection) according to the outside and controlled.MV prediction section 215 is utilized the memory 101 of the motion vector MV that has stored each piece, generation forecast motion vector PMV.Subtracter 102 is obtained their difference MV-PMV.103 couples of MV-PMV of dextroposition handling part carry out dextroposition.In the decoding side, because the contrary processing of the side of encoding, so the memory 101 of having stored motion vector MV, shift left handling part 113 and adder 112 are set.

The device of Figure 27 and Figure 28, be made of the formula (1) of carrying out embodiment 1 and the circuit of (2) respectively, the decision of (mv_shift) value that is shifted imports (obtaining when the motion vector detection) according to the outside and the some of conditions such as type of macro block determines, controls.Selected in shift value selection portion 104 under the situation of specific block type, selected the mv_shift value, from mv_shift value, used corresponding to the mv_shift value of block type and handle dextroposition handling part 103 from this one by control circuit 105.

The device of Figure 29 and Figure 30 is made of the separate equations (1) of carrying out embodiment 1 and the circuit of (2) respectively, and the decision of displacement (mv_shift) value only determines, controls according to the conditions such as type of macro block.

The device of Figure 31 and Figure 32 is made of the formula (3) of carrying out embodiment 2 and the circuit of (5) respectively, and the decision of (mv_shift) value that is shifted is only imported (obtaining when the motion vector detection) according to the outside and controlled.Precision correcting process portion 106, carry out formula (3) (PMV＞＞mv_shift)＜＜processing of mv_shift.

The device of Figure 33 and Figure 34, be made of the formula (3) of carrying out embodiment 2 and the circuit of (5) respectively, the decision of (mv_shift) value that is shifted imports (obtaining when the motion vector detection) according to the outside and the some of conditions such as type of macro block determines, controls.

The device of Figure 35 and Figure 36 is made of the formula (3) of carrying out embodiment 2 and the circuit of (5) respectively, and the decision of (mv_shift) value that is shifted only waits according to the type of macro block and controls.

The device of Figure 37 and Figure 38 is made of the formula (4) of carrying out embodiment 2 and the circuit of (6) respectively, and the decision of (mv_shift) value that is shifted is only imported (obtaining when the motion vector detection) according to the outside and controlled.

The device of Figure 39 and Figure 40, be made of the formula (4) of carrying out embodiment 2 and the circuit of (6) respectively, the decision of (mv_shift) value that is shifted imports (obtaining when the motion vector detection) according to the outside and the some of conditions such as type of macro block determines, controls.

The device of Figure 41 and Figure 42 is made of the separate equations (4) of carrying out embodiment 2 and the circuit of (6) respectively, and the decision of displacement (mv_shift) value determines, controls according to the some of conditions such as type of macro block.

＜embodiment 5 〉

Figure 18 is the process chart of the coding method of embodiment 5.

Present embodiment 5 is in order to solve the problem that in the motion compensation of the filter that has used long tap number storage access zone enlarges, and the coding-decoding method that the precision of encoding motion vector is switched.

The predicted value of using long filter to generate is pixel precision or its following predict pixel.Thereby, be defined as integer pixel by encoding precision with motion vector, just can cut down the storage access zone.But, the encoding precision of motion vector is defined as integer pixel, estimated performance will be reduced.Therefore, in present embodiment 5, device is realized or software is realized effect, limited pixel precision according to the expansion of storage access scope.Particularly only under the less situation of the block size when motion compensation, the encoding precision of constrained motion vector.

The influence that the associative memory accessing zone enlarges, block size is more little just to become big more.The pixel count that the less piece of size carries out access originally is just few.For this reason, if the relatively piece that size is bigger and little piece, then for the ratio of the expansion accessing zone of original access range, block size is more little then big more.

The code device method of present embodiment is being cut apart the piece of Fig. 7 under the situation of the motion compensation of carrying out 1/4 pixel precision, as shown in Figure 18, at first carries out block size and differentiates (handling 801), selects the mv_shift value.Under the situation of present embodiment, in 8 * 8Partition2 (8 * 4) and 8 * 8Partition3 (4 * 8), select 1/2 pixel precision (handling 811), among 8 * 8Partition3 (4 * 4), select integer-pel precision (handling 812), in 8 * 8 pixels and size is bigger more than this piece selection 1/4 pixel precision (processing 813 of Figure 18), and drive dextroposition handling part 103 with selected mv_shift and carry out motion compensation.When the coding of motion vector, establishing mv_shffi in the piece of 8 * 4 and 4 * 8 pixel sizes is 1, and establishing mv_shffi in the piece of 4 * 4 pixel sizes is 2, and establishing mv_shift in the piece of in addition pixel size is 0.

In addition, in the mode of using P-picture and B-picture, also can only be suitable for above-mentioned precision and switch, in P-picture, how all carry out motion compensation with 1/4 pixel precision regardless of block size to B-picture.In addition,, only will be made as 4 * 4, also it can be expanded to 8 * 4 and 4 * 8 with the block size that integer-pel precision is carried out motion compensation although here.In the case, establishing mv_shift in the piece of 8 * 4,4 * 8 and 4 * 4 pixel sizes is 2, and establishing mv_shift in the piece of in addition pixel size is 0.

According to the code device of the moving image of present embodiment and the encoding section and the lsb decoder of the motion vector in the decoding device, similarly constitute with above-mentioned Figure 41 and Figure 42 respectively.

＜embodiment 6 〉

Figure 19, Figure 45 and Figure 46 are respectively the coding method process chart of embodiments of the invention 6, the major part structure chart of code device and the major part structure chart of decoding device.

Present embodiment is also in order to solve the problem that in the motion compensation of the filter that has used long tap number storage access zone enlarges, and by with different encoding precision each composition of vertical-horizontal of motion vector being encoded, comes the limits storage access range.

The displacement selection portion 124 of code device as shown in Figure 19, judges that whether motion vector is vertical composition (handling 801), and select mv_shift.Type adaptation portion 124 drives dextroposition handling part 103 with selected mv_shift and carries out motion compensation.In the present embodiment, motion compensation by 1/4 pixel precision, the vertical one-tenth of motion vector is divided into integer-pel precision (handling 812), level is become to be divided into 1/4 pixel precision (handling 813), if the mv_shift when vertical composition is carried out coding/decoding is 2, the mv_shift that establishes when horizontal composition carried out coding/decoding is 0.

The method effect that the encoding precision of the vertical composition of the motion vector that the influence to the storage access expanded range of present embodiment is bigger limits is very big.View data generally is that the order by raster scan is kept at the memory towards the bottom right from upper left.For this reason, under situation about beginning from certain pixel adjacent pixels expansion access range, 1 pixel of horizontal direction means 1 pixel degree, and 1 pixel of vertical direction then means 1 stroke degree.Thereby, follow number to the storage access of vertical direction, bigger than going back of horizontal direction.And then, not with byte unit's (1 pixel) but preserving with word unit's (per 2 pixels) or dword unit's (per 4 pixels) in the implementation method of pixel data of 1 pixel 1byte, because expansion to the storage access scope of horizontal direction, take place with word unit or dword unit, so its influence is not too big.With respect to this, even in the access of 1 pixel degree of vertical direction, the storage access scope also must enlarge.Thereby to the method that the encoding precision of the motion vector of vertical direction limits, it is resultful dwindling such viewpoint with the storage access scope.

In addition, also the comprehensive method of embodiment 5 and embodiment 6 is carried out processing as shown in Figure 23.That is, displacement selection portion 124 only to the piece of 8 * 4,4 * 8 and 4 * 4 pixel sizes, is constrained to integer pixel with the vertical composition of motion vector.And then according to block size, the means of switching the vector composition that scope is limited also are effective.

As described above, the influence that the associative memory accessing zone enlarges, block size is more little then big more, and this also is said to the aspect rate of piece.Just, under the different situation of the big or small in length and breadth ratio of piece, the influence that the composition access range that size is more little enlarges just becomes big more.If cut apart with the piece of Fig. 7 and to consider, then we can say to size to be that the composition of 4 pixels applies that to limit be effective.

＜embodiment 7 〉

Figure 24, Figure 43 and Figure 44 are respectively the coding method process chart of embodiments of the invention 7, the major part structure chart of code device and the major part structure chart of decoding device.

Present embodiment, code device (Figure 43), the shift value selection portion 124 of decoding device (Figure 44) is such shown in the process chart of Figure 24, when block size is 8 * 4 (vertical * horizontal stroke), level one-tenth is divided into integer-pel precision (mv_shift=2) and will vertically becomes to be divided into 1/4 pixel precision (mv_shift=0), when block size is 4 * 8 (vertical * horizontal stroke), level one-tenth is divided into 1/4 pixel precision (mv_shift=0) and will vertically becomes to be divided into integer-pel precision (mv_shift=2), when block size is 4 * 4 (vertical * horizontal stroke), the horizontal vertical composition all is made as integer-pel precision (mv_shift=2).In addition, from the viewpoint of estimated performance, also can replace the processing of the horizontal vertical composition of 8 * 4 and 4 * 8 modes.This results from the more little high effective more such notion of Forecasting Methodology of composition precision of size.Be to be suitable for which method, then the necessity that limits according to storage access, coding efficiency, bit rate etc. should be used for determining.

＜embodiment 8 〉

Figure 20, Figure 47 and Figure 48 are respectively the coding method process chart of embodiments of the invention 8, the major part structure chart of code device and the major part structure chart of decoding device.Present embodiment is in embodiment 7, is interlace signal at input picture, and macro block is under the situation of frame structure, the vertical composition of motion vector is constrained to the method for even pixel precision.

Figure 17 represents interlace signal to be carried out macro block under the situation of coding/decoding with frame structure.Solid line on the figure means odd field (top field), and dotted line means even field (bottomfield).The motion vector of odd number value is as shown in Figure 17, and the even field from the parametric image generates the predicted value on the odd field, and the odd field from the parametric image generates the predicted value on the even field.In interlace signal, because odd field shows that with even field the moment is different, so in fact the vertical one-tenth resolute of odd number value does not produce.

Therefore, in the present embodiment, be interlace signal to input picture, macro block is the situation of frame structure, and the vertical composition of motion vector is constrained to the even pixel precision.In coding/decoding device, the mv_shift that establishes when vertical composition carried out coding/decoding is 3 (processing 803,814 of Figure 20), and the mv_shift that establishes when horizontal composition carried out coding/decoding is 0 (processing 813 of Figure 20).Even under the frame structure situation of this interlace signal, also can be shown in the flow chart of Figure 21 and Figure 22 like that, make up with the method for embodiment 7, only, the vertical composition of motion vector is constrained to integer pixel the piece (processing 801 of Figure 21) of 8 * 4,4 * 8 and 4 * 4 pixel sizes.And then, as the coded system of interlace signal, the accommodation mode to each macro block switch frame structure, a structure and frame structure and a structure is arranged, also can change the method for selecting of the encoding precision of motion vector according to this method.For example, in the coding specification of handling interlace signal and progressive signal, no matter how all identical the coded system that the frame structure is arranged signal scanning situation.In the case, also can be in a frame structure and a structure, the piece less to size, the encoding precision of the motion vector of vertical direction is defined as integer-pel precision carries out coding/decoding (processing 812 of Figure 20 and Figure 21 and Figure 22), in accommodation mode to each a macro block switch frame structure and a structure, the piece less to size, encoding precision with the motion vector of vertical direction in frame structure is defined as even pixel precision (processing 814 of Figure 20 and Figure 21), is defined as integer-pel precision in the structure on the scene and carries out the such method of coding/decoding (processing 812 of Figure 20 and Figure 21).In addition, at this moment also can be in frame structure not the size of plumber block how all the encoding precision of the motion vector of vertical direction is defined as even pixel precision (processing 814 of Figure 22), in the structure on the scene the less piece of size is defined as integer-pel precision and carries out the such method of coding/decoding (processing 801 of Figure 22 and handle 812).

In the storage access scope reduction mode of

embodiment

7 and 8, do not need the coding/decoding of mv_shift value to handle.Thereby, just do not need MVD Prediction13 that Fig. 1, Fig. 2 dash and the mv_shift among Figure 11, Figure 12.In addition, in Figure 13, just do not need the mv_shift value is passed to from motion compensator 211 processing of demultiplexing device 206.And then, in Figure 14, just do not need the mv_shift value is deciphered the processing that device 501 is passed to MV fallout predictor 508 from code.The mv_shift value is determined by MV fallout predictor 508 according to prediction mode.

In addition, the storage access scope reduction mode of present embodiment 8 is, the means that use is switched the precision of encoding motion vector and predictive coding motion vector " are defined as integer pixel according to block size with the encoding precision of motion vector ", " encoding precision of the vertical composition of motion vector is defined as integer pixel or even pixel ", block size about the encoding precision of the kind of the motion vector encoder precision of the precision of motion compensation, preparation and motion switch vector is not limited to the foregoing description.In addition, making in the motion compensation mode that the storage access scope reduces being defined as 1/2 pixel precision by the precision with motion vector, is not motion vector to be defined as integer-pel precision but the method that is defined as 1/2 pixel precision also can realize in enough the present invention.

The present invention is transmitting, in the storage, the device that shows moving image, system, encoding image information need become in the field on the industry of technology of digital signal of size of code still less to be utilized.

Claims

1. the coding method of a moving image has following steps,

It is characterized in that: this method also has following steps,

2. the coding method of moving image according to claim 1 is characterized in that: also have following steps,

Prepare the kind of a plurality of block types as encoding block, and the motion vector number in the precision of selected differential motion vector and the encoding block and corresponding to their block type of combination, wherein block type comprise expression the precision of differential motion vector is encoded as additional information or be specified to the identifying information of fixing precision and the number of the motion vector that under these a plurality of block types, should encode inequality;

Under the situation of representing the precision of differential motion vector is encoded as additional information at the block type of encoding block, the information of block type and selected precision and the differential motion vector of this precision are encoded;

Under the situation of representing differential motion vector to be encoded at the block type of encoding block, the differential motion vector of block type and specified accuracy is encoded with the precision of regulation.

3. the coding method of moving image according to claim 1 is characterized in that: also have following steps,

Preparation is different multiple little block-shaped of the number of the motion vector of piece planted agent coding, whether decision encodes the precision of precision, fritter shape kind and the above-mentioned differential motion vector of above-mentioned differential motion vector as additional information, the identifying information that each block encoding and output are determined uniquely according to above-mentioned decision.

4. the coding method of moving image according to claim 1 is characterized in that: also have following steps,

Preparation is different multiple little block-shaped of the number of the motion vector of piece planted agent coding, above-mentioned multiple little block-shaped for each, regulation fixes or is made as variable value with the differential motion vector precision with default value, and whether decision encodes the precision of precision, fritter shape kind and the differential motion vector of above-mentioned differential motion vector as additional information; And

Encode to above-mentioned default value with according to the identifying information that above-mentioned decision is determined uniquely, and the identifying information behind the above-mentioned coding is included in the coded data exports.

5. the coding method of a moving image has following steps,

It is characterized in that: this method also has following steps,

6. the coding method of moving image according to claim 5 is characterized in that: also have following steps,

Prepare the kind of a plurality of block types as encoding block, and the motion vector number in the precision of selected motion vector and the encoding block and corresponding to their step of block type of combination, wherein block type comprise expression the precision of motion vector is encoded as additional information or be specified to the identifying information of fixing precision and the number of the motion vector that under these a plurality of block types, should encode inequality;

Under the situation of representing the precision of motion vector is encoded as additional information at block type, encode to the information of block type, selected precision and by the differential motion vector that the motion vector computation of motion vectors that is modified to selected precision and selected precision goes out;

Under block type is represented precision situation that motion vector is encoded with regulation, encode to block type and by the differential motion vector that the motion vector computation of motion vectors that is modified to specified accuracy and specified accuracy goes out.

7. the coding method of moving image according to claim 5 is characterized in that: also have following steps,

Preparation is different multiple little block-shaped of the number of the motion vector of piece planted agent coding, and whether decision encodes the precision of precision, fritter shape kind and the above-mentioned motion vector of above-mentioned motion vector as additional information;

To each piece under the situation that the precision of motion vector is encoded as additional information, to encoding and export according to the information of the well-determined identifying information of above-mentioned decision, selected precision and by the differential motion vector that the motion vector computation of motion vectors that is modified to selected precision and selected precision goes out;

Under the situation of precision of motion vector not being encoded as additional information, the differential motion vector that goes out according to the well-determined identifying information of above-mentioned decision, by the motion vector computation that is modified to the motion vectors of the corresponding precision of this identifying information and this precision is encoded and exported.

8. the coding/decoding method of a moving image has following steps,

It is characterized in that: this method also has following steps,

9. the coding/decoding method of moving image according to claim 8 is characterized in that: also have following steps,

Prepare the kind of a plurality of block types as decoding block, and the block type of the piece that should decode decodes, wherein block type comprise expression the precision of differential motion vector is decoded as additional information or be specified to the identifying information of fixing precision and the number of the motion vector that under these a plurality of block types, should decode inequality;

Under the situation of representing the precision of differential motion vector is decoded as additional information in the decoding block type, the information of precision and the differential motion vector of this precision are decoded;

Under the situation of representing differential motion vector to be decoded in the decoding block type, the differential motion vector of specified accuracy is decoded with the precision of regulation;

After the precision of the differential motion vector of will decoding is modified to the precision identical with motion vectors, the motion vector that calculates decoding block from the decoding differential motion vector of carrying out the precision correction and motion vectors.

10. the coding/decoding method of a moving image has following steps,

It is characterized in that: also have following steps:

11. the coding/decoding method of moving image according to claim 10 is characterized in that: also have following steps,

Prepare the kind of a plurality of block types as decoding block, and block type decoded, wherein block type comprise expression the precision of motion vector is decoded as additional information or be specified to the identifying information of fixing precision and the number of the motion vector that under these a plurality of block types, should decode inequality;

Under the situation of representing the precision of motion vector is decoded as additional information in the decoding block type, the information of precision and the differential motion vector of this precision are decoded;

Under the situation of representing motion vector to be decoded in the decoding block type, the differential motion vector of specified accuracy is decoded with the precision of regulation;

Calculate the motion vector of decoding block from decoding differential motion vector and the motion vectors of carrying out the precision correction.