Predicting motion vector of current image section indicating reference zone overlapping multiple sections of reference image, encoding and decoding using said prediction

FIELD: physics, computer engineering.

SUBSTANCE: invention relates to means of encoding and decoding images. In the method, the motion vector of a reference section has the same shape as the current section and belongs to a reference image which is different from the current image and is broken down in advance as a result of encoding with subsequent decoding into a plurality of sections. When a reference section overlaps a set of reference sections from said plurality of sections of the reference image, said motion vector of the current image section is determined based on a reference motion vector function belonging to a set of reference motion vectors associated with k overlapped reference sections.

EFFECT: high accuracy of predicting the motion vector of an image section.

15 cl, 6 dwg

The level of technology

The present invention generally relates to the field of image processing and, in particular, to encoding and decoding by comparing digital images and sequences of digital images.

To transfer images there are several methods of encoding and decoding. In particular, there are large types of coding, such as coding, called "intra"when the image encode offline, i.e. without reference to other images, or encoding, called "inter", which is the encoding of the current image relative to the previous images to Express and transmit only the differences between these images.

Typically, the encoding method of the above type include stage predictive coding, in which parts of the image, called blocks or macroblocks, predict the current image relative to the other thrust blocks or macroblocks, that is previously encoded and then decoded blocks or macroblocks.

For example, in the case of the use of standard H264/MPEG-4 AVC (Advanced Video Coding") predictive coding macroblock consists in the breakdown of macroblocks on a variety of topics, which usually have the form of blocks of smaller size.

In private the tee, in the case of inter-coding according to the above standards used for coding the current macroblock can be divided into sections according to options 16×16, 8×16, 16×8 and 8×8. If you choose option 8×8, each block of 8×8 is again divided into sections on options 8×8, 4×8, 8×4 and 4×4. Each current block is compared with one or more blocks, respectively, one or more reference images. Thus, receive the motion vector describes motion between the current block and the reference block, which occupies the same position as the current macroblock in the previous image. This calculates the predicted value of the motion vector to encode the balance between the above-mentioned motion vector and the calculated predicted motion vector.

Such a prediction motion vector is not suitable for all types of partitioning and, in particular, to the case when the reference macroblock overlaps several reference sections reference image. This situation is shown in figa, which illustrates the case of the provisional prediction to encode the current macroblock indicated by MBC_Nimages N designed for encoding according to the above standards. In the presented example of such a macroblock MBC_Nhas a square shape and it matches the type of 4×4. Macroblock MBC_NROC the wives of other macroblocks BR1, BR2, BR3, BR4, which are in the immediate vicinity and have the same shape and size as the macroblock MBC_N.

In the present example, the motion vector of the current macroblock MBC_Ndenoted by MV, indicates the reference macroblock MBC_N-1the reference image is denoted by N-1, which is, for example, the previous image. Reference MBC_N-1has the same position as the current macroblock MBC_Nin the previous image N-1. Feature reference macroblock MBC_N-1is that it overlaps the already encoded and then decoded reference sections, denoted by BR'1, BR'2, BR'3 and BR'4 figa.

In accordance with the standard H264/AVC the above-mentioned motion vector MV predict only spatial. In particular, calculate the reference motion vector, which is equal to the average of the motion vectors MV1, MV3, MV4, respectively associated with the reference macroblocks BR1, BR3, BR4. In some situations, the motion vector MV2 associated with the reference macroblock BR2, can be used instead of one of the vectors MV1, MV3, MV4.

In addition, recently, new types of partitioning intended for encoding the current macroblock, which were not provided by the standard H264/AVC. As shown in figv encoded current MBC_Ncan be divided into several section is in P1-PP, having a linear shape, G-shape or an arbitrary shape.

Standard H264/AVC does not provide predictions for different types of partitioning, shown in figv, and for a particular case, when broken thus the reference macroblock overlaps several sections of the reference image. This situation presents on pigs, which illustrates the case of the provisional prediction to encode the current macroblock indicated by MBC_Nimages N designed for encoding according to the above standards. In the presented example of such a macroblock MBC_Ndivided into several smaller partitions P1, P2, P3, which have any geometric shape.

In the present example, the motion vector of the first partition P1 of the current macroblock MBC_Nmarked MVp1, point to the partition P'1 of the reference macroblock MBC_N-1in the previous image N-1. Feature reference macroblock MBC_N-1is that it overlaps the already encoded and then decoded reference sections, denoted by BR'1, BR'2, BR'3 and BR'4 figs.

In accordance with the standard H264/AVC to predict the above-mentioned motion vector MVp1, calculate the reference motion vector, which is usually equal to the spatial average of the motion vectors MV1, MV3, MV4, with the NGOs associated with the reference macroblocks BR1, BR3, BR4.

Such spatial prediction of motion vector may be inaccurate, since the image N-1, there are differences in shape and size between the reference section R'1 and reference macroblocks BR'1, BR'2, BR'3 and BR'4.

In addition, other known methods of calculating the predicted motion vector with the purpose of inter-coding of the current macroblock.

One of them is described in the publication IEEE Transactions on Circuits and System for Video Technology, volume 18, 1247-1257 (September 2008), G.Laroche, J.Jung and .Pesquet-Popescu, and true when, as in the standard H264/AVC, the macroblock is divided into many sections, which usually have the form of blocks of smaller size. According to this method, predicting the motion vector of the macroblock of the current image relative to the reference vector that is chosen as the vector pointing to the pixel at the top extreme left of the macroblock having the same position as the current macroblock in the previous image.

If you try to use this last method for predicting vector MV on figa or for prediction vector MVp1 on figs, each of the vectors MV and MVp1 is obtained on the basis of the reference vector MV2 associated with the reference macroblock BR'2, with the leftmost pixel of the reference macroblock MBC_N-1is the reference macroblock BR'2, associated with the motion vector MV.

The prediction motion vector using this method too is not exact for the same reasons that were stated above.

Task and disclosure of inventions

One of the objectives of the present invention is to eliminate the above disadvantages of the prior art.

In this regard, the first object of the present invention is a method of predicting a motion vector of the current partition of the image relative to the motion vector of the reference section, having the same form as the current section, and belonging to a reference picture different from the current image and the previously broken after encoding and subsequent decoding on the set of n partitions.

According to the invention, in the case when the reference section covers a set of k reference partition of the set n of sections of the reference image, where k≤n, the motion vector of the current partition image is determined on the basis of the function of at least one reference motion vector belonging to k than k reference motion vectors respectively associated with k overlapped reference sections.

This distinctive feature makes it possible to significantly improve the prediction accuracy, given:

- a special partitioning of the current macroblock or a reference macroblock,

only PE ecrivaine zone support partition.

In addition, the prediction in accordance with the present invention can be applied to any method of calculating the predicted motion vector of the current partition, for example, for the method according to the standard H264/AVC or to the method described in the link above IEEE.

In an embodiment, the determining of the motion vector of the current section contains the following stages:

- calculate the total number of pixels between the reference section and k respectively overlapping the supporting sections

- comparison of the calculated total number of pixels between the overlapped k reference sections on the basis of a predetermined criterion of comparison.

This distinctive feature allows you to accurately choose the reference motion vector on the basis of certain characteristics, which in this case is based on the number of common pixels between the reference section and overlapping the supporting sections.

According to the first variant, the criterion of comparison is the choice of k reference motion vectors respectively associated with k overlapping the supporting sections, the reference motion vector associated with the reference section, which calculated the total number of pixels is the largest.

According to the second variant, the criterion of comparison consists in weighting, using the calculated number of total pixels, Wed is the last value of k reference motion vectors, respectively associated with k overlapped reference sections.

Under the third option, the criterion of comparison is the choice of k reference motion vectors respectively associated with k overlapping the supporting sections, the reference motion vector that is associated with the overlapped reference section with more pixels within the reference section than outside of this reference section.

In another embodiment, the determining of the motion vector of the current section contains the following stages:

the calculation for each k, the overlap of the reference partition coefficient, which is a function of the spatial gradient of the reference section,

the choice of the coefficient having the largest calculated value

the choice of the reference motion vector, which corresponds to the overlap of the reference section with the selected ratio.

This distinctive feature allows you to accurately choose the reference motion vector based on a certain characteristic, which in this case is based on calculating the coefficient expressing the degree of reliability in the choice of the reference motion vector, which is assumed to be more accurate in the area of the image containing stutter than in the area of the whole image.

According to a variant implementation, the motion vector of the current partition determined in accordance with the ATA stage of calculating the average value of k reference motion vectors, respectively associated with k overlapped reference sections, this average value is weighed using k calculated coefficients.

In yet another embodiment, the determining of the motion vector of the current section contains the following stages:

- the selection of a single point of reference section

the choice of the reference motion vector associated with the overlapped reference section containing the selected single point.

This distinctive feature allows you to accurately choose the reference motion vector based on a certain characteristic, which in this case is based on the assessment of the location of the supporting section relative to k, the overlap of the reference sections.

According to another variant implementation, the determination of the motion vector of the current section contains the following stages:

- identification of the reference section characteristics associated with the image content

the choice of the reference motion vector associated with the overlapped reference section that contains the specified feature.

This distinctive feature allows you to accurately choose the reference motion vector based on a certain characteristic, which in this case is based on the identification of pattern, color, contour, etc. in the reference section.

The second object of the invention is a method encoded in the I image or image sequence, which generate a data stream containing data characterizing at least one partition image, the method includes a step of predicting a motion vector for the partition of the image.

According to the invention, the step of predicting this method of encoding is carried out according to the above method of prediction.

The third object of the invention is a method of decoding a stream of data describing the image or sequence of images, the stream contains data characterizing the at least one partition image, the method includes a step of predicting a motion vector for the partition of the image.

According to the invention, the step of predicting this method of decoding is carried out according to the above method of prediction.

Accordingly, the fourth object of the present invention is a device for predicting a motion vector for the current section of the image relative to the motion vector for the reference section, having the same form as specified in the partition, and belonging to a reference picture different from the current image and the previously broken after encoding and subsequent decoding in a multiple n of the sections.

According to the invention, in the case when the reference section covers sookun is here from k reference partition of the set n of sections of the reference image, where k≤n, such a prediction device includes a computing module configured to determine the motion vector of the current partition image based on the function of at least one reference motion vectors belonging to a set of k reference motion vectors respectively associated with k overlapped reference sections.

Accordingly, the fifth object of the invention is a device for encoding an image or a sequence of images generating a data stream containing data characterizing at least one partition image, with such a device includes means for predicting a motion vector of a partition of the image.

According to the invention, the device of the predictions of such a device coding complies with the above prediction device.

Accordingly, the sixth object of the invention is a device for decoding a stream of data describing the image or sequence of images, and the data stream contains data characterizing the at least one partition image, with such a device includes means for predicting a motion vector for the partition of the image.

According to the invention, the device of the predictions of such a device decoding corresponds vishey is related to the prediction device.

The object of the invention is also a computer program containing commands to apply one of the methods in accordance with the present invention, when it takes the computer.

The encoding method, the method of decoding a prediction device, the encoding device and the decoding device have at least the same advantages as the advantages provided by the method of prediction in accordance with the present invention.

Brief description of drawings

Other distinctive features and advantages of the invention will be more apparent from the following description of the preferred options of execution with reference to the accompanying drawings, on which:

figa - known example of a temporary predictions, which use temporal relationship between allocated for encoding the current macroblock image N and the reference macroblock in the previous image N-1, while the reference macroblock has a square shape and covers several adjacent reference macroblocks;

figw - macroblock, divided into sections of different types according to known solutions;

figs - known example of a temporary predictions, which use temporal relationship between allocated for encoding the current macroblock image N and the reference macroblock of the previous and what the considerations applying N-1, when the reference macroblock is divided into several sections of arbitrary shape and covers several adjacent reference macroblocks;

figure 2 - stages of the encoding method in accordance with the present invention;

figure 3 embodiment of the encoder in accordance with the present invention;

figure 4 is an example of temporal prediction in accordance with the present invention, in which use temporal relationship between allocated for encoding the current macroblock image and the reference macroblocks of the previous image;

5 is a decoding device in accordance with the present invention;

6 - stages of a method of decoding in accordance with the present invention.

The implementation of the invention

What follows is a description of a variant implementation of the invention in which is used a method of encoding in accordance with the present invention for inter-coding a sequence of images as a binary stream, close to the stream obtained according to the standard H264/MPEG-4 AVC. In this embodiment, the encoding method in accordance with the present invention is carried out using software or hardware through changes in the encoder, originally corresponding to the standard H264/MPEG-4 AVC. The encoding method in accordance with h is a worthwhile invention is presented in the form of the algorithm, containing phases C0-C7, as shown in figure 2.

The encoding method in accordance with the present invention carried out using the encoder CO, shown in figure 3.

The first stage C0, shown in figure 2, is a choice for a macroblock belonging to the image intended for encoding an image sequence, denoted by I_Nfigure 3 - the particular partitioning corresponding macroblock.

It should be noted that the phase C0 may be optional, and the prediction motion vector of the current macroblock can be performed, considering the last one in its entirety, that is, as the only and single partition.

During phase C0 macroblock MB_Nfor example, having a size of 4×4 and belongs to the image of I_N, is directed to the input module SP choice partitioning shown in figure 3.

In this module SP binning is used, for example, the method of choice by redundant mapping or selection method using the apriori algorithm. Such methods are well known in the art (see: G.F.Sullivan and T.Wiegand, "Rate-distorsion optimization for video compression", IEEE Signal Proc. Mag., p.74-90, 1998). Therefore, their description is omitted.

Different types of algorithms partitioning grouped in the database BD encoder CO. They allow the breakdown current is it macroblock into multiple sections or rectangular or square shape, or other geometric shapes, such as a substantially linear shape, or completely free-form.

In the present example, the selection module SP selects the partitioning of an arbitrary shape.

The next step S1, as shown in figure 2, represents a partitioning of the macroblock MB_Non the number of p are designed for predicting the partition.

For example, the macroblock MBn is divided into three sections P1, P2 and P3 of arbitrary shape. This breakdown provides a module RMSO breakdown of macroblocks shown in figure 3, which uses classical algorithm partitioning.

Figure 4 shows the macroblock MB_Nreceived after this breakdown.

After step S1 partitioning during step S2, shown in figure 2, the module breakdown RMSO passes just partitioned macroblock MB_Nin the module PREDCO predictions, shown in figure 3.

Classically this module PREDCO prediction is to predict the chaptered current macroblock MB_Nabout the already encoded and then decoded macroblock indicated by MBr_N-1in figure 4, which has the same position as the current macroblock MB_Nin the previous image I_N-1which was previously broken after encoding and subsequent decoding on many n section is in r'1, r'2, ..., R n.

According to the invention, the reference macroblock MBr_N-1covers a set of k reference sections r'1, r'2, ..., R k, for k≤n. In the present example, the reference macroblock MBr_N-1partially overlaps the four reference section r'1, r'2, r'3 and r'4. Of course, other possible versions of the reference macroblock MBr_N-1can completely block one or more anchor sections r'1, r'2, r'3 and r'4.

As shown in figure 3, such a reference macroblock MBr_N-1code according to the standard H264/MPEG-4 AVC, that is, as is well known, it is subjected to:

coding by discrete cosine transform and quantization that is performed by the module TQCO transformation and quantization,

- then decoded by the inverse discrete-cosine transform and quantization that is performed by the module TQICO inverse transform and quantization.

As shown in the same figure 3, according to the invention, the module PREDCO prediction contains:

module RMV partitioning, designed to split the reference macroblock MBr_N-1on a set of reference topics

computing the CAL module designed to calculate each motion vector MVp1, MVp2, ..., MVpp, which are respectively associated with sections P1, P2, ..., PP of the current macroblock MBr_N-1on the basis of function, what about the least one reference motion vectors belonging to a set of k reference motion vector MVr'1, MVr'2, ..., MVr'k, which are respectively associated with k overlapped reference sections r'1, r'2, ..., R k.

During C3, shown in figure 2, the module RMV partitioning shown in figure 3, produces a breakdown of the reference macroblock MBr_N-1p reference sections. In the example shown in figure 4, the reference macroblock MBr_N-1smash is identical to the current macroblock MB_N, three section Pr'1, Pr'2 and Pr'3, which have different shape and size.

During phase C4, shown in figure 2, the computing module CAL, shown in figure 3, calculates for each current partition P1, P2 and P3 associated with the predicted motion vector MVp1, MVp2 and MVp3 by means described below, various methods in accordance with the present invention.

According to the first method, the CAL module determines a predicted motion vector MVpl the current partition P1 depending on the reference motion vector MVr'1, MVr'2, MVr'3 and MVr'4, respectively associated with four overlapping reference sections r'1, r'2, r'3 and r'4, shown in figure 4. This definition includes, for example, in the calculation of the average values of the reference motion vector MVr'1, MVr'2, MVr'3 and MVr'4 according to the following equation:

MVp1=Moy(MVr'1, MVr'2, MVr'3, MVr'4)

According to the second IU the ode, presented with reference to figure 4, the CAL module determines a predicted motion vector MVp1 as equal to the reference motion vector associated with the overlapped base section having the largest number of pixels in common with the supporting topic Pr'1 reference macroblock MBr_N-1.

In the example shown in figure 4, MVp1=MVr'2.

According to the first version of this second method, the CAL module determines a predicted motion vector MVp1 as equal to the reference motion vector associated with the overlapped base section having the largest percentage of pixels in common with the reference section, WG'1 reference macroblock MBr_N-1.

In the example shown in figure 4, MVp1=MVr'2.

According to the second variant of this second method, the CAL module determines the average value of the reference motion vector MVr'1, MVr'2, MVr'3 and MVr'4, which is weighed by the number of common pixels between the reference section Pr'1 macroblock MBr_N-1and each of the overlapped reference sections r'1, r'2, r'3 and r'4. This definition consists in calculating the predicted motion vector MVp1 according to the following equation:

$$MVP1=\frac{1}{T}\left[{\displaystyle \underset{k=1}{\overset{K}{\sum }}\lceil pr\text{'}1\cap r_k^{\text{'}}\rceil \cdot MVr\text{'}k}\right]$$

where:

- K=4,

- T is the number of pixels forming the reference macroblock MBr_N-1

-$$\lceil \mathrm{Pr}\text{'}1\cap r\text{'}k\rceil$$is the number of pixels shared between the supporting topic Pr'1 macroblock MBr_N-1and each of the overlapped reference sections r'1, r'2, r'3 and r'4,

In the example shown in figure 4, MVp1=MVr'2.

Alternative to the above average can be weighted by the number of total pixels minus the number of common pixels.

Another alternative consists in weighting the average of the percentage of pixels in common between the topic Pr'1 reference macroblock MBr_N-1and each of the reference sections of r'1, r ' 2, r'3 and r'4.

Another alternative is the determination of the predicted motion vector MVp1 as equal to the reference motion vector associated with the overlapped reference section, which has more pixels inside Pr'1 than outside Pr'1.

According to the third method, is presented with reference to figure 4, the computing module CAL:

- defines, for each k, the overlap of the reference sections of r'1, r'2, r'3 and r'4, - coefficient of C_k(k=4), the cat is which is a function of the spatial gradient g of the specified reference section Pr'1,

- selects the coefficient Cj (1≤j≤k), the calculated value which is the largest, according to the following equation:

$$C_j=\underset{k}{\arg \max }\left\{C_k\right\}$$where$$(C_k=\frac{1}{\lceil \mathrm{Pr}\text{'}1\cap r_k^{\text{'}}\rceil }{\displaystyle \underset{i=1}{\overset{\lceil \mathrm{Pr}\text{'}1\cap r_k^{\text{'}}\rceil }{\sum }}\sqrt{g_x^2\left(i\right)+g_y^2\left(i\right))}}$$

This third method involves first alternative, according to which computing the CAL module determines a predicted motion vector MVp1 as equal to the reference motion vector associated with the overlapped reference section corresponding to the computed coefficient C_j.

This third method involves the second al is ternative, according to which computing the CAL module determines a predicted motion vector MVp1 equal to the average value of the reference motion vector MVr'1, MVr'2, MVr'3 and MVr'4, which is weighed by using the calculated coefficients C₁C₂C₃With₄.

According to the fourth method, presented with reference to figure 4, the computing module CAL primarily used to identify a single point of reference section Pr'1, for example, its center is denoted by CTr'1. Center CTr'1 is calculated using an algorithm that minimizes the sum of distances with respect to all points in the reference section Pr'1.

After computing the CAL module determines a predicted motion vector MVpl as equal to the reference motion vector associated with the overlapped reference section that contains the selected single point, i.e. the center CTr'1. In the present example, the center CTr'1 is contained in the reference section r'2, and, therefore, MVp1=MVr'2.

According to the fifth method, presented with reference to figure 4, the computing module CAL first identifies in the reference section Pr'1 individual characteristic associated with the image content. In the present example, such a characteristic is shown a cross inserted in the image of I_N-1and marked CIr'1.

Then computing the CAL module determines the forecast is hydrated motion vector MVp1 as equal to a reference motion vector, associated with the overlapped reference section that contains the selected characteristic, that is, the cross CIr'1. In the present example, the cross CIr'l is contained in the reference section r'1, and, therefore, MVp1=MVr'1.

Alternatively, this fifth method of characteristics associated with the image content can be a single color, single picture, a path which intersects the reference section Pr'1, or any other feature of the image that violates its homogeneity.

Upon completion of this step C4 calculations on one or another of the above methods in accordance with the present invention the computing module predictions PREDCO generates a first predicted vector MVp1, which when selected by the encoder CO as the optimal motion vector immediately coded module TQCO transformation and quantization, and then decoded by the module TQICO inverse transformation and quantization, which are presented in figure 3.

Then the above step S4 is repeated to predict other motion vectors MVp2 and MVp3, which are respectively associated with sections P2 and P3 of the current macroblock MB_N.

After a variety of possible predictions calculated by a compute module predictions PREDCO, at step S5, shown in figure 2, the module DCNCO decision, shown in figure 3, looking at the broken the and sections of the macroblocks of the image I _Nand selects at this stage C5 variant prediction to encode each of these macroblocks. Among the possible predictions for the macroblock module decision DCNCO selects the optimal prediction according to the criterion of the degree of distortion is well known to specialists.

As shown in figure 2, each predicted macroblock encode at step C6, as specified in the standard H264/MPEG-4 AVC.

As shown in figure 3, after the implementation of structural encoding module DCNCO decision factors residues, if they exist, correspond to the blocks of the image I_Nsend in the module TQCO transformation and quantization, where they are subjected to discrete cosine transform, and then the quantization. After that, lots of macroblocks with quantized coefficients are passed to the module CE entropy coding to generate, together with other images of the sequence, already encoded the same way as the image of I_N- the binary stream video F, encoded in accordance with the invention.

Encoded binary stream F is passed through the communication network to the remote terminal. It contains the decoder DO in accordance with the present invention, is shown in figure 5.

First binary stream F is sent to the module DE entropy decoding, which is the inverse Kadirova the Oia, implemented by the module CE entropy encoding, shown in figure 3. Then, for each macroblock reconstructing image coefficients decoded by the module DE, sent to the module QTIDO inverse quantization and transformation.

Module RI playback image receives the decoded data corresponding to the data produced by the module DCNCO (3) in step C5 coding in accordance with the present invention, taking into account transmission errors. Module RI performs the stages D0-D6 of the method of decoding in accordance with the present invention, shown in Fig.6. This method of decoding in accordance with the present invention can also be implemented using software or hardware modifications to the decoder, corresponding to the original standard H264/MPEG-4 AVC.

The first stage DO is a decoding patterns encoded data in the first area of the current macroblock decoded image I_N. As is known, the module RI playback determines on the basis of data of the specified area of the macroblock:

- the encoding type of the specified data, Intra or Inter: in accordance with the present invention is Inter-coding,

- type partitioning playing macroblock, Inter 4×4, 8×8, linear, etc.: in the described embodiment, the Inter 4×4,

- the index of the optimal predictor, the selected module DCNCO decision-making at the above step S5.

The next step D1 shown in Fig.6, is a breakdown intended for decoding the current macroblock in accordance with the partitioning determined at step D0. For this purpose, as shown in figure 5, the module PMBDO dividing the macroblock into partitions, which is absolutely similar to the module shown in figure 3, divides the macroblock on the set p of partitions, i.e. three sections of arbitrary shape in the present example.

During D2, shown in Fig.6, the module PMBDO partitioning passes intended for decoding the current macroblock that has just been split into p=3 sections in the module PREDDO predictions, shown in figure 5, which is identical to module PREDCO the predictions of the encoder, shown in figure 3, and therefore further detailed description is omitted.

During D3 and D4 shown in Fig.6, the module PREDDO predictions, shown in figure 5, performs the same algorithm as the module PREDCO the above encoder CO to get the current macroblock, the corresponding motion vectors which were predicted according to one or another of the methods described above.

At step D5 module DCNDO decision selects the optimal prediction under criterion degree of IP is Azania, well-known to experts.

After that, at step D6 each predicted macroblock decode as it is provided by the standard H264/MPEG-4 AVC.

After decoding all of the macroblocks of the image I_Nas shown in figure 5, the module RI image playback outputs of the decoder DO the image ID_Ncorresponding to the decoding image I_N.

Given that the prediction is performed by the decoder DO, according to all indicators is similar to the algorithm implemented by the coder CO, cost information, depending on the used predictors significantly reduced.

Of course, the described embodiments of the provided solely as examples and are not restrictive, and the specialist may make various changes without departing from the scope of the invention.

1. Way prediction motion vector of the current partition of the image relative to the motion vector of the reference section, having the same form as the current section, and belonging to a reference image different from the current image and the previously broken by the encoding and subsequent decoding on the set n of partitions (r'1, r'2, ..., R n), containing the stage at which the case, when the specified reference section covers a set of k reference section the crystals from the specified set of n sections (r'1, r'2, ..., R n) of the reference image, where k≤n, define the specified motion vector of the current partition image based on the function of at least one reference motion vectors belonging to the aggregate of the k reference motion vector (mvr'1, mvr'2, ..., mvr'k), respectively, associated with k overlapped reference sections.

2. The way the predictions of claim 1, wherein the step of determining the motion vector of the current section contains the steps are:
calculate the total number of pixels between the reference section and k respectively overlapping the supporting sections,
compare the calculated total number of pixels between the overlapped k reference sections on the basis of a given criterion of comparison.

3. Way prediction according to claim 2, in which the criterion of comparison is based on the choice of k reference motion vectors respectively associated with k overlapping the supporting sections, the reference motion vector associated with the reference section, which calculated the total number of pixels is the greatest.

4. Way prediction according to claim 2, in which the criterion of comparison is based on weighting, using the calculated total number of pixels, the average value of k reference motion vectors respectively associated with k overlapped reference sections.

5. Way prediction according to claim 2, in which the criterion of comparison is based on the ora of the k reference motion vectors, respectively associated with k overlapping the supporting sections, the reference motion vector associated with the overlapped reference section with more pixels within the reference section, than outside the support section.

6. The way the predictions of claim 1, wherein determining the motion vector of the current section contains the steps are:
compute for each k, the overlap of the reference partition coefficient, which is a function of the spatial gradient of the specified reference section,
choose the coefficient having the largest calculated value,
choose the reference motion vector corresponding to the overlapped reference section with the selected ratio.

7. The way the predictions of claim 1, wherein the step of determining the motion vector of the current section contains the steps are:
compute for each k, the overlap of the reference partition coefficient, which is a function of the spatial gradient of the specified reference section,
calculates the mean value of the k reference motion vectors respectively associated with k overlapped reference sections, with the specified mean value is weighed by means of these k calculated coefficients.

8. The way the predictions of claim 1, wherein the step of determining the motion vector of the current section contains the steps are:
choose individual p is th point of the reference section,
choose the reference motion vector associated with the overlapped reference section containing the selected single point.

9. The way the predictions of claim 1, wherein the step of determining the motion vector of the current section contains the steps are:
identify in the reference section of the characteristic associated with the image content,
choose the reference motion vector associated with the overlapped reference section that contains the specified feature.

10. A method of coding an image or a sequence of images, characterized by the fact that generate a flow (F) data containing data characterizing at least one partition image, the method includes a step of predicting the motion vector of the specified partition image,
moreover, the specified prediction is carried out according to the method of prediction according to any one of claims 1 to 9.

11. The way to decode the stream (F) data describing the image or sequence of images, and the specified stream contains data characterizing the at least one partition image, while the method includes a step of predicting the motion vector of the specified partition image,
moreover, the specified prediction is carried out according to the method of prediction according to any one of claims 1 to 9.

12. Device (PREDCO) prediction vector movement is possible in the current section of the image relative to the motion vector of the reference section, having the same form as specified in the partition, and belonging to a reference image different from the current image and the previously broken by the encoding and subsequent decoding on the set n of partitions (r'1, r'2, ..., R n),
in the case when the reference section covers a set of k reference sections from the specified set of n sections (r'1, r'2, ..., R n) of the reference image, where k≤n, the prediction device includes a computing module (CAL), is configured to determine the specified motion vector of the current partition image based on the function of at least one reference motion vectors belonging to the aggregate of the k reference motion vector (mvr'1, mvr'2, ..., mvr'k), respectively, associated with k overlapped reference sections.

13. Device (CO) encoding an image or sequence of images, characterized in that arranged to generate a flow (F) data containing data characterizing at least one partition image, when the specified device contains a means of predicting the motion vector of the specified partition image,
moreover, the use of predictions contained in the device (PREDCO) predictions corresponding to the device according to item 12.

14. The device (DO) decoding stream is (F) data characterized in that the specified data flow describes the image or sequence of images, and the specified stream (F) data contains data characterizing the at least one partition image, when the specified device contains a means of predicting the motion vector of the specified partition image,
moreover, the use of predictions contained in the device (PREDDO) predictions corresponding to the device according to item 12.

15. A computer program containing a command to execute one of the methods according to any one of claims 1 to 11 when executed by the computer.