CN1744719A - Block prediction method - Google Patents

Abstract

In the method for predicting an image, a reference image block for a current image block of a moving picture is obtained using motion vector information, and a variable weight factor is applied to the reference image block to form a product value of the weight factor and the reference image block. The current image block of the moving picture is predicted using the product value.

Description

The Forecasting Methodology of piece

The application is that application number is that 03101657.X, denomination of invention are divided an application for " adopting the block predication that improves Direct Model ".

Technical field

The present invention relates to the moving image encoding system, relate in particular to a kind of B image block Forecasting Methodology of improving Direct Model of using.

Background technology

In the moving image encoding system, use an advantage of B image to be to have selected more not increase the direct predictive mode of extraneous information than any other predictive mode (forward prediction, back forecast, bi-directional predicted and infra-frame prediction or the like).Code efficiency height when the code efficiency ratio when therefore, the moving image encoding system uses the B image only uses the P image.

In the B image, use the block predication of Direct Model, calculating forward motion vector and backward motion vector, as the zoom version of Direct Model the back motion vector of same position piece in reference picture; Use above-mentioned motion vector and obtain two different motion compensation blocks; And by average two motion compensation blocks finally obtain the prediction piece.

With reference to figure 1, hereinafter the block predication of aforesaid use Direct Model will be described in more detail.

Fig. 1 has shown an image model, to explain the block predication of employing Direct Model of the prior art.As shown in Figure 1, this image model comprises: use the prediction that is only come by the sampling of the decoding in the same image and the I image (not shown) of encoding, maximum motion vectors that utilization is come by the reference picture of early decoding, P image P1, the P4 and the P7 that encode by inter prediction, and the B image B 2, B3, B5 and the B6 that encode by two intra-frame prediction block that the reference picture by early decoding comes.

In addition, for simplicity, hereinafter parameter shown in Figure 1 will be described at first.TR _DBe illustrated in the time gap between reference picture P7 after Direct Model forward direction reference picture P1 and the Direct Model, TR _BBe illustrated in the time gap between Direct Model forward direction reference picture P1 and the current B image B 5, MV represents after the Direct Model motion vector of same position piece in reference picture P7, MV _fThe Direct Model forward motion vector of Direct Model forward direction reference picture is pointed in expression, and MV _bThe Direct Model backward motion vector to reference picture is pointed to after the Direct Model in expression.Here, Direct Model forward direction reference picture is meant after the Direct Model motion vector reference picture pointed of same position piece in reference picture.

Hereinafter will use above-mentioned parameter to describe the block predication of Direct Model.

At first, by using following formula (1), according to after the Direct Model in reference picture P7 same position piece B _sMotion vector MV, obtain Direct Model forward motion vector MV _f

${\mathrm{MV}}_f=\frac{{\mathrm{TR}}_B\×\mathrm{MV}}{{\mathrm{TR}}_D}\·\·\·\left(1\right)$

In addition, by using following formula (2), according to the same position piece B in reference picture (P7) after the Direct Model _sMotion vector MV, obtain Direct Model backward motion vector MV _b

${\mathrm{MV}}_b=({\mathrm{TR}}_B-{\mathrm{TR}}_D)\frac{\mathrm{MV}}{{\mathrm{TR}}_D}\·\·\·\left(2\right)$

Thereby, use the MV that calculates by formula (1) and (2) _fAnd MV _bTo piece B _fAnd B _bThe benefit of moving then, by average these two pieces of following formula (3), obtains B image current block B _cPredicted value B _c'.

$B_c^{\′}=\frac{B_f+B_b}{2}\·\·\·\left(3\right)$

Yet, according to the Direct Model block predication of prior art, obtain the Direct Model forward motion vector according to the motion vector of the same position piece in reference picture after the Direct Model, therefore, resulting value is an approximation, is not the accurate motion vector of B figure current block.

In addition, Direct Model block predication according to prior art, be very similar to the B image even approach the reference picture of B image on the time, also use the mean value of two different motion compensation blocks to carry out the piece prediction, and do not consider the time gap between the reference picture.Therefore, the accuracy of prediction piece is lower.

Especially, in sequence with the scene of fading out, because the continuously brightness of B image deepening or brighten gradually gradually, exist very big difference between predicted value that obtains by average two motion compensation blocks simply and the original value, thereby the code efficiency of whole system is very low.

Summary of the invention

Therefore, the object of the present invention is to provide a kind of block predication that uses Direct Model, this method by Direct Model after in reference picture the same position piece motion vector and obtain the Direct Model forward motion vector, and two different motion compensation blocks are carried out interpolative prediction and obtain the prediction piece of the current B image that will encode, thereby improve code efficiency.

In addition, another object of the present invention is to provide a kind of block predication that uses Direct Model, obtain the Direct Model forward motion vector by the reference picture that approaches most current B image, and, predict the accuracy of piece and improved code efficiency thereby improved by two different motion compensation blocks are carried out the prediction piece that interpolative prediction obtains the current B image that will encode.

Such as here embodiment and broadly described, to achieve the object of the present invention, the invention provides a kind of bi-directional predicted Forecasting Methodology, comprise: first motion vector of the current block that acquisition will be encoded or decode and second motion vector; Use first motion vector and second motion vector to obtain two different motion compensation blocks; By coefficient being applied to each of two different motion compensation blocks, from bi-directional predicted of two different motion compensation block predictions, wherein, based on distributing to reference picture that belongs to two different motion compensation blocks and the display sequence information that belongs to bi-directional predicted present image, obtain this coefficient.

By following detailed description, in conjunction with the accompanying drawings, it is clearer that above and other objects of the present invention, feature, aspect and advantage can become.

Description of drawings

Accompanying drawing helps to understand better the present invention, and at this in conjunction with a part that constitutes the application, the description of drawings embodiments of the invention and and specification explain principle of the present invention together, in the accompanying drawing:

Fig. 1 has shown an image model, is used to describe the Direct Model block predication of prior art;

Fig. 2 has shown an image model, is used to describe Direct Model block predication of the present invention;

Fig. 3 has shown an image model, is used to describe the interpolation prediction method of the embodiment of the invention; And

Fig. 4 has shown an image model, is used to describe the interpolation prediction method of another embodiment of the present invention.

The preferred embodiment explanation

Hereinafter will describe the preferred embodiments of the present invention in detail, the preferred embodiment is meant embodiment shown in the accompanying drawing.

In Direct Model block predication of the present invention, motion vector by the same position piece in reference picture after the Direct Model calculates Direct Model forward motion vector and backward motion vector, use above-mentioned motion vector to obtain two motion compensation blocks, at last, by being carried out interpolation, two motion compensation blocks obtain predicting piece.

In addition, in the block predication of use Direct Model of the present invention, by calculating backward motion vector to reference picture after the Direct Model, calculate the Direct Model forward motion vector by the reference picture that approaches most current B image in the forward direction reference picture, and obtain motion compensation block by above-mentioned motion vector, at last, by being carried out interpolation, these two motion compensation blocks obtain to predict piece.

With reference to the accompanying drawings, hereinafter embodiment of the present invention will be described.

Fig. 2 has shown an image model, so that Direct Model block predication of the present invention to be described.As shown in Figure 2, this image model comprises: use the prediction that is only come by the sampling of the decoding in the same image and the I image (not shown) of encoding, maximum motion vectors that utilization is come by the reference picture of early decoding, P image P1, the P4 and the P7 that encode by inter prediction, and the B image B 2, B3, B5 and the B6 that encode by two inter prediction pieces that the reference picture by early decoding comes.

For simplicity, hereinafter parameter shown in Figure 2 will be described at first.TR _DTime gap after expression Direct Model forward direction reference picture P1 and the Direct Model between reference picture P7, TR _BThe forward direction reference picture P1 of expression Direct Model and the time gap between the current B image B 5, TR _NExpression approaches the reference picture P4 of current B image and the time gap between the current B image B 5 most, and MV represents the motion vector of the same position piece in reference picture P7 after the Direct Model, MV _fThe Direct Model forward motion vector of the reference picture P4 that approaches current B image is most pointed in ' expression, and MV _BThe Direct Model backward motion vector to reference picture P7 is pointed to after the Direct Model in expression.

At this moment, current B image is being encoded (or decoding) before, in reference picture is encoded the process of (or decoding), obtaining the same position piece B in reference picture P7 after the Direct Model after to Direct Model _sMotion vector MV.

The Direct Model block predication of the present invention that as above makes up hereinafter will be described.

Can obtain pointing to the forward motion vector MV of reference picture P4 according to following formula (4) _f', this reference picture has the shortest time gap in the forward direction reference picture.

${\mathrm{MV}}_f^{\′}=\frac{{\mathrm{TR}}_N\×\mathrm{MV}}{{\mathrm{TR}}_D}\·\·\·\left(4\right)$

In addition, can obtain to point to after the Direct Model backward motion vector MV according to the following formula (2) that uses prior art to reference picture P7 _b

${\mathrm{MV}}_b=({\mathrm{TR}}_B-{\mathrm{TR}}_D)\×\frac{\mathrm{MV}}{{\mathrm{TR}}_D}\·\·\·\left(2\right)$

Therefore, use the motion vector MV that calculates by formula (2) and formula (4) _f' and MV _bCan obtain motion compensation block B _fAnd B _b

On the other hand, according to above-mentioned two motion compensation block B _fAnd B _b, can obtain piece B _cPredicted value B _c'.At this moment, motion compensation block B can more be approached to contain in the position of B image _fReference picture and contain motion compensation block B _bDirect Model after in reference picture one.

The block predication of use Direct Model of the present invention can be applicable to Fig. 1 and Fig. 2, therefore, contains motion compensation block B _fReference picture be Direct Model forward direction reference picture (for example, the P1 image among Fig. 1) or from the nearest reference picture of B image (for example, the P4 image among Fig. 2).

In addition, in the sequence with the scene of fading out, therefore the continuously brightness of B image deepening or brighten gradually gradually, resembles the prior art by two motion compensation block B of simple average _fAnd B _bAnd between predicted value that obtains and the actual original value bigger difference is arranged.Therefore, its code efficiency significantly reduces.

Therefore, the present invention uses the block predication of Direct Model to carry out interpolative prediction, considers current B image and contains motion compensation block B _fReference picture (promptly, Direct Model forward direction reference picture or from the nearest reference picture of B image) between time gap, and consider the time gap between reference picture after current B image and the Direct Model, thereby improve the accuracy of the prediction piece of Direct Model.

As shown in Figure 3, obtain the Direct Model forward motion vector, Direct Model motion compensation block B if use prior art _fBe present among the forward direction reference picture P1 Direct Model motion compensation block B _bBe present in the back in reference picture P7, then carry out the interpolative prediction shown in the following formula (5).Here, TR _DTime gap after finger Direct Model forward direction reference picture P1 and the Direct Model between reference picture P7, TR _BRefer to the time gap between Direct Model forward direction reference picture P1 and the current B image B 5.Especially, when the B image after Direct Model forward direction reference picture and the Direct Model during mid point between reference picture, this interpolation prediction method comprises the average computation the same with prior art.

$B_c^{\′}=B_f\×\frac{({\mathrm{TR}}_D-{\mathrm{TR}}_B)}{{\mathrm{TR}}_D}+B_b\×\frac{{\mathrm{TR}}_B}{{\mathrm{TR}}_D}\·\·\·\left(5\right)$

In addition, as shown in Figure 4, when obtaining the Direct Model forward motion vector according to the present invention, motion compensation block B _fBe present in the nearest reference picture P4 of current B image motion compensation block B _bBe present in after the Direct Model in reference picture P7.Therefore, carry out following formula (6) and carry out interpolative prediction.Here, TR _DTime gap after finger Direct Model forward direction reference picture P1 and the Direct Model between reference picture P7, TR _BRefer to the time gap between Direct Model forward direction reference picture P1 and the current B image, TR _NFinger is from nearest reference picture P4 of current B image and the time gap between the current B image.

$B_c^{\′}=B_f\×\frac{({\mathrm{TR}}_D-{\mathrm{TR}}_B)}{({\mathrm{TR}}_N+{\mathrm{TR}}_D-{\mathrm{TR}}_B)}+B_b\×\frac{{\mathrm{TR}}_N}{({\mathrm{TR}}_N+{\mathrm{TR}}_D-{\mathrm{TR}}_B)}\·\·\·\left(6\right)$

On the other hand, use the picture order count value, promptly display sequence information is represented each image.

Therefore, utilize the picture order count value of each image, i.e. display sequence information, formula (5) and the available following formula of formula (6) (7) expression.Here, T _cBe the picture order count value, promptly be assigned to the display sequence information of current B image, T _fIt is the picture order count value, promptly be assigned to the display sequence information of Direct Model forward direction reference picture, or when obtaining forward motion vector, be a picture order count value by formula (4), promptly distribute to display sequence information from the nearest reference picture of B image, and T _bBe a picture order count value, promptly distribute to after the Direct Model display sequence information to reference picture.

$B_c^{\′}=B_f\×\frac{(T_b-T_c)}{(T_b-T_f)}+B_b\×\frac{(T_c-T_f)}{(T_b-T_f)}\·\·\·\left(7\right)$

As mentioned above, according to the present invention, motion vector according to the same position piece in reference picture after the Direct Model obtains the Direct Model forward motion vector, and obtains the prediction piece of the B image that will encode by the motion compensation block value being carried out interpolative prediction.Therefore, compared with prior art, code efficiency of the present invention is improved.

In addition, according to the present invention, according to can obtaining the Direct Model forward motion vector from the nearest reference picture of the B image of be about to be encoded (or decoding) with the reference picture of B image similarity, and by the above-mentioned forward motion vector that comes from Direct Model and the motion compensation block of backward motion vector are carried out the prediction piece that interpolative prediction obtains the B image.Therefore, the accuracy of prediction piece can be improved, and code efficiency also can be improved.

Under the situation that does not break away from spirit of the present invention or substantive characteristics, the present invention may be embodied as various ways, should also be appreciated that, unless specify in addition, above embodiment is not limited to above-mentioned any details, and should explain broadly in the spirit and scope that claim limited that therefore, appended claim contains the changes and improvements in all boundaries that drop on claim or its equivalent.

Claims (6)

1. method of predicting the piece in the present image comprises:

For the piece in the present image, obtain the value of the motion compensation block in the reference picture; And

By coefficient being applied to the value of motion compensation block, according to the value of motion compensation block, the piece in the prediction present image.

2. according to the process of claim 1 wherein, prediction steps is predicted the piece in the present image according to the distance between present image and the reference picture.

3. according to the process of claim 1 wherein, this coefficient depends on distance.

4. method of predicting the piece in the present image comprises:

Use forward motion vector to obtain forward direction reference block in the forward direction reference picture, and use backward motion vector to obtain back back in reference picture to reference block; And

By coefficient being applied to forward direction reference block and back to reference block, the piece in the prediction present image.

5. according to the method for claim 4, wherein, prediction steps is according to the distance between each and the present image in reference picture of forward direction reference picture and back, the piece in the prediction present image.

6. according to the method for claim 4, wherein, coefficient depends on distance.

Images