CN104394423A - Calculation method of macroblock motion vector in spatial resolution video transcoding - Google Patents

Abstract

The invention relates to a calculation method of a macroblock motion vector in spatial resolution video transcoding. The calculation method makes full use of rich information in a source code stream to quickly calculate the macroblock motion vector and ensures the correctness of a calculated result to the greatest extent; the calculation of the motion vector is the most time-consuming part of the transcoding; the method can quickly accomplish the part, so that the transcoding speed is higher. The method employs the video transcoding of a pixel domain, no drift error appears, and no drift effect is caused, so that the quality of a recoded image is ensured.

Description

The computational methods of macroblock motion vector in a kind of Spatial Resolution Video transcoding

Technical field

The present invention relates to the computational methods of macroblock motion vector in a kind of Spatial Resolution Video transcoding, belong to the technical field of multimedia signal dispose.

Background technology

In the application of video request program, in order to provide abundant video resource can to user widely, video content provider needs to preserve the video resource under various form, as high-definition quality, and SD quality, general quality etc.Such demand that could meet all kinds of different user, as PC user, cellphone subscriber etc.In order to the difference that different user terminals brings can be solved, Video Transcoding Technology can be used.In the art, can in real time to the image size in video flowing, frame per second, each parameter such as picture quality adjusts, thus meets the requirement of access network and playback terminal.Such as in the video request program described in Fig. 1, just can add video code conversion module on video server, user just can complete program request by wireless terminal, thus solve that wireless channel is narrow cannot the problem of program request.

The input of video code conversion is a kind of bitstream format (as image size, frame per second, code check, coding standard etc.), through transcoding module, can obtain another output bit flow form.According to input and output bitstream format, video code conversion is divided into transcoding two kinds in standard room transcoding and standard usually.Standard room transcoding refers to the standard that incoming bit stream and output bit flow belong to different.In standard, transcoding refers to that incoming bit stream and output bit flow belong to same standard, is often divided into picture size transcoding again, frame per second transcoding, bit-rate transcoding three aspects.

Summary of the invention

For the deficiencies in the prior art, the invention provides the computational methods of macroblock motion vector in a kind of Spatial Resolution Video transcoding.The method is applicable to wireless video monitoring, and internet video is monitored, and the fields such as video request program, this method not only can significantly accelerate transcoding speed, and ensure that the picture quality after recompile, meets the requirement of real-time.

Technical scheme of the present invention is as follows:

Spatial resolution transcoding i.e. picture size transcoding, namely from large-size image transcoding to reduced size image.Picture size zoom factor is generally divided into integer and arbitrary value two kinds.The present invention supports the situation of arbitrary value zoom factor, and before arranging picture size convergent-divergent, picture size is N1 × M1, and after picture size convergent-divergent, picture size is N2 × M2, and therefore horizontal and vertical image scale factor is respectively N1/N2 and M1/M2.

The corresponding relation of sub-block in image before and after picture size convergent-divergent is given in accompanying drawing 2.Fig. 2 b is a sub-block after picture size convergent-divergent, width and be highly B pixel, and this explanation is referred to as B × B sub-block.Fig. 2 a is the position that B × B sub-block corresponds to before picture size convergent-divergent in image, wherein shadow region, namely being designated region A2 in figure is the corresponding region of B × B sub-block before picture size convergent-divergent in image, if the top left co-ordinate of B × B sub-block after picture size convergent-divergent in image is (x, y), then the top left co-ordinate of region A2 before picture size convergent-divergent in image is (xN1/N2, yM1/M2).Region A1 is defined as: its horizontal and vertical size is the integral multiple of B pixel, and just complete inclusion region A2, black thick line institute enclosing region is A1 in fig. 2 a

Select the sub-block of 16 × 16 pixels in image as a macro block, be called current macro.Each current macro is divided into the sub-block of 16 4 × 4, and division methods is shown in accompanying drawing 3.The macro block (mb) type that the present invention relates to is P16 × 16, P16 × 8, P8 × 16, P8 × 8, divides signal and sees accompanying drawing 4.

Computational methods for macroblock motion vector in Spatial Resolution Video transcoding, comprise step as follows:

1) to each 4 × 4 sub-block estimated motion vectors: method is zoning A1, the wherein median of all motion vectors during B=4; In the A1 of region, each 4 × 4 sub-blocks have a motion vector, extract obtain when decoding from video flowing; The motion vector calculated amounts to 16, is mv _i, i=0,1 ..., 15, result of calculation is table 1:

The result of calculation of the motion vector of each 4 × 4 sub-block in table 1 macro block

mv 0	mv 1	mv 2	mv 3
				mv 4	mv 5	mv 6	mv 7
mv 8	mv 9	mv 10	mv 11
				mv 12	mv 13	mv 14	mv 15

2) by step 1) in 16 motion vectors be divided into 9 different set, division methods is as follows:

MV ₁＝{mv ₀,mv ₁,mv ₂,mv ₃,mv ₄,mv ₅,mv ₆,mv ₇,mv ₈,mv ₉,mv ₁₀,mv ₁₁,mv ₁₂,mv ₁₃,mv ₁₄,mv ₁₅}

MV ₂＝{mv ₀,mv ₁,mv ₂,mv ₃,mv ₄,mv ₅,mv ₆,mv ₇}

MV ₃＝{mv ₈,mv ₉,mv ₁₀,mv ₁₁,mv ₁₂,mv ₁₃,mv ₁₄,mv ₁₅}

MV ₄＝{mv ₀,mv ₁,mv ₄,mv ₅,mv ₈,mv ₉,mv ₁₂,mv ₁₃}

MV ₅＝{mv ₂,mv ₃,mv ₆,mv ₇,mv ₁₀,mv ₁₁,mv ₁₄,mv ₁₅}

MV ₆＝{mv ₀,mv ₁,mv ₄,mv ₅}

MV ₇＝{mv ₂,mv ₃,mv ₆,mv ₇}

MV ₈＝{mv ₈,mv ₉,mv ₁₂,mv ₁₃}

MV ₉＝{mv ₁₀,mv ₁₁,mv ₁₄,mv ₁₅}

3) for current macro selects candidate type, system of selection adopts existing disclosed algorithm;

4) if current macro type selecting is P16 × 16, wherein P16 × 16 type comprises initial motion vectors mv16 × 16 and refinement step-length re16 × 16, and its computational methods are:

$\mathrm{re}16\×16=\frac{1}{16}\underset{i=0}{\overset{15}{\mathrm{\Σ}}}|\mathrm{mv}16x16-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_1$

Wherein median{} represents the median calculating all numerical value in bracket;

5) if current macro type selecting is P16 × 8, wherein P16 × 8 type comprises upper and lower two initial motion vectors, is respectively: mv16 × 8 ₀with mv16 × 8 ₁, each motion vector is a corresponding refinement step-length respectively, is respectively: re16 × 8 ₀, re16 × 8 ₁, its computational methods are:

$\begin{array}{c}\mathrm{mv}16\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+2}\}\\ \mathrm{re}16\×8_k=\frac{1}{8}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|\mathrm{mv}16\×8_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+2}\end{array},k=\mathrm{0,1};$

6) if current macro type selecting is P8 × 16, wherein P8 × 16 type comprises left and right two initial motion vectors, is respectively: mv8 × 16 ₀with mv8 × 16 ₁, each motion vector is a corresponding refinement step-length respectively, is respectively: re8 × 16 ₀, re8 × 16 ₁, its computational methods are:

$\begin{array}{c}\mathrm{mv}8\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+4}\}\\ \mathrm{re}8\×{16}_k=\frac{1}{8}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|\mathrm{mv}8\×{16}_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+4}\end{array},k=\mathrm{0,1};$

7) if current macro type selecting is P8 × 8, wherein P8 × 8 type comprises four, upper and lower, left and right initial motion vectors, is respectively: mv8 × 8 ₀, mv8 × 8 ₁, mv8 × 8 ₂, mv8 × 8 ₃, each motion vector is a corresponding refinement step-length respectively, is respectively: re8 × 8 ₀, re8 × 8 ₁, re8 × 8 ₂, re8 × 8 ₃, its computational methods are:

$\begin{array}{c}\mathrm{mv}8\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+6}\}\\ \mathrm{re}8\×8_k=\frac{1}{4}\underset{i=0}{\overset{3}{\mathrm{\Σ}}}|\mathrm{mv}8\×8_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+6}\end{array},k=\mathrm{0,1,2,3};$

8) initial motion vectors adopting above-mentioned steps to calculate, adopts existing disclosed motion estimation algorithm to carry out motion refinement to it.

The invention has the beneficial effects as follows:

1) the present invention takes full advantage of the abundant information of source code flow, carries out the motion vector computation of macro block fast, and ensures the correctness of selection as far as possible; The calculating of motion vector is part the most consuming time in transcoding, and the present invention can complete this partial content fast, thus transcoding speed.

2) present invention employs the video code conversion based on pixel domain, there will not be drift error, also would not cause drift effect, thus ensure that the picture quality after recompile.

3) image scale factor that the present invention is suitable for is not limited to integer value, but arbitrary value, thus extend the range of application of video code conversion.

4) the present invention is in calculating kinematical vector, have employed median filter, thus precision is higher; Initial motion vectors is through and calculates, and only in motion refinement, has used estimation, therefore speed.

Accompanying drawing explanation

The application principle figure of Fig. 1 transcoding in video-on-demand service;

Sub-block corresponding relation, i.e. image before transcoding in image before and after Fig. 2 a transcoding;

Sub-block corresponding relation, i.e. sub-block after transcoding in image before and after Fig. 2 b transcoding;

The division of Fig. 3 macro block;

Each macro block (mb) type figure in Fig. 4 the present invention.

Embodiment

For the ease of understanding and implementing the present invention, below in conjunction with wireless video on-demand example, the present invention is described in further detail, but is not limited thereto.

As shown in figure 1 – Fig. 4.

Embodiment 1,

In wireless video on-demand, encoded video flowing leaves on video server, and these video flowings are all compress under the prerequisite of high bit rate, and namely picture size is large, and frame per second is high, and picture quality is better.When there being user to carry out certain video-frequency band of program request, corresponding required parameter can be sent to video server simultaneously, these parameters comprise: picture size, frame per second, bit rate etc.Video server, according to the requirement of these parameters, starts transcoding module, by the video stream transcoding that encoded under required form, and in real time by the video stream after transcoding to user terminal.

Video server, according to the parameter request of user side, starts online transcoding module, needs to select a type to each macro block in online transcoding, and the concrete implementation step of this module is as follows:

(1) line decode.Start decoder by online live video stream complete decoding, obtain pixel domain data.

(2) information extraction.Macro block (mb) type is extracted, the information such as residual error data from decoded information.

(3) downscaled images size.According to the needs of terminal, downscaled images size; The algorithm of downscaled images size described herein can use disclosed algorithm.

(4) macro block (mb) type is selected.Utilize disclosed algorithm, complete the type selecting of macro block.

(5) calculating kinematical vector.Use algorithm of the present invention for macro block calculating kinematical vector.

(6) recompile.Use the macro block (mb) type chosen, the motion vector recalculated, carries out recompile to video and exports.

Computational methods for macroblock motion vector in Spatial Resolution Video transcoding, comprise step as follows:

1) to each 4 × 4 sub-block estimated motion vectors: method is zoning A1, the wherein median of all motion vectors during B=4; In the A1 of region, each 4 × 4 sub-blocks have a motion vector, extract obtain when decoding from video flowing; The motion vector calculated amounts to 16, is mv _i, i=0,1 ..., 15, result of calculation is table 1:

The result of calculation of the motion vector of each 4 × 4 sub-block in table 1 macro block

mv 0	mv 1	mv 2	mv 3
				mv 4	mv 5	mv 6	mv 7
mv 8	mv 9	mv 10	mv 11
				mv 12	mv 13	mv 14	mv 15

2) by step 1) in 16 motion vectors be divided into 9 different set, division methods is as follows:

MV ₁＝{mv ₀,mv ₁,mv ₂,mv ₃,mv ₄,mv ₅,mv ₆,mv ₇,mv ₈,mv ₉,mv ₁₀,mv ₁₁,mv ₁₂,mv ₁₃,mv ₁₄,mv ₁₅}

MV ₂＝{mv ₀,mv ₁,mv ₂,mv ₃,mv ₄,mv ₅,mv ₆,mv ₇}

MV ₃＝{mv ₈,mv ₉,mv ₁₀,mv ₁₁,mv ₁₂,mv ₁₃,mv ₁₄,mv ₁₅}

MV ₄＝{mv ₀,mv ₁,mv ₄,mv ₅,mv ₈,mv ₉,mv ₁₂,mv ₁₃}

MV ₅＝{mv ₂,mv ₃,mv ₆,mv ₇,mv ₁₀,mv ₁₁,mv ₁₄,mv ₁₅}

MV ₆＝{mv ₀,mv ₁,mv ₄,mv ₅}

MV ₇＝{mv ₂,mv ₃,mv ₆,mv ₇}

MV ₈＝{mv ₈,mv ₉,mv ₁₂,mv ₁₃}

MV ₉＝{mv ₁₀,mv ₁₁,mv ₁₄,mv ₁₅}

3) for current macro selects candidate type, system of selection adopts existing disclosed algorithm;

4) if current macro type selecting is P16 × 16, wherein P16 × 16 type comprises initial motion vectors mv16 × 16 and refinement step-length re16 × 16, and its computational methods are:

$\begin{array}{c}\mathrm{mv}16\×16=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_1\}\\ \mathrm{re}16\×16=\frac{1}{16}\underset{i=0}{\overset{15}{\mathrm{\Σ}}}|\mathrm{mv}16\×16-\mathrm{mv}|,\mathrm{mv}\∈\mathrm{MV}\end{array}$

Wherein median{} represents the median calculating all numerical value in bracket;

5) if current macro type selecting is P16 × 8, wherein P16 × 8 type comprises upper and lower two initial motion vectors, is respectively: mv16 × 8 ₀with mv16 × 8 ₁, each motion vector is a corresponding refinement step-length respectively, is respectively: re16 × 8 ₀, re16 × 8 ₁, its computational methods are:

$\begin{array}{c}\mathrm{mv}16\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+2}\}\\ \mathrm{re}16\×8_k=\frac{1}{8}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|\mathrm{mv}16\×8_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+2}\end{array},k=\mathrm{0,1};$

6) if current macro type selecting is P8 × 16, wherein P8 × 16 type comprises left and right two initial motion vectors, is respectively: mv8 × 16 ₀with mv8 × 16 ₁, each motion vector is a corresponding refinement step-length respectively, is respectively: re8 × 16 ₀, re8 × 16 ₁, its computational methods are:

$\begin{array}{c}\mathrm{mv}8\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+4}\}\\ \mathrm{re}8\×{16}_k=\frac{1}{8}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|\mathrm{mv}8\×{16}_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+4}\end{array},k=\mathrm{0,1};$

7) if current macro type selecting is P8 × 8, wherein P8 × 8 type comprises four, upper and lower, left and right initial motion vectors, is respectively: mv8 × 8 ₀, mv8 × 8 ₁, mv8 × 8 ₂, mv8 × 8 ₃, each motion vector is a corresponding refinement step-length respectively, is respectively: re8 × 8 ₀, re8 × 8 ₁, re8 × 8 ₂, re8 × 8 ₃, its computational methods are:

$\begin{array}{c}\mathrm{mv}8\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+6}\}\\ \mathrm{re}8\×8_k=\frac{1}{4}\underset{i=0}{\overset{3}{\mathrm{\Σ}}}|\mathrm{mv}8\×8_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+6}\end{array},k=\mathrm{0,1,2,3};$

8) initial motion vectors adopting above-mentioned steps to calculate, adopts existing disclosed motion estimation algorithm to carry out motion refinement to it.

Claims (1)

1. the computational methods of macroblock motion vector in Spatial Resolution Video transcoding, comprise step as follows:

1) to each 4 × 4 sub-block estimated motion vectors: method is zoning A1, the wherein median of all motion vectors during B=4; In the A1 of region, each 4 × 4 sub-blocks have a motion vector, extract obtain when decoding from video flowing; The motion vector calculated amounts to 16, is mv _i, i=0,1 ..., 15, result of calculation is table 1:

The result of calculation of the motion vector of each 4 × 4 sub-block in table 1 macro block

mv ₀ mv ₁ mv ₂ mv ₃ mv ₄ mv ₅ mv ₆ mv ₇ mv ₈ mv ₉ mv ₁₀ mv ₁₁ mv ₁₂ mv ₁₃ mv ₁₄ mv ₁₅

2) by step 1) in 16 motion vectors be divided into 9 different set, division methods is as follows:

MV ₁＝{mv ₀,mv ₁,mv ₂,mv ₃,mv ₄,mv ₅,mv ₆,mv ₇,mv ₈,mv ₉,mv ₁₀,mv ₁₁,mv ₁₂,mv ₁₃,mv ₁₄,mv ₁₅}

MV ₂＝{mv ₀,mv ₁,mv ₂,mv ₃,mv ₄,mv ₅,mv ₆,mv ₇}

MV ₃＝{mv ₈,mv ₉,mv ₁₀,mv ₁₁,mv ₁₂,mv ₁₃,mv ₁₄,mv ₁₅}

MV ₄＝{mv ₀,mv ₁,mv ₄,mv ₅,mv ₈,mv ₉,mv ₁₂,mv ₁₃}

MV ₅＝{mv ₂,mv ₃,mv ₆,mv ₇,mv ₁₀,mv ₁₁,mv ₁₄,mv ₁₅}

MV ₆＝{mv ₀,mv ₁,mv ₄,mv ₅}

MV ₇＝{mv ₂,mv ₃,mv ₆,mv ₇}

MV ₈＝{mv ₈,mv ₉,mv ₁₂,mv ₁₃}

MV ₉＝{mv ₁₀,mv ₁₁,mv ₁₄,mv ₁₅}

3) for current macro selects candidate type, system of selection adopts existing disclosed algorithm;

4) if current macro type selecting is P16 × 16, wherein P16 × 16 type comprises initial motion vectors mv16 × 16 and refinement step-length re16 × 16, and its computational methods are:

mv16×16＝median{mv∈MV ₁}

$\mathrm{re}16\×16=\frac{1}{16}\underset{i=0}{\overset{15}{\mathrm{\Σ}}}|\mathrm{mv}16\×16-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_1$

Wherein median{} represents the median calculating all numerical value in bracket;

5) if current macro type selecting is P16 × 8, wherein P16 × 8 type comprises upper and lower two initial motion vectors, is respectively: mv16 × 8 ₀with mv16 × 8 ₁, each motion vector is a corresponding refinement step-length respectively, is respectively: re16 × 8 ₀, re16 × 8 ₁, its computational methods are:

$\left\{\begin{array}{c}\mathrm{mv}16\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+2}\}\\ \mathrm{re}16\×8_k=\frac{1}{8}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|\mathrm{mv}16\×8_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+2}\end{array}\right.,k=\mathrm{0,1};$

6) if current macro type selecting is P8 × 16, wherein P8 × 16 type comprises left and right two initial motion vectors, is respectively: mv8 × 16 ₀with mv8 × 16 ₁, each motion vector is a corresponding refinement step-length respectively, is respectively: re8 × 16 ₀, re8 × 16 ₁, its computational methods are:

$\left\{\begin{array}{c}\mathrm{mv}8\×{16}_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+4}\}\\ \mathrm{re}8\×{16}_k=\frac{1}{8}\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|\mathrm{mv}8\×{16}_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+4}\end{array}\right.,k=\mathrm{0,1};$

7) if current macro type selecting is P8 × 8, wherein P8 × 8 type comprises four, upper and lower, left and right initial motion vectors, is respectively: mv8 × 8 ₀, mv8 × 8 ₁, mv8 × 8 ₂, mv8 × 8 ₃, each motion vector is a corresponding refinement step-length respectively, is respectively: re8 × 8 ₀, re8 × 8 ₁, re8 × 8 ₂, re8 × 8 ₃, its computational methods are:

$\left\{\begin{array}{c}\mathrm{mv}8\×8_k=\mathrm{median}\{\mathrm{mv}\∈{\mathrm{MV}}_{k+6}\}\\ \mathrm{re}8\×8_k=\frac{1}{4}\underset{i=0}{\overset{3}{\mathrm{\Σ}}}|\mathrm{mv}8\⊗8_k-\mathrm{mv}|,\mathrm{mv}\∈{\mathrm{MV}}_{k+6}\end{array}\right.,k=\mathrm{0,1},\mathrm{2,3};$

8) initial motion vectors adopting above-mentioned steps to calculate, adopts existing disclosed motion estimation algorithm to carry out motion refinement to it.