CN101668197A - Code rate control method in scalable video coding based on linear model - Google Patents

Abstract

The invention discloses a code rate control method in scalable video coding based on a linear model, comprising the following steps: allocating a code rate to the groups of pictures in each layering of the scalable video coding; allocating an initial code rate to each frame of picture in the groups of pictures; adjusting the initial code rate allocated to each frame of picture; obtaining the quantization parameter required by the coding of each frame of picture in the method of linear model; and carrying out coding on each frame of picture by the obtained quantization parameter, wherein, the method for adjusting the initial code rate allocated to each frame of picture is as follows: (a) finding out the encoded picture which has identical position with the picture needing to be adjusted inthe encoded group of picture adjacent to the picture needing to be adjusted, and obtaining the relative code rate error of the encoded picture; and (b) obtaining the code rate finally allocated to thepicture needing to be adjusted according to the relative code rate error E. The invention overcomes the defects of the prior art; the matching of code rates is accurate; and the obtained code streamhas dramatic gain in the aspect of PSNR, thus improving the effect of code rate control.

Description

A kind of based on the bit rate control method in the scalable video coding of linear model

Technical field

The present invention relates to a kind of based on the bit rate control method in the scalable video coding of linear model.

Background technology

Be accompanied by the growing of the Internet and radio network technique, the application of digital video technology is seen everywhere.Along with popularizing and the granting of domestic 3G license of 3G (Third Generation) Moblie technology (3G), video coding technique becomes more and more important efficiently.At smart mobile phone, face different screen sizes today that hand-held intelligent equipment such as palmtop PC are widely applied, the disposal ability and the network bandwidth require the video transmission service that the code stream that can be applicable to the different application terminal can be provided.H.264/AVC the video compression coding standard of main flow can not meet the demands now.In order to satisfy application demand, the video coding expert group (VCEG) of joint video team (JVT) and ITU-T unites and has proposed scalable video coding (Scalable Video Coding).It is a kind of multi-layer coding technology, provides to comprise the space, and time and quality are at three kinds of interior scalabilities.Spatial scalability (Spatial Scalability) is to provide the code stream that contains different images resolution in order to satisfy the different screen size of Video Applications terminal, time scalability (Temporal Scalability) thus provide the code stream that contains different frame per second to obtain the smooth degree of different video, quality scalability (Quality Scalability) provide the code stream with different images fine degree, thereby obtain different picture quality.By the combination of three kinds of scalabilities, just can obtain to satisfy the scalability requirement of multiple different situations, preferably the new now Video service requirement of Xie Jueing.Therefore, scalable video coding has obtained increasing concern.

In real-time video transmission, Rate Control is most important.It makes the bit rate behind the coding can obtain the code stream that restructural goes out better picture quality under the situation that adapts to the network bandwidth of being given.At present still seldom for the research of the Rate Control in the scalable video coding.For based on the bit rate control method in the scalable video coding of linear model, (bit rate control method of ρ-domain), wherein ρ is the shared ratio of zero coefficient (zero coefficients) behind the image quantization to be also referred to as the ρ territory.The size of ρ is directly related with the code stream size behind the entropy coding.The ρ domain model utilizes the relation of ρ-rate to obtain target ρ from the code check of being given, and utilizes the relation of ρ-QP to select required QP again, thereby reaches the purpose of Rate Control.At present this method is still needing to improve aspect the PSNR gain of the accurate coupling of code check and gained code stream.

Summary of the invention

Technical problem to be solved by this invention provides a kind of bit rate control method based on the scalable video coding of linear model with better Rate Control effect.

Inventive concept of the present invention is: there is deviation in the encode code check that drawn of every two field picture with the code check of original allocation, for the code check that makes final gained mates more accurately with target bit rate, the initial bit rate that needs in real time each image to be distributed is in advance adjusted.The factor that initial bit rate that coded image distributes adjusts is treated in influence three: one, the relative code check error that encoded image produced identical with the picture position of need adjustment in the encoded image group adjacent with the image of need adjustment, the 2nd, the summation of the image place coding relative code check error that layering produced that need adjust, the 3rd, need the image place coding relative code check error change trend of adjusting that layering produced, feedback according to above three aspects is adjusted the code check of the image initial distribution of need adjustment, make this bit rate control method more accurately mate, improved the effect of Rate Control at bit rate connection.

Introduced gradable B frame (Hierarchical B) structure in the scalable video coding, as shown in Figure 5, it is a pyramid structure, the bottom is high frequency B frame, from low to high, frequency reduces successively, and its contribution to whole coding efficiency is also increasing, and pyramidal top is the I frame of intraframe coding or the P frame of interframe forward predictive coded.Therefore, we improve the distribution method of code check, each picture frame that makes full use of in scalable video coding employing multi-segment and the layering is different characteristics to coding efficiency contribution, having adopted based on coded image self complexity and the weight in image sets thereof is the Rate Distribution Strategy that the factor combines, make and utilize method gained code stream of the present invention aspect PSNR, obvious gain to be arranged, improved the effect of Rate Control.

The present invention solves the technical solution that its technical problem takes: should may further comprise the steps based on the bit rate control method in the scalable video coding of linear model: (1) distributes code check to the image sets in each layering of scalable video coding; (2) the every two field picture in the described image sets of step (1) is distributed initial bit rate; (3) initial bit rate that every two field picture is distributed is adjusted; (4) draw the required quantization parameter of coding of described every two field picture with the method for linear model; (5) with the quantization parameter of gained described every two field picture is encoded, wherein, the method that the initial bit rate that in the step (3) every two field picture is distributed is adjusted is:

(a) at first in the encoded image group adjacent, find and the identical encoded image in picture position that needs to adjust, obtain the relative code check error E of this encoded image by following formula with the image of need adjustment

$E=\frac{B_{t\arg \mathrm{et}}-B_{\mathrm{actual}}}{B_{t\arg \mathrm{et}}},$

In the formula, B _TargetBe the code check of this encoded image being intended distribute, B _ActualCode check for this encoded image real income;

(b), draw the code check T that the final image that need are adjusted distributes by following formula according to relative code check error E _{L, k}For

$T_{l,k}=T_{l,k,\mathrm{init}}(1+k_1{E}_{l,k}+k_2\underset{j=0}{\overset{k}{\mathrm{\Σ}}}{E}_{l,j}\frac{1}{F_l}+k_3(E_{l,k}-E_{l,k-1})F_l),$

In the formula, T _{L, k, init}Be the initial bit rate that image distributed to the need adjustment, E _{L, j}(j=0,1,2,3 ..., k) the relative code check error that produces for the image place layering encoded image that need adjust, l is the coding layering that needs the image place of adjustment, j (j=0,1,2,3 ..., k) be encoded image and the place, position that needs the image of adjustment, k _lFor to current relative code check error E _{L, k}The coefficient that feeds back, k ₂The summation of the relative code check error that produces for the image place layering that need are adjusted

The coefficient that feeds back, k ₃For to relative code check error change trend (E _{L, k}-E _{L, k-1}) coefficient that feeds back, F _lImage place layered encoded frame per second for the need adjustment.

Further, distribute the method for initial bit rate to be to every two field picture in the step of the present invention (2):

If this image is the I/P frame, then through type (I) obtains distributing to the code check T of this image _l ^I/PFor

$T_l^{I/P}=\frac{{\mathrm{\ω}}^{I/P}{G}_l}{{\mathrm{\ω}}^{I/P}+\underset{m=0}{\overset{d-1}{\mathrm{\Σ}}}{\mathrm{\ω}}_m{N}_b\left(m\right)}$

Formula (I),

In the formula (I), T _l ^I/PFor distributing to the code check of I/P frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, w ^I/PBe the weight coefficient of I/P frame in the image sets to be encoded, G _lBe the code check that the present image group is distributed, d represents total progression of time domain classification, the time domain progression at B frame place in the m presentation video group, N _b(m) be the quantity of uncoded B frame in the m level time domain classification, w _mIt is the weight coefficient of B frame in the m level time domain classification;

If this image is the B frame, then through type (II) obtains distributing to the code check T of this image ^b _{L, ji}For

$T_{l,j}^B=\frac{{\mathrm{\ω}}_j{G}_j}{\underset{m=j}{\overset{d-1}{\mathrm{\Σ}}}{\mathrm{\ω}}_m{N}_B\left(m\right)}$

Formula (II),

In the formula (II), t ^B _{L, j}For distributing to the code check of B frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, and j is the time domain classification at the B frame place of code check to be allocated, and d represents the progression in the time domain classification, N _B(m) be the quantity of uncoded B frame in the m level time domain classification, G _jBe the last code check of present image group, w _mIt is the weight coefficient of B frame in the m level time domain classification.

The present invention is based on the bit rate control method in the scalable video coding of linear model, can also be may further comprise the steps: (1) distributes code check to the image sets in each layering of scalable video coding; (2) the every two field picture in the described image sets of step (1) is distributed initial bit rate; (3) initial bit rate that every two field picture is distributed is adjusted; (4) draw the required quantization parameter of coding of described every two field picture with the method for linear model; (5) with the quantization parameter of gained described every two field picture is encoded, it is characterized in that distributing the method for initial bit rate to be to every two field picture in the described step (2):

If this image is the I/P frame, then through type (I) obtains distributing to the code check T of this image _l ^I/PFor

$T_l^{I/P}=\frac{{\mathrm{\ω}}^{I/P}{G}_l}{{\mathrm{\ω}}^{I/P}+\underset{m=0}{\overset{d-1}{\mathrm{\Σ}}}{\mathrm{\ω}}_m{N}_b\left(m\right)}$

Formula (I),

In the formula (I), T _l ^I/PFor distributing to the code check of I/P frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, w ^I/PBe the weight coefficient of I/P frame in the image sets to be encoded, G _lBe the code check that the present image group is distributed, d represents total progression of time domain classification, the time domain progression at B frame place in the m presentation video group, N _b(m) be the quantity of uncoded B frame in the m level time domain classification, w _mIt is the weight coefficient of B frame in the m level time domain classification;

If this image is the B frame, then through type (II) obtains distributing to the code check T of this image ^B _{L, ji}For

$T_{l,j}^B=\frac{{\mathrm{\ω}}_j{G}_j}{\underset{m=j}{\overset{d-1}{\mathrm{\Σ}}}{\mathrm{\ω}}_m{N}_B\left(m\right)}$

Formula (II),

In the formula (II), T ^B _{L, j}For distributing to the code check of B frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, and j is the time domain classification at the B frame place of code check to be allocated, and d represents the progression in the time domain classification, N _B(m) be the quantity of uncoded B frame in the m level time domain classification, G _jBe the last code check of present image group, w _mIt is the weight coefficient of B frame in the m level time domain classification.

Compared with prior art, bit rate control method of the present invention has the following advantages: (1) is adjusted the code check that image initial distributes according to the feedback of three aspects: the one, and the code check deviation that encoded image produced identical in the encoded image group adjacent with the picture position of need adjustment with the image of need adjustment; The 2nd, need the image place coding code check deviation summation of adjusting that layering produced; The 3rd, the image place coding layering that needs to adjust produces the variation tendency of code check deviation, thereby makes this bit rate control method more accurately mate at bit rate connection, has improved the effect of Rate Control; (2) primary distribution method to every two field picture code check improves, make full use of scalable video coding and adopted multi-segment, and contribution is different characteristic to each image in the layering to coding efficiency, having adopted based on coded image self complexity and the weight in image sets thereof is the Rate Distribution Strategy that the factor combines, make and utilize method gained code stream of the present invention aspect PSNR, obvious gain to be arranged, improved the effect of Rate Control.

Description of drawings

Fig. 1 is the FB(flow block) of first kind of execution mode of bit rate control method of the present invention;

Fig. 2 is the FB(flow block) of second kind of execution mode of bit rate control method of the present invention;

Fig. 3 is the FB(flow block) of the third execution mode of bit rate control method of the present invention;

Fig. 4 is the graph of a relation of code check and ρ in the high frequency B frame of scalable video coding of the present invention;

Fig. 5 is the exploded view of 4 grades of layerings in the scalable video coding;

Fig. 6 is the corresponding relation figure of the present invention's ρ-code check and ρ-quantization parameter in the process that draws the required quantization parameter of coded image;

Fig. 7 for the present invention on quality scalability with the performance comparison diagram of existing scalable video coding method based on linear model.

Embodiment

The invention will be further described below in conjunction with the drawings and specific embodiments, and the present invention has carried out emulation experiment to it on JSVM 12." soccer " cycle tests of selecting motion less " pairs " and moving violent for sequence " pairs ", is selected k as simulation object of the present invention ₁, k ₂, k ₃So that 1.0,0.15,0.2}; For sequence " soccer ", the present invention selects k ₁, k ₂, k ₃So that 1.0,0.25,0.3}.Then respectively to the spatial domain in the scalability, mass domain, and the associating scalability carried out emulation, and simulation result is being contrasted aspect the PSNR gain of the accuracy of code check and gained code stream with existing scalable video coding method based on linear model.

Embodiment 1:

As shown in Figure 1, present embodiment handle to as if having three sheaf space scalabilities { qcif, cif, the sequence of 4cif} " pairs ".Concrete configuration is: GOPSize=8, frame per second=30frames/s, each space delamination of cycle tests " pairs " qcif, cif, the pairing target bit rate of 4cif} be 192,576,1728}kbits/s.

The first frame coded image with the basic layer (qcif) of sequence " pairs " among this embodiment is the concrete implementation step that example illustrates this example:

(1) image sets in each layering of scalable video coding is distributed code check: according to every layer target bit rate, the coding frame per second, and the size of GOP distributes code check G for each GOP of every layer _l:

$G_l=S_l\×\frac{G_l}{F_l}$

Wherein, G _lFor need distribute the target bit rate of the image sets place layering of code check, l is for distributing the coding layering at the image sets place of code check, F _lBe the image sets place layered encoded frame per second that need distribute code check, S _lThe size of distributing the image sets of code check for need.

The target bit rate G of first space delamination _lBe 192kbits/s, Sl=8, frame per second F _lBe 30frames/s, owing to be first image sets in this layering, according to the processing method among the JSVM, the value of gop size need increase by a frame, and therefore, the size of this image sets is 9.First coded frame is the I frame, and to the contribution maximum of coding, so we distribute this image sets

${2G}_l=2S_l\×\frac{G_l}{F_l}=2\×8\×\frac{192}{30}=102.4\left(\mathrm{kbits}\right)$

Code check.

(2) the every two field picture in the described image sets of step (1) is distributed initial bit rate: adopt according to image self complexity to come it is carried out the initial bit rate distribution, the formula of image self complexity p is:

$p=2^{q/6}\×\underset{i,j}{\mathrm{\Σ}}{x}_{\mathrm{ij}}$

In the formula, the quantization parameter when q is this two field picture of coding, x _IjFor image is positioned at position (i, conversion coefficient j) in the sub-piece (block) of 4x4.Because the complexity of image self will just can obtain after it is encoded, the complexity of required image self is by distributing the complexity prediction of finding the encoded image identical with the picture position that needs to distribute initial bit rate in the adjacent encoded image group of the image of initial bit rate with need and coming when therefore image being distributed initial bit rate.And, there is no encoded image in its front because this two field picture is the first frame coded image, and unpredictable its image self complexity, therefore the initial bit rate to its distribution is 51.2kbits.

(3) initial bit rate that every two field picture is distributed is adjusted: there is deviation in the encode code check that drawn of every two field picture with the code check of original allocation, for the code check that makes final gained mates more accurately with target bit rate, the initial bit rate that needs in real time each picture frame to be distributed is in advance adjusted.The factor that initial bit rate that coded image distributes adjusts is treated in influence three, the one, the relative code check error that encoded image produced identical in the encoded image group adjacent with the picture position of need adjustment with the image of need adjustment, the 2nd, the summation of the image place coding relative code check error that layering produced that need adjust, the 3rd, need the image place coding relative code check error change trend of adjusting that layering produced, feedback according to above three aspects is adjusted the code check of the image initial distribution of need adjustment, and this method is specially:

(a) at first in the encoded image group adjacent, find and the identical encoded image in picture position that needs to adjust, obtain the relative code check error E of this encoded image by following formula with the image of need adjustment:

$E=\frac{B_{t\arg \mathrm{et}}-B_{\mathrm{actual}}}{B_{t\arg \mathrm{et}}}$

In the formula, B _TargetBe the code check of this encoded image being intended distribute, B _ActualCode check for this encoded image real income.

(b), draw the code check T that the final image that need are adjusted distributes by following formula according to relative code check error E _{L, k}For:

$T_{l,k}=T_{l,k,\mathrm{init}}(1+k_1{E}_{l,k}+k_2\underset{j=0}{\overset{k}{\mathrm{\Σ}}}{E}_{l,j}\frac{1}{F_l}+k_3(E_{l,k}-E_{l,k-1})F_l)$

In the formula, T _{L, k, init}Be the initial bit rate that image distributed to the need adjustment, E _{L, j}(j=0,1,2,3 ..., k) the relative code check error that produces for the image place layering encoded image that need adjust, l is the coding layering that needs the image place of adjustment, j (j=0,1,2,3 ..., k) be encoded image and the place, position that needs the image of adjustment, k _lFor to current relative code check error E _{L, k}The coefficient that feeds back, k ₂The summation of the relative code check error that produces for the image place layering that need are adjusted

The coefficient that feeds back, k ₃For to relative code check error change trend (E _{L, k}-E _{L, k-1}) coefficient that feeds back, F _lImage place layered encoded frame per second for the need adjustment.

Owing to be first frame, the front does not produce relative code check error, and therefore, the code check of this two field picture still is 51.2kbits after the adjustment.

(4) method with linear model draws the required quantization parameter of coding: for drawing the required quantization parameter of coding, what adopt is that (method of ρ-domain), wherein ρ is the shared ratio of zero coefficient behind the image quantization (zero coefficients) to linear model.The size of ρ is directly related with the code stream size behind the entropy coding.ρ-domain model utilizes the relation of ρ-rate to obtain target ρ from the code check of being given (rate), utilizes the relation of ρ-QP to select required quantization parameter (QP) again, thereby reaches the purpose of Rate Control.The formula that draws ρ is:

$\mathrm{\&rho;}\left(\mathrm{QP}\right)=\frac{1}{M}\underset{i,j}{\mathrm{\&Sigma;}}\underset{\left|x_{\mathrm{ij}}\right|<t(i,j,\mathrm{QP})}{\mathrm{\&Sigma;}}{D}_{\mathrm{ij}}\left(x_{\mathrm{ij}}\right)$

In the formula, M is the quantity of coefficient in the compilation picture to be encoded, x _IjFor in the sub-piece (block) of 4x4, being positioned at position (i, conversion coefficient j), D _Ij((i, j are to judge to be positioned at position (i, conversion coefficient x j) QP) to t for i, the block diagram of variation coefficient j) in order to be positioned at the position _IjWhether be the threshold values of zero coefficient.Have the better linearity relation between the bit rate of ρ with coding back code stream,, be defined as Fig. 4:

$\mathrm{\&rho;}\left(R\right)=\frac{R_0-R*(1-{\mathrm{\&rho;}}_0)}{{\mathrm{<figure-callout\; id="0"\; label="R"\; filenames="A2009101528960023H1.png,A2009101528960023H2.png"\; state="\{\{state\}\}">R</figure-callout>}}_0}$

In the formula, ρ ₀For can be by the computing formula of a given QP from ρ

$\mathrm{\&rho;}\left(\mathrm{QP}\right)=\frac{1}{M}\underset{i,j}{\mathrm{\&Sigma;}}\underset{\left|x_{\mathrm{ij}}\right|<t(i,j,\mathrm{QP})}{\mathrm{\&Sigma;}}{D}_{\mathrm{ij}}\left(x_{\mathrm{ij}}\right)$

In obtain R ₀By using the bit rate of given QP coding back gained.Utilize linear model to draw quantization parameter, as Fig. 6, its concrete steps are:

(a) by an initial quantization parameter QP _InitImage is encoded, then according to the computing formula of ρ:

$\mathrm{\&rho;}\left(\mathrm{QP}\right)=\frac{1}{M}\underset{i,j}{\mathrm{\&Sigma;}}\underset{\left|x_{\mathrm{ij}}\right|<t(i,j,\mathrm{QP})}{\mathrm{\&Sigma;}}{D}_{\mathrm{ij}}\left(x_{\mathrm{ij}}\right)$

Draw ρ ₀And R ₀, in the present embodiment by an initial quantization parameter QP _Init=28 pairs of images are encoded, and obtain ρ ₀=0.774897, R ₀=1.587121;

(b) according to (ρ that is drawn in (a) ₀, R ₀) and (1,0) draw suc as formula

$\mathrm{\&rho;}\left(R\right)=\frac{R_0-R*(1-{\mathrm{\&rho;}}_0)}{{\mathrm{<figure-callout\; id="0"\; label="R"\; filenames="A2009101528960023H1.png,A2009101528960023H2.png"\; state="\{\{state\}\}">R</figure-callout>}}_0}$

Shown ρ-rate model, (0.774897,1.587121) and (1,0) that is drawn in the basis (a) in the present embodiment draws the ρ-rate model of this image:

$\mathrm{\&rho;}\left(R\right)=\frac{1.587121-R*(1-0.774897)}{1.587121}$

(c) the code check T that each coded image is distributed by step (3) _{L, k}Can get R _{L, k}:

$R_{l,k}=\frac{T_{l,k}}{W_{l,k}\&times;H_{l,k}}$

W wherein _{L, k}, H _{L, k}Be respectively the width and height of image to be encoded.According to R _{L, k}Can draw ρ with the ρ-rate model of gained in (b) _{L, k}, the code check that image distributes in the present embodiment is:

$R_{l,k}=\frac{51.2}{176\&times;144}=2.020202(\mathrm{bits}-\mathrm{per}-\mathrm{pixel})$

Get ρ by the ρ in (b)-rate model _{L, k}=0.713473.

(d) according to the ρ in (c) _{L, k}And formula:

$\mathrm{\&rho;}\left(\mathrm{QP}\right)=\frac{1}{M}\underset{i,j}{\mathrm{\&Sigma;}}\underset{\left|x_{\mathrm{ij}}\right|<t(i,j,\mathrm{QP})}{\mathrm{\&Sigma;}}{D}_{\mathrm{ij}}\left(x_{\mathrm{ij}}\right)$

Can draw image to be encoded the quantization parameter QP that should select _{L, k}, with QP=28-1=27 this image is encoded in the present embodiment that to draw the ρ value be 0.756471, the ρ value of QP=26 is 0.735519, the ρ value of QP=25 is 0.708254, so QP _{L, k}=25.

(5) the quantization parameter QP of usefulness gained _{L, k}=25 pairs of images are encoded, and obtaining code check is 51.36kbits, are 0.31% compared to the error of target bit rate 51.2kbits.

According to identical step next image is encoded then.

It more than is the concrete implementation step of present embodiment.Table 1 is that cycle tests " pairs " is in the Rate Control result of method under spatial scalability who adopts embodiment.

Table 1

As can be seen from Table 1, bit rate control method of the present invention has reached good control effect on spatial scalability, and maximum code check error is less than 1%.

Table 2 has now based on the bit rate control method in the motions of scalable video of linear model in the control effect aspect the accurate coupling of code check for adopting.

Table 2

For having three sheaf space scalability (qcif, cif, sequence 4cif) " pairs ", by to adopting Rate Control effect of the present invention comparing aspect the accurate coupling of code check based on the bit rate control method in the motions of scalable video of linear model with existing, by table 1 and table 2 as can be seen: for basic layer (qicf), method of the present invention has reduced 0.68% than the relative code check error of prior art; Strengthen one deck (cif) for the space, method of the present invention has reduced 0.16% than the relative code check error of prior art; Strengthen and layer (4cif) for the space, method of the present invention has reduced 0.58% than the relative code check error of prior art.

Embodiment 2:

As shown in Figure 2, present embodiment handle to as if having a sequence " soccer " of three layers of quality scalability.Parameter configuration is: GOPSize=8, frame per second=30frames/s, the target bit rate of each quality layering correspondence of cycle tests " soccer " be 240,480,720}kbits/s, used sequence " soccer " is of a size of cif (352x288).

The step of present embodiment (1), step (4), step (5) are the same with embodiment 1, and step (2) and step (3) are different with embodiment's 1.

Every two field picture in each layering distributes that code check adopts in to scalable video coding in step (2) is that method is in the present invention: introduced gradable B frame structure in the scalable video coding, as shown in Figure 5, it is a pyramid structure, the bottom is high frequency B frame, from low to high, frequency reduces successively, and its contribution to whole coding efficiency is also increasing, and pyramidal top is the I frame of intraframe coding or the P frame of interframe forward predictive coded.Therefore, to have adopted based on coded image self complexity and the weight in image sets thereof be the Rate Distribution Strategy that the factor combines to method of the present invention.Code rate allocation method to each picture frame in each image sets in each layering is:

If this image is the I/P frame, then through type (I) obtains distributing to the code check T of this image _l ^I/PFor:

$T_l^{I/P}=\frac{{\mathrm{\&omega;}}^{I/P}{G}_l}{{\mathrm{\&omega;}}^{I/P}+\underset{m=0}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_b\left(m\right)}$

Formula (I)

T in the formula (I) _l ^I/PFor distributing to the code check of I/P frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, w ^I/PBe the weight coefficient of I/P frame in the image sets to be encoded, G _lBe the code check that the present image group is distributed, d represents total progression of time domain classification, the time domain progression at B frame place in the m presentation video group, N _b(m) be the quantity of uncoded B frame in the m level time domain classification, w _mIt is the weight coefficient of B frame in the m level time domain classification.

If this image is the B frame, then through type (II) obtains distributing to the code check T of this image ^B _{L, ji}For:

$T_{l,j}^B=\frac{{\mathrm{\&omega;}}_j{G}_j}{\underset{m=j}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_B\left(m\right)}$

Formula (II)

T in the formula (II) ^B _{L, j}For distributing to the code check of B frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate; J is the time domain classification at the B frame place of code check to be allocated; D represents the progression in the time domain classification; N _B(m) be the quantity of uncoded B frame in the m level time domain classification; G _jBe the last code check of present image group; w _mIt is the weight coefficient of B frame in the m level time domain classification, be defined as the product of coded image self complexity and the image importance in image sets, the importance of image in image sets is { 1.000,1.216,1.478,1.794} self complexity of coded image is provided by the prediction of the complexity of image at the same level among the adjacent GOP that has encoded.

The present invention adopts the relative code check error that encoded image produced identical with the picture position of need adjustment in the feedback encoded image group adjacent with the image that needs adjustment that the initial bit rate of every two field picture distribution is adjusted the adjusted code check T of every two field picture in step (3) _{L, k}For:

T _l，k＝T _l，k，init(1+k _lE _l，k)

In the formula, T _{L, k, init}Be the initial bit rate that image distributed to the need adjustment, E _{L, k}The relative code check error that the identical encoded image in picture position that is to adjust with need in the adjacent encoded image group of the image adjusted with need produces, l is the coding layering at the image place that needs to adjust, the place, position of the image that k need adjust, k _lFor to current relative code check error E _{L, k}The coefficient that feeds back.

Table 3 is that cycle tests " soccer " is adopting the Rate Control result of present embodiment method under the mass domain scalability.

Table 3

Bit rate control method of the present invention has as can be seen from Table 3 reached good effect on the mass domain scalability, the code check error is generally below 1%, and maximum code check error is less than 2%.

As shown in Figure 7, for the sequence with three layers of mass domain scalability " soccer ",, contrasting aspect the PSNR gain based on the bit rate control method in the motions of scalable video of linear model with existing to high code check from low code check adopting Rate Control effect of the present invention.As seen from Figure 7, for basic layer (cif), code check is between 240kbps and 600kpbs, and the PSNR value has improved 0.04-0.53dB; Strengthen one deck (cif) for quality, code check is between 480kbps and 1200kbps, and the PSNR value has improved 0.04-0.60dB; Strengthen two layers (cif) for quality, code check is between 720kbps and 1800kbps, and the PSNR value has improved 0.03-0.41dB.

Embodiment 3:

As shown in Figure 3, the object handled of present embodiment has three sheaf space scalabilities { qcif, cif, 4cif}, each space delamination has the sequence " pairs " of the associating scalability of two quality layerings.Parameter configuration is: GOPSize=8, frame per second=30frames/s, the target bit rate of cycle tests " pairs " from basic layer to each enhancement layer be 96,192,288,576,864,1728}kbits/s.

The step of present embodiment is the same with the step of embodiment 1 substantially, the method that every two field picture during different is in step (2) to scalable video coding in each layering distributes code check to adopt is: introduced gradable B frame structure in the scalable video coding, as Fig. 5, it is a pyramid structure, the bottom is high frequency B frame, from low to high, frequency reduces successively, its contribution to whole coding efficiency is also increasing, and pyramidal top is the I frame of intraframe coding or the P frame of interframe forward predictive coded.Therefore, to have adopted based on coded image self complexity and the weight in image sets thereof be the Rate Distribution Strategy that the factor combines to method of the present invention.Code rate allocation method to each picture frame in each image sets in each layering is:

If this image is the I/P frame, then through type (I) obtains distributing to the code check T of this image _l ^I/PFor:

$T_l^{I/P}=\frac{{\mathrm{\&omega;}}^{I/P}{G}_l}{{\mathrm{\&omega;}}^{I/P}+\underset{m=0}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_b\left(m\right)}$

Formula (I)

T in the formula (I) _l ^I/PFor distributing to the code check of I/P frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, w ^I/PBe the weight coefficient of I/P frame in the image sets to be encoded, G _lBe the code check that the present image group is distributed, d represents total progression of time domain classification, the time domain progression at B frame place in the m presentation video group, N _b(m) be the quantity of uncoded B frame in the m level time domain classification, w _mIt is the weight coefficient of B frame in the m level time domain classification.

If this image is the B frame, then through type (II) obtains distributing to the code check T of this image ^B _{L, ji}For:

$T_{l,j}^B=\frac{{\mathrm{\&omega;}}_j{G}_j}{\underset{m=j}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_B\left(m\right)}$

Formula (II)

T in the formula (II) ^B _{L, ji}For distributing to the code check of B frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, and j is the time domain classification at the B frame place of code check to be allocated, and d represents the progression in the time domain classification, N _B(m) be the quantity of uncoded B frame in the m level time domain classification, G _jBe the last code check of present image group, w _mIt is the weight coefficient of B frame in the m level time domain classification.

Under the situation of identical configuration, adopt and existingly cycle tests is encoded, and the result is compared with the method that adopts present embodiment based on the bit rate control method in the motions of scalable video of linear model.Table 4 is the control result of cycle tests " pairs " under the bit rate control method of present embodiment, table 5 be existing based on the result under the bit rate control method in the motions of scalable video of linear model, table 6 is comparing results of present embodiment and existing method.

Table 4

Table 5

Table 6

As can be seen, for sequence " pairs ", adopt bit rate control method of the present invention from table 4 and table 5, the code check error generally is no more than 1%, and maximum code check error is 1.31%, simultaneously the maximal rate error 3.77% of existing method; As can be seen from Table 6 aspect the code check error, the bit rate control method of the present invention existing methodical error of comparing has reduced 0.00-2.87%, and aspect the PSNR gain, bit rate control method of the present invention has also improved 0.11-0.64dB than existing method.Therefore for the associating scalability with space and quality layering, bit rate control method of the present invention has better Rate Control effect than existing method raising preferably being arranged all aspect the accurate coupling of code check and the PSNR value two.

Claims (3)

1. one kind based on the bit rate control method in the scalable video coding of linear model, may further comprise the steps: (1) distributes code check to the image sets in each layering of scalable video coding; (2) the every two field picture in the described image sets of step (1) is distributed initial bit rate; (3) initial bit rate that every two field picture is distributed is adjusted; (4) draw the required quantization parameter of coding of described every two field picture with the method for linear model; (5) with the quantization parameter of gained described every two field picture is encoded, it is characterized in that the method for in the described step (3) initial bit rate of every two field picture distribution being adjusted is:

(a) at first in the encoded image group adjacent, find and the identical encoded image in picture position that needs to adjust, obtain the relative code check error E of this encoded image by following formula with the image of need adjustment

$E=\frac{B_{t\arg \mathrm{et}}-B_{\mathrm{actual}}}{B_{t\arg \mathrm{et}}},$

In the formula, B _TargetBe the code check of this encoded image being intended distribute, B _ActualCode check for this encoded image real income;

(b), draw the code check T that the final image that need are adjusted distributes by following formula according to relative code check error E _{L, k}For

$T_{l,k}=T_{l,k,\mathrm{init}}(1+k_1{E}_{l,k}+k_2\underset{j=0}{\overset{k}{\mathrm{\&Sigma;}}}{E}_{l,j}\frac{1}{F_l}+k_3(E_{l,k}-E_{l,k-1})F_l),$

In the formula, T _{L, k, init}Be the initial bit rate that image distributed to the need adjustment, E _{L, j}(j=0,1,2,3 ..., k) the relative code check error that produces for the image place layering encoded image that need adjust, l is the coding layering that needs the image place of adjustment, j (j=0,1,2,3 ..., k) be encoded image and the place, position that needs the image of adjustment, k ₁For to current relative code check error E _{L, k}The coefficient that feeds back, k ₂The summation of the relative code check error that produces for the image place layering that need are adjusted

The coefficient that feeds back, k ₃For to relative code check error change trend (E _{L, k}-E _{L, k-1}) coefficient that feeds back, F _lImage place layered encoded frame per second for the need adjustment.

2. according to claim 1 a kind of based on the bit rate control method in the scalable video coding of linear model, it is characterized in that distributing the method for initial bit rate to be to every two field picture in the described step (2):

If this image is the I/P frame, then through type (I) obtains distributing to the code check T of this image _l ^I/PFor

$T_l^{I/P}=\frac{{\mathrm{\&omega;}}^{I/P}{G}_l}{{\mathrm{\&omega;}}^{I/P}+\underset{m=0}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_b\left(m\right)}$

Formula (I),

In the formula (I), T _l ^I/PFor distributing to the code check of I/P frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, w ^I/PBe the weight coefficient of I/P frame in the image sets to be encoded, G _lBe the code check that the present image group is distributed, d represents total progression of time domain classification, the time domain progression at B frame place in the m presentation video group, N _b(m) be the quantity of uncoded B frame in the m level time domain classification, w _mIt is the weight coefficient of B frame in the m level time domain classification;

If this image is the B frame, then through type (II) obtains distributing to the code check T of this image ^b _{L, ji}For

$T_{l,j}^B=\frac{{\mathrm{\&omega;}}_j{G}_j}{\underset{m=j}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_B\left(m\right)}$

Formula (II),

In the formula (II), T ^B _{L, j}For distributing to the code check of B frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, and j is the time domain classification at the B frame place of code check to be allocated, and d represents the progression in the time domain classification, N _B(m) be the quantity of uncoded B frame in the m level time domain classification, G _jBe the last code check of present image group, w _mIt is the weight coefficient of B frame in the m level time domain classification.

3. one kind based on the bit rate control method in the scalable video coding of linear model, may further comprise the steps: (1) distributes code check to the image sets in each layering of scalable video coding; (2) the every two field picture in the described image sets of step (1) is distributed initial bit rate; (3) initial bit rate that every two field picture is distributed is adjusted; (4) draw the required quantization parameter of coding of described every two field picture with the method for linear model; (5) with the quantization parameter of gained described every two field picture is encoded, it is characterized in that distributing the method for initial bit rate to be to every two field picture in the described step (2):

If this image is the I/P frame, then through type (I) obtains distributing to the code check T of this image _l ^I/PFor

$T_l^{I/P}=\frac{{\mathrm{\&omega;}}^{I/P}{G}_l}{{\mathrm{\&omega;}}^{I/P}+\underset{m=0}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_b\left(m\right)}$

Formula (I),

In the formula (I), T _l ^I/PFor distributing to the code check of I/P frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, w ^I/PBe the weight coefficient of I/P frame in the image sets to be encoded, G _lBe the code check that the present image group is distributed, d represents total progression of time domain classification, the time domain progression at B frame place in the m presentation video group, N _b(m) be the quantity of uncoded B frame in the m level time domain classification, w _mIt is the weight coefficient of B frame in the m level time domain classification;

If this image is the B frame, then through type (II) obtains distributing to the code check T of this image ^B _{L, ji}For

$T_{l,j}^B=\frac{{\mathrm{\&omega;}}_j{G}_j}{\underset{m=j}{\overset{d-1}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_m{N}_B\left(m\right)}$

Formula (II),

In the formula (II), T ^B _{L, j}For distributing to the code check of B frame in the image sets to be encoded, l is for distributing the coding layering at the image sets place of initial bit rate, and j is the time domain classification at the B frame place of code check to be allocated, and d represents the progression in the time domain classification, N _B(m) be the quantity of uncoded B frame in the m level time domain classification, G _jBe the last code check of present image group, w _mIt is the weight coefficient of B frame in the m level time domain classification.

Images