CN1186933C - Parallel image sequence bit rate controlling method for digital TV video coder - Google Patents

Abstract

The present invention relates to a control method for a parallel image sequence code rate of a digital TV video coder. Multiple paths of video program sequences belonging to different data sources or one path of parallel image sequence which is divided in a space domain or a frequency domain by a data sources is input, and a parallel coding unit respectively carries out the parallel real-time MPEG-2 variable bit rate coding for corresponding image sequences; an image frame in the same time of each of the image sequences is defined as a superframe, a super image group is defined, and the size is the least common multiple of the size of each of image groups; the stratification combined code rate control of the image frame, the superframe and the super image group can be carried out bv the main control code rate control unit, the code stream composition and the code rate transform of multiple paths of output compressed data flow are carried out by an after-treatment unit, and data is wrapped through a multiplexer and then enters a channel for the transmission through a buffer zone. The present invention enables the parallel image sequences to keep the image quality and the fine smoothness which are closely identical so that a phenomenon of inhomogeneous quality for all image sequences because the code rate is halved can be avoided.

Description

The digital TV video frequency encoder image sequence bit rate control method that walks abreast

Technical field:

The present invention relates to the parallel image sequence bit rate control method of a kind of digital TV video frequency encoder, can be applied in constant rate of speed (CBR) Network, carry out the Data Rate Distribution occasion of multiple data stream code multiplexing.Belong to technical field of video coding.

Technical background:

Current, general digital tv encoder technology is the MPEG-2 technology, it is by the ISO/IEC13818 file description, this document only stipulated code stream that video encoder is compiled out the syntax gauge that must observe, comprise group of picture, picture frame, macro block bar, macro block, piece hierarchical structure, wherein picture frame is divided into intracoded frame, forward-predictive-coded frames, bi-directional predictive coding frame again according to the characteristic of time domain classification, but standard is not made restriction provision to the concrete grammar of Rate Control.Bit rate control method commonly used is the TM5 algorithm, it is comprised the picture frame in the group of picture is made Bit Allocation in Discrete, carried out the macro block Rate Control, carries out adaptive quantizing adjustment three phases at the macro block complexity according to the exuberant degree of virtual buffering region by the ISO/IEC/JTC1/SC29/WG11/N0400 file description.Development along with Digital Television and Access Network, in constant rate of speed (CBR) Network, carry out the Polymera parallel encoding and multiplexing transmission becomes important application, experimental results show that simultaneously, the variable bit rate (VBR) that they carry out under the unicode rate control is encoded, with comparison they to carry out the average bit rate allocative decision of independent constant rate of speed coding more effective, not only improve channel utilization, and agree with the trend of following multimedia evolution.This means that handling structure at the image sequence parallel distributed need propose a kind of general and efficient unicode rate control method, can guarantee provides consistent image program quality to the user.And the TM5 method is to come quantization parameter is adjusted according to the exuberant degree and the macro block activity in the picture frame of virtual buffering region, the realization Data Rate Distribution, it can't be applied to the situation of a plurality of encoder concurrent collaborative Rate Control.Granted publication number be CN 1058360C patent application the code rate allocation method of a parallel image subsection in the high definition television video encoder, distribute code check according to the activity of image subsection according to logarithmic relationship.But it is too simplified because of the realization of looking after on the hardware, does not possess generality and generalization, can influence the performance of this type of encoder under most scene situations of change.

Summary of the invention:

The objective of the invention is at the deficiencies in the prior art, a kind of digital TV video frequency encoder image sequence bit rate control method that walks abreast is provided, can realize a plurality of encoder concurrent collaborative Rate Control, hardware configuration is simple, and has a good divergence, be applicable to separate multi-channel video program, and the height correlation degree subgraph sequence in the parallel processing plan of high definition digital television employing.

For realizing such purpose, the present invention proposes a kind of multiplexing effective code check allocative decision of multiple data stream variable bit rate coding of carrying out in the constant rate of speed Network, particularly is layering unicode rate control method and the implementation structure of digital TV video frequency encoder by fixed channel parallel encoding transmission multichannel image program.It adopts a kind of host-guest architecture, the master control Rate Control unit of realizing by high-speed dsp adheres to separately under the cooperation of DSP unit at each parallel encoder, total bitrate is carried out reasonable distribution according to the consistent distortion factor between parallel image sequence, make and to keep between the parallel image sequence avoiding because of dividing the uneven phenomenon of each image sequence quality that code check causes equally near consistent image quality and good smoothness.

Digital TV video frequency encoder of the present invention comprises: multichannel image sequence input unit, parallel encoding unit (comprising m parallel encoder), post-processing unit, multiplexer, buffering area and master control Rate Control unit.Multichannel image sequence input unit can refer to that the multichannel of importing is under the jurisdiction of the video frequency program sequence in different pieces of information source, also can refer to the parallel image subsection sequence that is marked off at spatial domain or frequency domain by a circuit-switched data source; To walk abreast real-time MPEG-2 coding of corresponding image sequence, master control Rate Control unit adopts layering unicode rate control method of the present invention to carry out Data Rate Distribution on the data syntax hierarchy basis of Moving Picture Experts Group-2 definition respectively in the parallel encoding unit; Post-processing unit can carry out the synthetic and rate conversion of code stream to the compressed and multiplexed data flow of parallel encoding unit output, carries out packing data by multiplexer, enters Channel Transmission through buffering area.

Master control Rate Control unit adopts the digital TV video frequency encoder layering unicode rate control method according to parallel map architecture design.Data syntax structure according to the Moving Picture Experts Group-2 definition, it is a superframe that the present invention defines the picture frame that each image sequence is positioned at synchronization, on its basis, group of picture at each parallel image sequence different structure, size, define a hyper-image group, its size is the least common multiple of each group of picture size.Picture frame, superframe, hyper-image group constitute the unit of layering unicode rate control, and correspondence comprises following stratification step with it:

1, at first, the parallel image sequence of importing is carried out various preliminary treatment, comprise each image sequence is taken a sample and scene detection.

When detecting the image sequence occurrence scene conversion of a certain road, judge whether the current group of picture of this sequence is last group of picture that this image sequence is positioned at current hyper-image group: if then this scene change picture frame is become intracoded frame and compress; If not, then residual graph picture frame in the current group of picture of this image sequence and its next group of picture are merged to constitute a new group of picture, this scene change picture frame is handled by intracoded frame, and the intra-coded of next group of picture then becomes forward-predictive-coded frames and handles.

2, then, make the unit, distribute the target code word according to channel code rate with the hyper-image group:

Whenever superframe coding is finished, need to upgrade the characteristic parameter of target code word and each sequence of hyper-image group, as average quantisation parameter etc.

3, in the hyper-image group, the composite image superframe is carried out Data Rate Distribution:

The complexity that defines a picture frame in each sequence is estimated, and is the last same number of frames type of this sequence (intracoded frame, forward-predictive-coded frames or the bi-directional predictive coding frame) generated codeword of picture frame and the product of its each macro block average quantisation parameter; Superframe is carried out Data Rate Distribution according to the ratio that its synthetic complexity accounts for current hyper-image group, and the hypothesis complexity of each picture frame of not encoding in the superframe is identical with the last same number of frames type map of corresponding sequence picture frame therebetween;

4, then, each parallel picture frame in the superframe is carried out Data Rate Distribution:

According to transition coding rate distortion statistical model, based on the average quantisation parameter and the generated codeword of code pattern picture frame, calculate the average pixel bit number of present encoding picture frame, each picture frame is carried out Data Rate Distribution according to the product proportion of its average pixel bit number and picture resolution;

5, the upper and lower threshold value of Data Rate Distribution is set, when by above-mentioned 3) and 4) when the code check that calculates of step exceeds this upper and lower thresholding, all by threshold value.

6, in macroblock layer, according to carrying out the macro block Rate Control, carry out the adaptive quantizing adjustment according to the exuberant degree of virtual buffering region in the TM5 method at the macro block complexity.

Digital TV video frequency encoder of the present invention is by the layering unicode rate control method of constant rate of speed network channel parallel encoding transmission multichannel image program, the algorithm of reasonable distribution is carried out total bitrate in employing according to the consistent distortion factor between parallel image sequence, make and to keep between the parallel image sequence avoiding because of dividing the uneven phenomenon of each image sequence quality that code check causes equally near consistent image quality and good smoothness.The hardware configuration that computing of the present invention realizes is simple, and has good divergence.No matter be with respect to separate multi-channel video program, still for the height correlation degree subgraph sequence in the parallel processing plan that high definition digital television adopts, method of the present invention all has applicability.

Description of drawings:

Fig. 1 is the digital TV video frequency coder structure block diagram of the parallel coding sequence of pictures multiplexing transmission in m road.

Wherein, comprise multichannel image sequence input unit, by the parallel encoding unit that m parallel encoder constitutes, post-processing unit, multiplexer, buffering area and master control Rate Control unit.Multichannel image sequence input unit input multichannel or be under the jurisdiction of the video frequency program sequence in different pieces of information source, perhaps parallel image subsection sequence for a circuit-switched data source is marked off at spatial domain or frequency domain; Parallel encoding unit (1-m) is respectively to walk abreast real-time MPEG-2 variable bit rate (VBR) coding of corresponding image sequence, master control Rate Control unit is on the data syntax hierarchy basis of Moving Picture Experts Group-2 definition, for guaranteeing consistent image quality, adopt layering unicode rate control method of the present invention to carry out Data Rate Distribution.Post-processing unit can carry out code stream synthetic (contrasting a circuit-switched data source) and rate conversion (contrast channel conditions) to the compressed and multiplexed data flow of parallel encoding unit output, carry out packing data (carrying out respective handling) by multiplexer, enter Channel Transmission through buffering area according to program stream PS or transport stream TS.

Fig. 2 is the high definition television encoder parallel organization block diagram according to first embodiment of the invention.

Fig. 3 divides schematic diagram according to the subgraph of first embodiment of the invention.

Fig. 4 is the sub-encoders structure chart according to first embodiment of the invention.

Fig. 5 is the master control Rate Control cell schematics according to first embodiment of the invention.

Fig. 6 is the video encoder structured flowchart according to four road image program parallel encodings transmission of second embodiment of the invention.

Embodiment:

Below in conjunction with accompanying drawing and specific embodiment technical scheme of the present invention is further described.

Fig. 2 is the parallel organization block diagram of one embodiment of the invention, is applied in the multiple data stream parallel encoding that derives from No. one high definition TV data source (picture resolution 1920 * 1080).

Because the high-resolution of high-definition screen, desired amount of calculation is very huge, needs high disposal ability and high speed data transfer ability, and this has brought great difficulty for the design of hardware system.Usually, for the complex calculation that realizes this big data quantity adopts parallel processing technique, be about to a high definition TV picture and be divided into a few height pieces, with a sub-encoders piece is carried out real-time coding, the concurrent collaborative work of several sub-encoders can be finished the real-time coding to the high definition video sequence, and the pattern of division can be selected by the designer, as: cross is divided, vertical division, horizontal bar is graded, and the subgraph partition mode is not in category of the present invention.

Among Fig. 2: the high definition video input signal in the image acquisition unit (A/D conversion) sample, be divided into 6 image subsections (as Fig. 3) through subgraph then: middle 4 sub-image size are 1920 * 192, and upper and lower image subsection size is 1920 * 160 (respectively mending four lines).Subgraph sequential coding module (1-6) is corresponding to the parallel encoder among Fig. 1 (1-m), they carry out real-time MPEG-2 coding (see figure 4) to corresponding subgraph sequence respectively, each subgraph MPEG-2 compressed bit stream of output carries out code stream in unit, subgraph sequential coding code stream interflow synthetic, formation meets the high definition video stream of MPEG-2 MP@HL standard, and its picture resolution is consistent with the high definition video image of input.The mode of operation of encoder is " transparent " to decoding end, does not reflect the feature of its parallel encoding from its output code flow.Master Control Unit plays a part initialization setting and monitoring to all working module.Master control Rate Control unit carries out unicode rate control to each sub-encoders.Fig. 3 is the master control Rate Control DSP cellular construction figure that realizes unicode rate control core Data Rate Distribution, it is according to the characteristic parameter that the DSP unit is sent that adheres to separately in 6 parallel subgraph sequential coding modules (1-6), through jointly controlling the computing of algorithm,, each distributes the phase code rate for adhering to DSP separately.

For keeping the consistency and the flatness of HDTV picture frame quality, establish the group of picture size and be N:

1) the code check adaptive equalization of scene change is carried out in preliminary treatment to the hyper-image group;

2) hyper-image group size also is N, carries out the target code assignment according to channel code rate, and after every coding is finished a superframe real-time update;

3) the synthetic superframe to the HDTV image subsection carries out Data Rate Distribution;

At first, the HDTV subgraph picture frame in the superframe is set up from recurrence image complexity measure model:

$C_{[j,n,t]}={\stackrel{\‾}{Q}}_{[n-1,t]}\×S_{[j,n-1,i]},{\stackrel{\‾}{Q}}_{[n-1,t]}=\frac{1}{6}\underset{j=1}{\overset{6}{\mathrm{\Σ}}}{Q}_{[j,n-1,t]},t\∈\{1,P,B\}$

C wherein _{[j, n, t]}The complexity that refers to j subgraph frame in n the HDTV superframe is estimated (frame type is t), Q _{[j, n-1, t]}And S _{[j, n-1, t]}Refer to that respectively the frame type of having encoded also is the average quantization step-length and the number of coded bits of the last subgraph frame of t.

This HDTV superframe is carried out Data Rate Distribution by following formula;

$T_{\sup -\mathrm{fram}{e}_{[n,t]}}=\frac{\underset{j=1}{\overset{6}{\mathrm{\Σ}}}{C}_{[j,n,t]}/K_t}{\underset{j=1}{\overset{6}{\mathrm{\Σ}}}[N_i\frac{C_{[j,i]}}{k_i}+N_p\frac{C_{[j,p]}}{K_p}+N_b\frac{C_{[j,b]}}{K_b}]}{T}_{\sup -G}$

C wherein _{[j, i]}, C _{[j, p]}And C _{[j, b]}Be each HDTV subgraph sequence correspondence the up-to-date complexity of three kinds of picture frame types estimate, every coding needs real-time update after finishing a subgraph frame; N _i, N _pAnd N _bBe three kinds of remaining separately frame number of frame type in the current hyper-image group; K _i, K _p, K _bBe the frame type factor; T _Sup-GTarget code word for current hyper-image group.

4) the HDTV subgraph picture frame in the synthetic superframe is carried out Data Rate Distribution;

At first, to j subgraph frame in n the HDTV superframe (frame type is t), according to its last characteristic parameter of having encoded the subgraph frame set up the average pixel bit rate from recursive models:

${\hat{b}}_{[j,n,t]}=\frac{1}{1.386}\ln \frac{P_{[j,n-1]}}{{\hat{D}}_{n-1}},$

Wherein

The information source distributed constant $P_{[j,n-1]}=\frac{{Q_{[j,n-1]}}^2}{\mathrm{\β}\×L}{e}^{\mathrm{\α}{\stackrel{\‾}{b}}_{[j,n-1]}}\underset{k=0}{\overset{L-1}{\mathrm{\Σ}}}{W_k}^2,$ ${\stackrel{\‾}{b}}_{[j,n-1]}=\frac{S_{[j,n-1]}}{H_j\×V_j},$ H _i* V _iBe the resolution (number of picture elements) of i subgraph, β is signal probability distributed constant (generally getting 12 in the high-speed data service);

Average distortion measure ${\hat{D}}_{n-1}=\frac{1}{6}\underset{j=1}{\overset{6}{\mathrm{\Σ}}}{\stackrel{\‾}{D}}_{[j,n-1]}=\frac{1}{6\×\mathrm{\β}\×L}\underset{j=1}{\overset{6}{\mathrm{\Σ}}}{Q_{[j,n-1]}}^2\underset{k=0}{\overset{L-1}{\mathrm{\Σ}}}{W_k}^2;$

W _kBe based on the weight coefficient of block transform coding (dct transform among the MPEG-2) piece transformation matrix, L is the transformation matrix size.

This subgraph frame is carried out Data Rate Distribution by following formula;

$T_{\mathrm{sub}-\mathrm{fram}{e}_{[j,n,t]}}=\frac{{\hat{b}}_{[j,n,t]}\×H_j\×V_j}{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\hat{b}}_{[i,n,t]}\×H_i\×V_i}{T}_{\sup -\mathrm{fram}{e}_{[n,t]}}$

5) the upper and lower threshold value of Data Rate Distribution is set, when by above-mentioned 3) and 4) when the code check that calculates of step exceeds this upper and lower thresholding, all by threshold value.

6) macroblock layer is according to carrying out the macro block Rate Control, carry out the adaptive quantizing adjustment at the macro block complexity according to the exuberant degree of virtual buffering region in the TM5 method.

Experimental result shows: if do not adopt bit rate control method of the present invention, and only be that to fix each sub-encoders output average bit rate be 1/6 HDTV channel code rate, then at original picture especially when certain image subsection scene is complicated, can cause final image reconstruction that tangible distortion is arranged in subjective assessment.If adopted unicode rate control method of the present invention, adopt the HDTV channel code rate of 18Mbits/s～60Mbits/s experimentally, above-mentioned human eye interference phenomenon disappears.

Fig. 3 is the encoder block diagram of another embodiment of the present invention, is used for realizing that four way word TV programme transmit at constant Network parallel encoding.Encoder wherein (1-4) carries out the MPEG-2 real-time coding respectively to four road program sequences of input respectively, the encoder here can be that main chipset constitutes with special chip, also can be made of programmable logic device, high-speed dsp unit and movement estimation chip.Based on same theory and method, master control Rate Control unit is optimized Data Rate Distribution and multiplexing to four road program sequences, and this moment, least common multiple was got in the setting of hyper-image group because each road program sequence has different group of picture sizes.The code stream of output outputs to channel by multiplexer, and according to the channel conditions of buffering area feedback and user's qos requirement, master control Rate Control unit can be adjusted Rate Distribution Strategy neatly, does smoothly or takes statistics multiplexing in spatial domain in time domain.Obtained excellent control effect equally, final four road image programs have shown the image quality of approaching unanimity in the output of decoder.

Claims (2)

1, a kind of digital TV video frequency encoder image sequence bit rate control method that walks abreast, it is characterized in that being under the jurisdiction of the video frequency program sequence in different pieces of information source by multichannel image sequence input unit input multichannel, or the parallel image subsection sequence that marks off at spatial domain or frequency domain of circuit-switched data source, the parallel encoding unit is respectively to corresponding image sequence real-time MPEG-2 variable bit rate coding that walks abreast, defining the picture frame that each image sequence is positioned at synchronization is a superframe, at each parallel image sequence different structure, the group of picture of size, define a hyper-image group, its size is the least common multiple of each group of picture size, by master control Rate Control unit to picture frame, superframe, the hyper-image group is carried out the control of layering unicode rate, post-processing unit carries out the synthetic and rate conversion of code stream to the compressed and multiplexed data flow of parallel encoding unit output, carry out packing data by multiplexer, enter Channel Transmission through buffering area.

2,, it is characterized in that the control of layering unicode rate comprises following stratification step as the parallel image sequence bit rate control method of the said digital TV video frequency encoder of claim 1:

1) the parallel image sequence to input carries out various preliminary treatment, comprise each image sequence is taken a sample and scene detection, when detecting the image sequence occurrence scene conversion of a certain road, whether the current group of picture of judging this sequence is last group of picture that this image sequence is positioned at current hyper-image group, if, then this scene change picture frame being become intracoded frame compresses, if not, then residual graph picture frame in the current group of picture of this image sequence and its next group of picture are merged to constitute a new group of picture, this scene change picture frame is handled by intracoded frame, and the intra-coded of next group of picture then becomes forward-predictive-coded frames and handles;

2) make the unit with the hyper-image group, distribute the target code word,, need the target code word of renewal hyper-image group and the characteristic parameter of each sequence whenever superframe coding is finished according to channel code rate;

3) in the hyper-image group composite image superframe is carried out Data Rate Distribution: the complexity that defines a picture frame in each sequence is estimated, be the generated codeword of the last same number of frames type map of this sequence picture frame and the product of its each macro block average quantisation parameter, superframe is carried out Data Rate Distribution according to the ratio that its synthetic complexity accounts for current hyper-image group, and the hypothesis complexity of each picture frame of not encoding in the superframe is identical with the last same number of frames type map of corresponding sequence picture frame therebetween;

4) each parallel picture frame in the superframe is carried out Data Rate Distribution: according to transition coding rate distortion statistical model, based on the average quantisation parameter and the generated codeword of code pattern picture frame, calculate the average pixel bit number of present encoding picture frame, each picture frame is carried out Data Rate Distribution according to the product proportion of its average pixel bit number and picture resolution;

5) the upper and lower threshold value of Data Rate Distribution is set, when the code check that calculates exceeds this upper and lower thresholding, all by threshold value;

6) in macroblock layer, according to carrying out the macro block Rate Control, carry out the adaptive quantizing adjustment according to the exuberant degree of virtual buffering region in the TM5 method at the macro block complexity.

Images