CN104683804A - Parameter-adaptive multidimensional bit rate control method based on video content characteristics - Google Patents

Abstract

The invention relates to the field of video decoding, in particular to a parameter-adaptive multidimensional bit rate control method based on video content characteristics. The method comprises the following steps: determining an initial coding frame rate according to a channel bandwidth and an initial frame complexity of a video sequence; determining a model parameter update cycle according to a dynamically transformed video, and extracting a video time-space complexity from a sliding window of the current update cycle; meanwhile, determining the length of the next update cycle by combining the bit rate control accuracy of the current update cycle, and updating the multidimensional bit rate control model parameters and the coding parameters. The parameter-adaptive multidimensional bit rate control method based on video content characteristics is superior to internationally similar relevant methods, can effectively improve the subjective quality and objective quality, especially for the video sequence with high time-space complexity, and can slightly reduce the computational complexity.

Description

Based on the parameter adaptive multidimensional bit rate control method of video content features

Technical field

The present invention relates to video decoding filed, specifically, relate to the bit rate control method of technical field of video coding, particularly based on the parameter adaptive multidimensional bit rate control method of video content features, the network video coder multidimensional bit rate control method in Multielement integration heterogeneous network, band broad range dynamic change occasion can be applied to.

Background technology

Along with the development of multimedia communication and network technology, people are more and more extensive to the application demand of video, and video information just plays a part more and more important to the life style of people and social development; Video Coding Compression Technology is the prerequisite of the links such as video storage, transmission, broadcasting, is the core technology of the application such as Digital Television, video monitoring, Internet video; Rate control techniques is one of video compression and the most important link in transmitting, the object of Rate Control is under certain actual channel bandwidth, improve the subjective and objective quality of Video coding reconstructed image as much as possible, and the code check being responsible for making video encoder produce and actual channel bandwidth match, therefore its effect is in the encoder very important.

Existing bit rate control method is all balance code stream based on the adjustment of quantization parameter, and existing method generally can only provide a kind of video frequency output of credit rating, this way is inadequate for the applying flexible degree that bandwidth fluctuation is large, cannot make comprehensive video optimal quality in the real-time low bit-rate situation of bandwidth; Multidimensional Rate Control is exactly be suggested under this background, multidimensional Rate Control organically combines regulating the basic means of code check, and code rate adjustment means such as the conversion of traditional code check (change quantization step), spatial resolution (sdi video size), temporal analytical density (frame per second) etc. are combined the Automatic adjusument of having worked in coordination with video code flow; Due to each dimension size of multidimensional Rate Control, to determine to depend on a period of time code check how many, do not need to be divided in advance some grades in actual applications, effectively can solve SVC computation complexity and the high problem of redundancy.

But current, multidimensional rate control techniques is also in the starting stage, also have the research of more tool expansions to have pending, the research of its key technology has important practical significance for the development of video industry; There are the following problems for current multidimensional bit rate control method: (1) is main stream approach at present, such as Q-R-TQ etc. are based on whole video information or by the fixed cycle when calculating video content information, and become when the content of video is in actual applications, for changing in time, its accuracy of model coefficient that violent sequence obtains is not high, is unfavorable for real-time Transmission; (2) initial being chosen in multidimensional Rate Control of frame per second plays an important role; If the initial frame per second chosen is excessive, after initial frame of only encoding, the actual bit number consumed will exceed preallocated target bit, thus the bit number leaving next code frame for will be little, makes reconstruction video image quality decrease; Otherwise, choose too small initial parameter, not only make image quality decrease, and the waste of bandwidth resources can be caused; Postpone a small amount of a few frame to come close to real time communication when not affecting vision, but almost do not have correlation technique to go to pay close attention to the selection of initial frame per second from the angle of real-time Transmission at present.

Summary of the invention

The object of the present invention is to provide the parameter adaptive multidimensional bit rate control method based on video content features, to solve the problem.

The update cycle of model parameter makes corresponding adjustment according to video Space-time Complexity, is taking sliding window as elementary cell when extracting Space-time Complexity; The parameter adaptive multidimensional bit rate control method of the video content features that this invention is proposed by the invention has higher accuracy, as shown in Fig. 3,4,5,6, be better than intra-class correlation method in the world, can effectively improve subjective quality and objective quality, especially for the higher video sequence of Space-time Complexity; In addition, compared to intra-class correlation method in the world, the present invention also can reduce computation complexity a little.

The method mainly comprises the work of two parts: the first, for ensureing that initial parameter can show good performance for different video sequences; According to initial bandwidth sum initial video content characteristic, the initial code frame per second of dynamic calculation optimum, the initial frame rate value calculated meets the genuine property of sequence more; Second, for different video sequences, there is different Time and place correlations, the different section of identical video sequence has different Space-time Complexities, dynamically update the multidimensional code check Controlling model parameter update cycle based on video content, adopt the complexity forecast model of spatial temporal adaptive to go to obtain the model parameter update cycle; In this process, take sliding window as base unit, extract video content features, predict next update cycle complexity, in conjunction with the accuracy of current update cycle Rate Control, obtain next update cycle length, and then determine the model parameter of next update cycle and quantization parameter and coding frame per second.

Embodiments provide a kind of parameter adaptive multidimensional bit rate control method based on video content features, comprise the steps:

Step (1), input video sequence, obtain initial bandwidth, judges whether it is initial frame;

Step (2) is if initial frame, and the video content features according to initial two frames of initial bandwidth sum calculates initial code frame per second, and then obtains initial code quantization parameter according to initial frame per second;

Step (3), if non-initial frame, to the coding video frames in the current update cycle, simultaneously using the update cycle as unit, and extracts associated video content information, calculates multidimensional code check Controlling model parameter; Make unit with sliding window, extract the content information of video in time-domain and spatial domain, and in conjunction with the accuracy of current update cycle Rate Control, calculate the length of next update cycle;

Step (4), after getting multidimensional code check Controlling model parameter, ask for optimized parameter according to model parameter, the present invention selects correlation model to remove to calculate frame per second f and the initial quantization parameters q of next update cycle.

In certain embodiments, be preferably, in described step (2), specifically comprise the steps:

Step (21), basis calculate preliminary initial frame per second to constant bit rate and video resolution;

Step (22), according to the video content information of front cross frame, the preliminary initial frame per second got to be adjusted, obtain optimum initial frame per second;

Step (23), initial frame per second according to optimum, obtain initial quantization parameter.

In certain embodiments, be preferably, in described step (22), specifically comprise the steps:

Step (221), extract the first frame entropy as initial airspace complexity;

Step (222), extract the first frame and the second frame frame difference as initial time domain complexity, simultaneously by initial time domain complexity and initial airspace complexity normalized;

Step (223), according to the first frame entropy and the first frame and the second frame frame difference, the preliminary initial frame per second got to be adjusted.

In certain embodiments, be preferably, in described step (3), specifically comprise the steps:

Step (31), be elementary cell with update cycle, extract frame difference FD, the MVM mean value u of motion-vector magnitude of consecutive frame _mVM, motion-vector magnitude u _mVMwith motion vector angle u _mDAratio η (μ _mVM,μ _mDA), ask for model parameter by forecast model P=HF, wherein P=[a, b, c, d, R _max] be each model parameter, F is for extracting video content, and H is prediction matrix, fixed value;

Step (32), be elementary cell with sliding window, obtain the complexity forecast model of spatial temporal adaptive, sliding window is made up of last continuous a few frame of current update cycle, each update cycle terminates rear sliding window and moves backward, and in described sliding window, video sequence is divided into according to time-domain and spatial domain is the ensemble space of two kinds of different scales;

Step (33), after getting the complexity forecast model of spatial temporal adaptive, ask for the length of next update cycle, the length of next update cycle adjusts according to code check precise control in the current update cycle, adaptively to increase the adaptability of forecast model simultaneously.

In certain embodiments, be preferably, in described step (32), specifically comprise the steps:

Step (321), according to spatial domain, then the entropy of each frame is asked for the image sequence in input sliding window, comprise first frame of next update cycle, get weighted entropy model;

Step (322), according to time-domain, then frame difference is asked for the image consecutive frame in the image sequence in input sliding window, get weighting frame poor, with the description of weighting frame difference as time domain complexity, with the description of the airspace complexity of weighted entropy model prediction, obtained the complexity forecast model of spatial temporal adaptive by weighting frame difference and weighted entropy model.

In certain embodiments, be preferably, in described step (4), according to the model parameter of step (3) gained, remove with Q-R-TQ model the frame per second f and the initial quantization parameters q that obtain next update cycle.

The parameter adaptive multidimensional bit rate control method based on video content features that the embodiment of the present invention provides, compared with prior art, on the one hand, in prior art, the acquisition of initial code frame per second all adopts empirical value, the present invention obtains initial frame per second according to the Space-time Complexity of given given initial bit rate, video resolution and initial two frames, and then obtain initial quantization parameter, ensure and ensure that initial parameter can show good performance for different video sequences.On the other hand, the model parameter update cycle of prior art takes fixed value or remains unchanged in an encoding process, the present invention dynamically updates the update cycle of model parameter in an encoding process according to video content, from the current update cycle, be that unit extracts video content features with sliding window, go out subsequent time update cycle length according to Space-time Complexity model prediction; Experiment shows, the parameter adaptive multidimensional bit rate control method of the video content features that the method proposes can effectively improve subjective quality and objective quality, especially for the higher video sequence of Space-time Complexity, in addition, the present invention also can reduce the complexity of calculating a little.

Accompanying drawing explanation

Fig. 1 is the frame diagram based on the parameter adaptive multidimensional bit rate control method of video content features in one embodiment of the invention;

Fig. 2 is sliding window schematic diagram in one embodiment of the invention;

Fig. 3 is that the method model parameter that the present invention proposes accurately is schemed, model coefficient a, PC=0.9839, RMSE=0.02646;

Fig. 4 is that the method model parameter that the present invention proposes accurately is schemed, model coefficient b, PC=0.99997, RMSE=0.004552;

Fig. 5 is that the method model parameter that the present invention proposes accurately is schemed, model coefficient c, PC=0.9826, RMSE=0.006722;

Fig. 6 is that the method model parameter that the present invention proposes accurately is schemed, model coefficient d, PC=0.9493, RMSE=0.6202;

Fig. 7 is optimum frame per second and bpp relation in one embodiment of the invention;

Fig. 8 is optimum frame per second and initial video content relation in one embodiment of the invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.

The present invention proposes a kind of parameter adaptive multidimensional bit rate control method based on video content features, method flow as shown in Figure 1, mainly comprises following four parts.

Step 1, input video sequence, obtain initial bandwidth, judges whether it is initial frame;

Step 2 is if initial frame, and the entropy extracting the first frame is as initial airspace complexity, and the frame difference extracting the first frame and the second frame, as initial time domain complexity, obtains initial frame per second, and then obtains initial quantization parameters according to initial frame per second;

Step 3, if non-initial frame, asks for length and next corrigendum modulus of periodicity shape parameter of next update cycle; To the coding video frames in the current update cycle, and extract associated video content information, according to extraction video content Confirming model parameter; The associated video content information extracted in sliding window, in conjunction with the accuracy of current update cycle Rate Control, determines the length of next update cycle;

Step 4, ask for optimized parameter according to model parameter.After getting model parameter, correlation model is selected to go to calculate optimum frame per second and quantization parameter.

This method effectively can improve subjective quality and the objective quality of encoded video, is particularly useful for video and changes violent sequence.Launch below to illustrate.

For step 1, first, initial bandwidth is obtained to the image sequence of input, in addition, after coding video frames terminates within each update cycle, need again obtain corresponding bandwidth.

For step 2, optimum initial parameter divides two parts, and namely first gets preliminary initial frame per second, then obtains final frame per second according to content characteristic; During Computed-torque control complexity, the entropy extracting the first frame, as initial airspace complexity, extracts the difference of the first frame and the second frame as initial time domain complexity; Optimum initial frame per second is as the function of initial bit rate and image complexity

FR _initial＝F(bpp)+G(Si _1,1,Ti _1,2) (1)

Wherein Si _1,1refer to the entropy of the first frame, Ti _1,2the frame referring to the first frame and the second frame is poor.Entropy is defined as:

N _lrepresent the length of l update cycle, EI _l,krepresent the entropy of the k frame in the l update cycle, can obtain by following formula:

${\mathrm{EI}}_{l,k}=-\underset{x=0}{\overset{I-1}{\mathrm{\Σ}}}{p}_{l,k}\left(x\right){\log }_2\left[p_{l,k}\left(x\right)\right](k=\mathrm{1,2}...N_l)---\left(2\right)$

Wherein, L is maximum grey level, and for 8 bit pixel precision, the value of I is 256.P (x) is the probability of occurrence of grey level x;

The definition of frame difference:

Use F _l,krepresent the gray value of the k frame in the l update cycle.Frame difference TI _l,kbe expressed as:

${\mathrm{TI}}_{l,k}=F_{l,k}-F_{l,k-1}=\frac{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}|F_{l,k}(i,j)-F_{l,k-1}(i,j)|}{M\×N\×255}(k=\mathrm{2,3}..N_l)---\left(3\right)$

Wherein M and N represents line number and the columns of frame of video respectively;

Specifically comprise the steps:

In coarse calculation stages:

Step (21): according to given target bit rate, by formula calculate the value of bpp;

Step (22): the bpp value obtained in step (1) is substituted into formula

$F\left(\mathrm{bpp}\right)=\left\{\begin{array}{cc}{a}_{11}\×\mathrm{bpp}+a_{12}& \mathrm{bpp}\≥0.6\\ a_{21}\×\mathrm{bpp}+a_{22}& 0.2>\mathrm{bpp}>0.6\\ a_{31}\×\mathrm{bpp}+a_{32}& \mathrm{bpp}\≤0.2\end{array}\right.---\left(4\right)$

Obtain the value of F (bpp), wherein ε ₁={ a ₁₁, a ₁₂, a ₂₁, a ₂₂, a ₃₁, a ₃₂be coefficient correlation, {-8.254,13.88 ,-18.54,19.74 ,-51.66,26.99} have higher accuracy to value, as shown in Figure 7;

In the Exact calculation stage:

Step (23): for real-time application, calculates the entropy Si of the first frame _1,1, calculate the frame difference Ti of the first frame and the second frame simultaneously _1,1;

Calculate the trim values of initial frame per second further wherein { m, n, h, z} are coefficient correlation, and value is that {-3 ,-5,75,3} have higher accuracy, as shown in Figure 8;

Step (24): step (22) and the result of calculation of step (23) are substituted into formula FR _initial=F (bpp)+G (Si _1,1, Ti _1,2) in, obtain optimum FR _initialvalue.

For step 3, specifically comprise the steps:

Step (31) is the video content information that elementary cell extracts encoded frame with update cycle, asks for model parameter;

Frame difference FD, (MVM) mean value u of motion-vector magnitude of consecutive frame is extracted within the update cycle _mVM, motion-vector magnitude u _mVMwith motion vector angle (u _mDA) ratio η (μ _mVM, μ _mDA), η (μ _mVM, μ _mDA)=μ _mVM/ μ _mDA;

Model parameter is asked for, wherein P=[a, b, c, d, R by forecast model P=HF _max] be each model parameter, F is for extracting video content, and H is prediction matrix, and value is

$P=\left[\begin{array}{cccc}1.1406& -0.0330& -0.0611& 0.1408\\ 0.4462& 0.0112& 0.0680& -0.0667\\ 0.1416& -0.0008& -0.0001& -0.0036\\ \text{8.9757}& -0.5728& -0.8516& 2.0528\\ 67.73& 49.45& 281.7& -245.6\end{array}\right]$

Step (32) take sliding window as base unit, extracts Space-time Complexity, according to complexity forecast model, asks for the length of next update cycle in conjunction with current update cycle Rate Control accuracy simultaneously;

According to the length of next update cycle of acquisition of current update cycle Space-time Complexity forecast model and current update cycle Rate Control

N _l＝N _l(Si,Ti,φ,N _max,N _min) (5)

Si refers to time domain complexity, and Ti refers to airspace complexity, and φ refers to the accuracy of Rate Control, N _maxthe maximum length of update cycle, N _minit is the minimum update cycle.

The calculating of A, airspace complexity:

By frame difference information as a kind of Complexity Measurement time; The elementary cell that sliding window is predicted as Space-time Complexity, as shown in Figure 2.Sliding window refers to last continuous a few frame of current update cycle; Each update cycle terminates rear sliding window and moves backward, as follows in sliding window frame difference gradient mean value calculation:

${\mathrm{Ti}}_l^{\mathrm{Avg}}=\frac{\underset{i=0}{\overset{d}{\mathrm{\Σ}}}{\mathrm{Ti}}_{n,N_l-i}}{d}---\left(6\right)$

Wherein d represents the size of sliding window, and because sliding window is larger, accuracy is higher, but consuming time higher, and compromise consuming time and accuracy, d value is 4.

With final airspace complexity as prediction:

$N_l^{\mathrm{Ti}}=\frac{e^{{\mathrm{Ti}}_l^{\mathrm{Avg}}}-(1-{\mathrm{Ti}}_l^{\mathrm{Avg}})}{e}---\left(7\right)$

When sequence motion is very fast or frame poor larger time, now spatial coherence is higher, and corresponding value is less; Otherwise it is contrary.

The calculating of B, time domain complexity;

With the spatial information of weighted entropy model prediction, this equation is as follows:

${\mathrm{Si}}_l^{\mathrm{Avg}}=\left\{\begin{array}{cc}\frac{\underset{i=0}{\overset{d}{\mathrm{\Σ}}}{\mathrm{Si}}_{l,N_l-i}+{\mathrm{Si}}_{l+\mathrm{1,1}}}{d+1},& n>1\\ \frac{\underset{i=0}{\overset{d}{\mathrm{\Σ}}}{\mathrm{Si}}_{l,N_l-i}}{d}& n=1\end{array}\right.---\left(8\right)$

With final airspace complexity as prediction:

$N_l^{\mathrm{Si}}=\frac{e^{{\mathrm{Si}}_l^{\mathrm{Avg}}}-(1-{\mathrm{Si}}_l^{\mathrm{Avg}})}{e}---\left(9\right)$

C, adjust the length of next update cycle in conjunction with update cycle Rate Control accuracy;

If the code check of actual consumption and there is larger difference to constant bit rate, show that scene changes is relatively violent, shorten the change that the update cycle adapts to scene;

The length of next update cycle adjusts adaptively according to code check precise control in the current update cycle, and increase the adaptability of forecast model, the judgement of accuracy is as follows

$\mathrm{\φ}\left(l\right)=\min (\frac{B_l}{O_l},\frac{O_l}{B_l})---\left(10\right)$

Wherein O _lit is the preallocated code check of l update cycle; B _lit is the code check of actual consumption in l update cycle;

Steps A, B, C result are substituted into N _l=N _l(Si, Ti, φ, N _max, N _min), obtain

N _l＝N _l(Si，Ti，φ，N _max，N _min)

(11)

＝N _max-N _l(Si)N _l(Ti)N _l(φ)×(N _max-N _min)

The present invention takes full advantage of Time and place correlation, by the normalized of time prediction model and spatial prediction model, obtain the complexity forecast model of spatial temporal adaptive, obtain the length of next update cycle according to the accuracy of complexity forecast model and current update cycle Rate Control; In addition the complexity forecast model of the spatial temporal adaptive of the present invention's proposition is limited in sliding window, and thus computation complexity is very little.

For step 4, specifically comprise the steps:

Because Q-R-TQ has higher accuracy, after getting model parameter, the present invention selects Q-R-TQ to go to calculate optimum frame per second and quantization parameter

Q-R-TQ code check model

$R(q,t)=R_{\max }{\left(\frac{q}{q_{\min }}\right)}^{-a}{\left(\frac{t}{t_{\max }}\right)}^{-b}---\left(12\right)$

Wherein R _maxbe when q get minimum value, t gets maximum time actual bit rate, a, b are model parameters, R _maxcan measure;

Q-R-TQ quality model

$Q(q,t)=\frac{e^{-c\frac{q}{q_{\min }}}}{e^{-c}}\frac{1-e^{-d\frac{t}{t_{\max }}}}{1-e^{-d}}---\left(13\right)$

Wherein c, d are model parameter;

Optimized parameter (f ^*, q ^*) obtain by such as under type

$\begin{array}{c}{f}^{*},q^{*}=\underset{q\&Element;[q_{\min },51],f\&Element;[0,f_{\max }]}{\arg \min }\left\{Q(q,t)\right\}\\ s.t.R(q,t)<R_{\max }\end{array}---\left(14\right)$

In the forecast model of spatial temporal adaptive of the present invention, mostly adopt plus and minus calculation to realize operation, thus the amount of calculation of algorithm is very little; Adopt multidimensional code check Rate Control structure of the present invention and method, for the application of extensive dynamic network and multi-source heterogeneous network, provide very flexible and good rate control techniques, can subjective quality be improved; And reduce the complexity of calculating.

These are only the preferred embodiments of the present invention, be not limited to the present invention, for a person skilled in the art, the present invention can have various modifications and variations.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (6)

1., based on a parameter adaptive multidimensional bit rate control method for video content features, it is characterized in that comprising the steps:

Step (1), input video sequence, obtain initial bandwidth, judges whether it is initial frame;

Step (2) is if initial frame, and the video content features according to initial two frames of initial bandwidth sum calculates initial code frame per second, and then obtains initial code quantization parameter according to initial frame per second;

Step (3), if non-initial frame, to the coding video frames in the current update cycle, simultaneously using the update cycle as unit, and extracts associated video content information, calculates multidimensional code check Controlling model parameter; Make unit with sliding window, extract the content information of video in time-domain and spatial domain, and in conjunction with the accuracy of current update cycle Rate Control, calculate the length of next update cycle;

Step (4), after getting multidimensional code check Controlling model parameter, ask for optimized parameter according to model parameter, the present invention selects correlation model to remove to calculate frame per second f and the initial quantization parameters q of next update cycle.

2. the parameter adaptive multidimensional bit rate control method based on video content features according to claim 1, is characterized in that in described step (2), specifically comprises the steps:

Step (21), basis calculate preliminary initial frame per second to constant bit rate and video resolution;

Step (22), according to the video content information of front cross frame, the preliminary initial frame per second got to be adjusted, obtain optimum initial frame per second;

Step (23), initial frame per second according to optimum, obtain initial quantization parameter.

3. the parameter adaptive multidimensional bit rate control method based on video content features according to claim 2, is characterized in that in described step (22), specifically comprises the steps:

Step (221), extract the first frame entropy as initial airspace complexity;

Step (222), extract the first frame and the second frame frame difference as initial time domain complexity, simultaneously by initial time domain complexity and initial airspace complexity normalized;

Step (223), according to the first frame entropy and the first frame and the second frame frame difference, the preliminary initial frame per second got to be adjusted.

4. the parameter adaptive multidimensional bit rate control method based on video content features according to claim 1, is characterized in that in described step (3), specifically comprises the steps:

Step (31), be elementary cell with update cycle, extract frame difference FD, the MVM mean value u of motion-vector magnitude of consecutive frame _mVM, motion-vector magnitude u _mVMwith motion vector angle u _mDAratio η (μ _mVM,μ _mDA), ask for model parameter by forecast model P=HF, wherein P=[a, b, c, d, R _max] be each model parameter, F is for extracting video content, and H is prediction matrix, fixed value;

Step (32), be elementary cell with sliding window, obtain the complexity forecast model of spatial temporal adaptive, sliding window is made up of last continuous a few frame of current update cycle, each update cycle terminates rear sliding window and moves backward, and in described sliding window, video sequence is divided into according to time-domain and spatial domain is the ensemble space of two kinds of different scales;

Step (33), after getting the complexity forecast model of spatial temporal adaptive, ask for the length of next update cycle, the length of next update cycle adjusts according to code check precise control in the current update cycle, adaptively to increase the adaptability of forecast model simultaneously.

5. the parameter adaptive multidimensional bit rate control method based on video content features according to claim 4, is characterized in that in described step (32), specifically comprises the steps:

Step (321), according to spatial domain, then the entropy of each frame is asked for the image sequence in input sliding window, comprise first frame of next update cycle, get weighted entropy model;

Step (322), according to time-domain, then frame difference is asked for the image consecutive frame in the image sequence in input sliding window, get weighting frame poor, with the description of weighting frame difference as time domain complexity, with the description of the airspace complexity of weighted entropy model prediction, obtained the complexity forecast model of spatial temporal adaptive by weighting frame difference and weighted entropy model.

6. the parameter adaptive multidimensional bit rate control method based on video content features according to claim 1, it is characterized in that in described step (4), according to the model parameter of step (3) gained, remove with Q-R-TQ model the frame per second f and the initial quantization parameters q that obtain next update cycle.