CN1281066C - Video frequency compression encoding method for partitional arithmetics of three dimentional hierarchical tree sets - Google Patents

Abstract

The present invention relates to a method that a three-dimensional wavelet transform grade tree set with a plurality of wavelet bases divides compressed coding of a time domain with alternate video frame groups and residual frame groups, and is an improvement on a three-dimensional wavelet transform compressed coding system of a video image. In consideration of the real time of video processing, the storage capacity of the system, the statistic characteristics of a coefficient after wavelet transform, and the characteristics of an SPIHT coding algorithm, the present invention provides a video stream image with more completed textures and not too intense movement, and further utilizes the time relativity of the video stream so that the video stream is divided into an original frame group (OFG) and a residual frame group (RFG), which serve as a coding unit for processing. Simultaneously, the time domain adopts a method that a Haar wavelet basis and a Daubechies 9/7 wavelet basis are combined. The present invention obviously enhances the performance of a compressed coding algorithm system divided by a three-dimensional wavelet transform grade tree set.

Description

A kind of video compressing and encoding method of 3 D wavelet hierarchical tree set partitioning algorithm

Technical field

The invention belongs to the video compression technology field, be specifically related to a kind of three-dimensional hierarchical tree set and divide (3D-DWT-SPIHT) video compressing and encoding method.

Technical background

Because wavelet transformation overcome under the low bit rate environment, based on the blocking artifact that minute block transform coding exposed and the shortcoming of non-mosquito effect, it has obtained significant progress on image processing in recent years.Fig. 2 (a) provides the one-level wavelet decomposition schematic diagram (totally 4 subband: LL1, LH1, HL1 and HH1) of two dimension, and LH1, HL1 and HH1 are the HFSs of first order conversion, and LL1 is the low frequency part that one-level is decomposed.Fig. 2 (b) is three grades of wavelet decomposition schematic diagrames (totally 10 subbands) of two dimension, and it is LL1 to be carried out wavelet decomposition produce LL2, LH2, HL2 and HH2, again LL2 is carried out wavelet decomposition and produces LL3, LH3, HL3 and HH3.Separable 3 D wavelet decomposition is the result at space two-dimensional binding time one-dimensional wavelet transform.

Through behind the wavelet transformation, the image transform domain coefficient has tangible statistics characteristics, and how organizing these wavelet conversion coefficients to quantize to encode with the back effectively is the key of image and video compression.Before EZW algorithm EZW (Embedded Zerotree Wavelet) [1] occurred, image compression particularly had the distortion compression, and its computational complexity and code efficiency are grown simultaneously.In 1993, the EZW algorithm that J.M.Shapiro proposed was broken through this restriction, it the coefficient of hypothesis on the thick yardstick always greater than thin its relevant position of yardstick under the prerequisite of coefficient, introduce the notion of zero tree, realized the high coding efficiency under the low-complexity.1996, the hierarchical tree set partitioning algorithm SPIHT (Set Partitioning in Hierarchical Trees) [2] that Amir Said and William A.Pearlman propose has improved EZW, even do not add the arithmetic coding, spiht algorithm still can be suitable with the performance of many complicated algorithms.The calendar year 2001 Yang Chun tinkling of pieces of jade and Yu Yinglin [3] are generalized to D S PIHT with SPIHT, it for simple background and motion comparatively slowly video flowing very good compression performance is arranged.

Why the SPIHT coding can be obtained preferably, and the compressed encoding effect is the statistical property that it has taken into full account wavelet coefficient: sky-frequency localization; Energy compaction property; The cluster characteristic of wavelet coefficient in the subband; The similitude of intersubband wavelet coefficient; The attenuation characteristic of wavelet coefficient amplitude from the low frequency sub-band to the high-frequency sub-band.The simple more image of texture, its energy concentrates on low frequency part more behind wavelet transformation, and high frequency just detail section energy is few more.Consider that SPIHT adopts embedded bit-planes quantization encoding, zero tree that the image that level and smooth more energy is low more produces when coding is many more, and then needed transmitted bit is few more, and compression ratio is high more.In order to reduce the energy of complicated image, the video flowing that frame of video group and residual frame group replace has been proposed.

Introduce some notions related to the present invention below:

1, separable 3 D wavelet transformation

Separable 3 D wavelet transformation is made up of two peacekeeping interframe one-dimensional transforms in the frame.It can be divided into two kinds on realizing, a kind of is to make the space two-dimensional wavelet transformation earlier, then the result after the conversion is made the one-dimensional wavelet transform of time shaft; Another kind method is made the time one-dimensional transform earlier just conversely, remakes two-dimensional transform in the frame.These two kinds of methods do not have the difference of essence, and transformation results is close.

2, spiht algorithm

1) formation of set membership and tree: set membership is defined as arrow direction (arrow points to children by father) among Fig. 3, with in scheming more arbitrarily (i j) is root, (and i, j) and all descendants just constitute one tree.(i, children's j) finding method are divided into two kinds of situations:

A, if (i, i) (n refers to highest decomposition stage at LLn, as the LL3 among Fig. 3), spiht algorithm is divided into LLn the piece of 2 * 2 sizes, in the piece point in the upper left corner do not have children (as among the LL3 among Fig. 3 by the point of stain mark), the children of other point are 2 * 2 the point (as one 2 * 2 the corresponding LH3 in the upper right corner, the corresponding HL3 in the lower left corner, the corresponding HH3 in the lower right corner of piece among the LL3 among Fig. 3) of relevant position in the current decomposition level.

B, if (i, i) in the subband beyond the LLn, then its children are following four points:

{(2i，2j)，(2i+1，2j)，(2i，2j+1)，(2i+1，2j+1)，}

2) set and division thereof

Fig. 4 is the schematic diagram of gathering in the spiht algorithm, and wherein each circle is represented a point, and solid expression point exists, and hollow representative point does not exist, and is father above the lines, following finger children.

T (i j) is a complete tree, C (i j) is root node, O (i j) is the children of root node, D (i j) is all descendants, and L (i ' j) be all descendants except that children.As can be seen from the figure

T(i，j)＝C(i，j)+D(i，j) (2)

D(i，j)＝O(i，j)+L(i，j) (3)

Set partitioning be exactly with D (i, j) be split into O (i, j) and L (i, j), or (i, (k, l) (wherein (k l) belongs to O (i, j)) to C j) to be split into four D with L.In spiht algorithm, (i j) is identified as the category-A set to D, and (i j) is identified as the category-B set to L.

3) threshold value and importance are judged

The initial threshold of spiht algorithm is calculated by formula (4), c in the formula _{I, j}(symbol  a  represents that logarithm a rounds, for example  6.87 =6 for i, wavelet coefficient j) to be illustrated in node.Threshold value Threshold dynamically changes.

Threshold＝2 ⁿ，n＝log ₂(max _(i，j){|c _i，j|})i，j (4)

A bit whether importantly calculate for certain by formula (5), important if the coefficient of present node greater than current threshold value, is then thought, otherwise think inessential.

$\mathrm{Sn}\left(C_{i,j}\right)=\left\{\begin{array}{cc}1,& \left|C_{i,j}\right|>\mathrm{Threshold}\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(5\right)$

Calculated by formula (6) for whether a set is important, meaning promptly has any node coefficient greater than current threshold value in the set, and then this set is important, otherwise inessential.

$\mathrm{Sn}\left(T\right)=\left\{\begin{array}{cc}1,& \underset{(i,j)\∈T}{\max }\left\{\right|C_{i,j}\left|\right\}>\mathrm{Threshold}\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(6\right)$

4) LSP, LIP and LIS

These three chained lists are work chained lists of using in the SPIHT encoding-decoding process.Significant coefficient chained list LSP (List ofsignificant pixels) deposits the coordinate of significant coefficient; Inessential coefficient chained list LIP (List of insignificant pixels) deposits the coordinate of inessential coefficient; Inessential set chained list LIS (List of insignificant sets) deposits unessential set (with the coordinate representation of root node).

5) SPIHT coding

The SPIHT coding is made up of three parts: initialization, sequencer procedure and thinning process.It is with the ordering of three chained lists realizations to wavelet coefficient information.

Initialization: calculate initial threshold; Put LSP for empty; LIP is initialized as the highest level low frequency part of wavelet decomposition (as the LL3 of three grades of wavelet decomposition in Fig. 3); LIS is initialized as the node (as lowest frequency subband among Fig. 3 is divided into 2 * 2 node unit, except that the node of beating asterisk, all the other 3 points have children) that children are arranged among the LIP, and it is made as the category-A set.

Ordering: algorithm at first scans LIP.Then LIS is scanned.Sequencer procedure is seen Fig. 5, and representation node coefficient (or set) is important and inessential respectively for two lines of the depth among Fig. 5, the bit that the bit on the line (0,1 and S) indicates to export.1. for each the node coefficient among the LIP, if inessential output 0, otherwise export 1 and S (symbol), then this node coordinate is moved to LSP table tail.2. for each set among the LIS, if important output 1 otherwise export 0.If important and for category-A set then by formula (2) division become children (if importantly then export 1 and S and add the LSP end to, otherwise export 0 and add the LIP end to) and a category-B gather (moving to the LIS end); If important and be category-B set, then it is split into four subclass (category-A set) and adds LIS to and show tail, delete current set then.

Refinement: after ordering is finished, need carry out thinning process,, do test of significance and output test result promptly to be sorted the coefficient on all nodes that add to the LSP except that current threshold value.Then threshold value is reduced by half, repeat ordering and thinning process, till output reaches needed output bit number.

List of references

[1]J.M.Shapiro，“Embedded image coding using zerotrees of waveletscoefficients，”IEEE Trans.Signal Processing，Vol.41，pp.3445-3462，Dec.1993.

[2]A.Said and W.A.Pearlman，“A New Fast and Efficient Image Codec Basedon Set Partitioning in Hierarchical Trees，”IEEE Trans.Circ.and Syst.forVideo Technology，Vol.6，pp.243-250，June 1996

[3] Yang Chun-Ling, YU Ying-Lin.Research on Embedded Video Compression Based on3D-Wavelet Transformation.Acta Electronica Sinica, 2001,29 (10): 1381-1383 (Yang Chunling, Yu Yinglin. based on the research of 3 D wavelet transformation embedded video compression algorithm. electronic letters, vol, 2001,29 (10): 1381-1383)

[4] Yang Chun-Ling, YU Ying-Lin.An improved wavelet video coder using three-dimensionalset partitioning in hierarchical trees.Acta Electronica Sinica, 2001,22 (5): 86-92 (Yang Chunling, Yu Yinglin. the coding method of the three-dimensional hierarchical tree set of improved wavelet field divided video. the communication journal, 2001,22 (5): 86-92)

[5] Zhang Zongping, Liu Guizhong, Hou Xingsong.Improved 3D-Wavelet Video Coder.Journalof Xi ' an Jiaotong University, 2001,35 (6): 595-599. (Zhang Zongping, Liu Guizhong, Hou Xingsong. a kind of improved 3 D wavelet video coding. XI AN JIAOTONG UNIVERSITY Subject Index, 2001,35 (6): 595-599).

[6]Kin BJ，Pearlman W A.An embedded wavelet video coder using three-dimensional setpartitioning in hierarchical trees(SPIHT)[J].Proc Data Compression Conf，1997：251-260.

[7] Ding Wenqi, the D S PIHT virtual tree method for organizing that time dimension is limited.(Chinese patent application number: 200410017386.3, the applying date is 2004.4.1)

Summary of the invention

The objective of the invention is to propose the higher video lossy compression method coding method of a kind of compression efficiency, this method makes under much compression and video quality condition internal memory more economize, and time of delay is shorter.

The 3D-DWT-SPIHT video lossy compression method coding method that the present invention proposes, it is a kind of compression method based on 3 D wavelet hierarchical tree set division (3D-DWT-SPIHT), it specifically is the temporal correlation that utilizes video flowing, video flowing is divided into primitive frame group (OFG) and residual frame group (RFG) handles, simultaneously the method that on time-domain, adopts Haar wavelet basis and Daubichies9/7 wavelet basis to combine as coding unit.The steps include:

1, time shaft multi-wavelet bases 3 D wavelet transformation

In existing wavelet compression method, all adopt at spatial domain and the good Daubechies 9/7 biorthogonal wavelet base of frequency domain slickness, in concrete operations in order to reduce boundary effect, must do symmetrical continuation at the room and time epigraph handles, and the filter order height of 9/7 biorthogonal wavelet base, for time-domain, if the frame number of image sets is T, then need be in the continuation of carrying out suitable 4 frames end to end of T two field picture group, difficulty can take place when decomposed class is high, during as T=16, doing 4 grades of time shafts decomposes, only surplus 4 frames behind the 3rd level, just can't carry out continuation, therefore in document [3～5] at 9/7 small echo of spatial domain with Daubechies, and all adopt the Harr wavelet basis as the conversion on the time shaft, but because of Harr wavelet basis rough in time-domain in time-domain, frequency domain unlimited makes along the wavelet decomposition of time shaft to reach best.

By analyzing the wavelet transformation on time-domain, can find that high frequency mainly appears at the place of interframe movement, and self-similarity mainly is on the background of not moving in a large amount of trees, therefore adopts two kinds of different wavelet basiss not influence its similitude, can improve the deficiency of Harr wavelet basis simultaneously.Among the present invention, in the interframe one-dimensional wavelet transform,, adopt the Daubechies9/7 wavelet basis when the number of image frames in the frame group reaches in 4 the time, when number of image frames in the frame group less than 4 the time, in the time of can not carrying out continuation, just adopt the Haar wavelet basis.Specifically, be exactly when T=16, with 9/7 wavelet basis of twice Daubechies, use the Harr wavelet basis then, can use once 9/7 wavelet basis of Daubechies to T=8.Experiment shows that this method can improve the Y-PSNR PSNR value of reconstructed image under certain compression ratio condition.In frame, in the two-dimensional transform (in the spatial domain), adopt Daubechies 9/7 wavelet basis.

As seen, time shaft multi-wavelet bases 3 D wavelet transformation is united realization by two peacekeeping interframe one dimension multi-wavelet bases conversion in the frame.

2, the method that replaces of frame of video group and residual frame group

From the characteristics of spiht algorithm itself, the progression of wavelet decomposition is high more, and the organizable node number of each tree root is many more, tree is also just big more, and code efficiency is high more, but because of reasons such as internal memory and delays, frame number in the 3 D wavelet processing unit can not be too many, generally all gets T=16.In document [4], I3D-SPIHT gets T=8 coding effect and increases.The present invention proposes and adopt T=8, but handle video flowing with the method that video and residual frame replace as the 3 D wavelet processing unit.If residual frame is the difference frame of present frame and reference frame, mainly reflect the information of two frame moving boundaries.Among the present invention, it is that the primitive frame group (OFG) and the residual frame group (RFG) of coding unit carried out compressed encoding that the video flowing that will handle is divided into 8 frames.The primitive frame group is exactly 8 successive image frames directly obtaining from video flowing.The residual frame group is to obtain on the basis of the adjacent primitive frame group in its front.The concrete practice is to do difference with the subsequent frame that its adjacent primitive frame group is finished 8 original in last reconstruction frames that obtains after coding and decoding and video flowing primitive frame groups to obtain.Under the little situation of video motion, necessarily contain much zero in the residual frame, just the energy of frame of video has obtained reducing the SPIHT coding that helps the back.(system such as Fig. 1 show)

Specifically describe as follows: a video flowing is divided into primitive frame group and the residual frame group that 8 frames are unit, and system block diagram is seen Fig. 1 (a).At first make 8 frame original images at transmitting terminal during beginning and (be expressed as I _i) 3D-DWT then carry out the 3D-SPIHT compressed encoding, simultaneously preceding 8 frames are recovered out the 8th frame  that keep to recover ₈, each frame of back and the 8th frame of recovery are done difference, and the residual frame that can obtain the 9th frame and the 8th frame respectively (is expressed as R _i), the residual frame of the 10th frame and the 8th frame, up to the residual frame that obtains back 8 frames, available formula is described as (1) by that analogy.

R _i＝I _i- ₈(i＝9，......，16) (1)

Then, residual frame being carried out 3D-DWT-SPIHT handles.Remake 8I+8R, till video flowing finishes (shown in Fig. 1 (b)) next time.The 8 two field picture frame groups that 16 two field pictures in the video flowing are divided into two characteristics are handled, and 16 frames among the present invention are dimerous by OFG and RFG, do not go not that 16 frames or 8 frames all are the situations of primitive frame in basic skills.

Advantage of the present invention:

The method of time shaft multi-wavelet bases has been carried out abundant decomposition, utilization to the frame of video group under the limited frame bar spare, has improved the accumulation of energy of the wavelet transformation of time-axis direction, and test shows that method [6] PSNR than independent employing Haar wavelet basis has improved 2-4dB.The resulting PSNR as a result of the method for alternate frame group obviously is better than directly carrying out more than the high 1dB of result of 3D-DWT-SPIHT processing with 8 frames, but the time delay of the two is identical; No matter the method for alternate frame group is in memory capacity, or the real-time aspect all is better than directly carrying out the three-dimensional compression with 16 frames, and the PSNR of its reconstructed image increases under identical compression ratio.Experimental result shows, is not influencing compression effectiveness, even under the condition that also PSNR of image is increased, the method for alternate frame group has effectively reduced time delay and needed memory cell.Consider the characteristics of video flowing and residual frame,, have similitude definitely between its frame and the frame, can remove time redundancy better, improve compression ratio with the residual error method according to changing slower video flowing.Export with bit plane because of the code stream of SPIHT simultaneously, be easy to control, the pro rate of the code stream of video group and residual frame group is different herein, in experiment we distribute 60% to video group 40% to the residual frame group, distribution that also can adaptive adjustment code stream is to obtain better PSNR.

Description of drawings

The method that Fig. 1, video and residual frame group replace.Wherein:

Fig. 1 (a) video and residual frame group method 3D-DWT-SPIHT flow chart.

The division of video flowing in Fig. 1 (b) video and the residual frame group method.

Fig. 2,2-d wavelet decomposing schematic representation [7].

Fig. 3, spiht algorithm set membership [7].

Fig. 4, set schematic diagram [7].

Fig. 5, ordering schematic diagram [7] (S represents symbol among the figure, 0 and 1 expression importance).

Embodiment

Salesman (352 * 288,30 frame per second) video flowing with public intermediate form CIF (Common Intermediate Format) is the concrete execution mode of example explanation below, and whole implementing procedure is consistent with Fig. 1.Encoding and decoding are an OFG or a RFG with 8 frames.

A: video flow processing

Input video stream is divided into the form of 8 frame primitive frames (OFG) and 8 frame residual frame (RFG).If the odd-multiple of 8 frames ((2n+1) * 8) promptly passes as the primitive frame group, then forward B to and handle.If the even-multiple of 8 frames (2n*8), then will be by the method (referring to summary of the invention 2) of alternate frame group, last frame after the decoding of the individual frame of video group of the 2nd (n-1) recovered be used as reference frame, subtract each other with each frame in 2n the frame of video group and this reference frame, obtain 8 continuous processing residual frame as the residual frame group, forward B then to.

B: wavelet transformation

Step 1. space field transformation

For each two field picture, it is made two-dimensional wavelet transformation with the Daubechies9/7 wavelet basis.Do 3 grades of decomposition altogether.

The conversion of step 2. time-domain

Each frame was made the wavelet coefficient of two-dimensional wavelet transformation, got the coefficient (totally 8 coefficients) of each frame relevant position, and carried out making 1 grade of wavelet decomposition behind the symmetric extension, got the low frequency sub-band after the decomposition, made 2 grades of Haar wavelet decomposition with Daubechies 9/7.

The D:SPIHT coding

Step 1. initialization

To all wavelet coefficient c _{I, j, k}, ask n= log ₂(max _{(i, j, k)}| c _{I, j, k}|) , and with 2 ⁿBe current threshold value Threshold, output n; LSP is initialized as sky; Add to the institute among the ROOT among the LIP a little; The tree set that with the node among the ROOT is tree root is added among the LIS.

Step 2. ordering and refinement

Step 2.1 for each node among the LIP (i, j, coefficient k):

Judge | c _{I, j, k}| whether greater than Threshold, and if greater than output " 1 " (important pixel), otherwise output " 0 " (inessential pixel).If important, judge c _{I, j, k}Whether greater than 0, if c _{I, j, k}Greater than 0 output " 1 " (positive sign) and with c _{I, j, k}Deduct Threshold, otherwise export " 0 " (negative sign) and add Threshold, (i, j k) move on to the LSP end with point then.

Step 2.2 for each set among the LIS (i, j k) (represent with tree root):

Judge (whether i, j have a coefficient greater than Threshold in k), if coefficient is arranged greater than Threshold then export " 1 " (important set), otherwise output " 0 " (inessential set) in set.

1) if gather important and be the category-A set, whether each children who judges tree root important (seeing step 2.1), if important then output " 1 ", otherwise would export " 0 ".For unessential children, it is moved to the LIP end.For important children, judge its coefficient symbols, if be positive sign then export " 1 ", and coefficient is deducted Threshold; Otherwise export " 0 ", and coefficient is added Threshold.Then important children are moved to the LSP end.If it is empty being removed the children's of root node set, then it is deleted from LIS, otherwise aggregate type changes the category-B type into, move to the LIS end then.

2) if gather important and be the category-B set, it is split into the root node children is the subclass of root, and setting its aggregate type is the category-A type, then these subclass has been added the LIS end, and deletes current set.

Step 3. quantizes.For the point among the LSP (i, j k) (add in current minor sort the point of LSP):

Judge | c _{I, j, k}| whether greater than Threshold, and if greater than output " 1 " (important pixel), otherwise output " 0 " (inessential pixel).

Step 4. is upgraded

Threshold＝Threshold/2。And turn to step 2.

E: compression ratio control and Data Rate Distribution

Repeat A and carry out compressed encoding to D.In coding, every output one bit all can carry out a compression ratio and judge, when reaching required compression ratio, can stop coding at once.

It is to be noted, the method of alternate frame group adopts 16 frames as a processing unit when being different from initial 3 D wavelet compressed encoding, just be that it divides into two parts with one 16 frame processing unit, OFG and RFG, be 8I+8R, be equivalent to handle the processing unit of two 8 frames during compressed encoding, so just effectively reduced time delay and saved amount of ram.Is that the 3 D wavelet compressed encoding of processing unit is that it utilizes video flowing interframe redundancy and the alternate frame group is better than basic with 8 frames, has improved the efficient of 3 D wavelet Coding Compression Algorithm.So because known compressed bit stream length, will consider the bit number that OFG and RFG respectively account in the alternate frame group of methods during specific coding, promptly will consider the Data Rate Distribution problem.Usually, the ratio height that the used bit number of OFG accounts in total code stream, the PSNR height of decoding compressed image then, compression ratio is low.Otherwise, the ratio height that RFG is shared, then the PSNR of decoding compressed image decreases, and big compression ratio improves.Through a large amount of experiments, in order under certain compression ratio condition, to obtain 16 frame mean P SNR preferably, we distribute 60% to video group 40% to the residual frame group.

0-47 frame with the Salesman sequence of CIF form carries out emulation experiment below.Table 1 first row is the method that adopts the time shaft multi-wavelet bases that primitive frame group and residual frame group replace, second row is that 3D-DWT-SPIHT directly adopts the simulation result of 8 frames as a codec unit, and the third line is the result that 3D-DWT-SPIHT directly adopts 16 frames.Under different compression ratios, the method PSNR of this paper (the peak value letter is made ratio) improves 0.39～1.65dB.

Method	PSNR			Amount of ram	Time delay
						CR＝24.369 7	CR＝52.792 4	CR＝136.62 77
	The method of alternate frame group	37.29	33.30						29.46	9 frames	8 frames
Directly use 8 frames				35.69	31.65	28.03	8 frames	8 frames
	Directly with 16 frames [3]	36.45	32.91						29.07	16 frames	16 frames

Claims (1)

1, a kind of 3 D wavelet hierarchical tree is gathered the video compressing and encoding method of partitioning algorithm, it is characterized in that utilizing the temporal correlation of video flowing, video flowing is divided into primitive frame group and residual frame group, as coding processing unit; The method that while adopts Haar wavelet basis and Daubechies 9/7 wavelet basis to combine on time-domain, concrete steps are as follows:

(1) time shaft multi-wavelet bases 3 D wavelet transformation is united by two peacekeeping interframe one dimension multi-wavelet bases conversion in the frame and is realized: in the interframe one-dimensional transform, when the two field picture number in the frame group greater than 4 the time, adopt Daubechies 9/7 wavelet basis, and when the two field picture number less than 4 the time, then adopt the filtering of Harr wavelet basis; In frame, in the two-dimensional transform, adopt Daubechies 9/7 small echo;

(2) method that replaces with frame of video group and residual frame group is handled video flowing: it is the primitive frame group and the residual frame group of coding unit that the video flowing that will handle is divided into 8 frames, carry out compressed encoding, wherein, the primitive frame group is 8 successive image frames directly obtaining from video flowing, and the residual frame group is to obtain on the basis of the adjacent primitive frame group in its front.

Images