CN101841705A - Video lossless compression method based on adaptive template - Google Patents

Abstract

The invention discloses a video lossless compression method based on adaptive model selection. The method comprises the following steps: judging the type of an input video, determining operand, performing video pre-processing operation; adopting the airspace redundancy-removing method, the time domain redundancy-removing method and the direct mode redundancy-removing method to perform redundancy-removing operation to the current sample and obtain the predictive value of the current sample; utilizing the redundant information of space, time and frequency domain to separately perform infraframe prediction and interframe prediction to the original YUV sequence; if the original sequence is RGB sequence, performing infraframe prediction, interframe prediction and direct prediction to the converted sequence; using an adaptive prediction model selector to judge the predictive value of the current sample and determine that the minimum is the optimal prediction error; and finally performing context-based arithmetic encoding to the map value of the final prediction error and outputting to obtain bit stream. The invention uses the adaptive prediction model selector to replace the additional bit prediction model, thus obviously increasing the compression efficiency of video; and the method can be used for video compression in the fields of aviation and navigation.

Description

Video lossless compression method based on adaptive template

Technical field

The invention belongs to the digital image processing techniques field, relate to video lossless compression method, this method can be used for the higher aviation of resolution requirement, navigation field.

Background technology

In recent years, digital picture and video compression have had very big development, MPEG-4, JPEG2000 have successively occurred, have H.264/MPEG-4AVC waited advanced compression standard.These standard great majority are conceived on the lossy compression method, but in a lot of practical applications, harmless digital picture and video compression but seem more important.For example, must compress to save the storage area, improve propagation efficiency the aviation of magnanimity and remote sensing images, if but use lossy compression method, the image that has been compressed detailed information will be omitted important target information, thereby causes error.Therefore, the research of harmless digital picture and video compression has progressively been obtained scholars' attention.

The redundant information the original video sequence is mainly removed in the compression of digital video from three aspects: colourity, room and time.The colourity redundancy generally removes by the conversion of rgb space to the YcbCr space.Spatial redundancy mainly is that this point is particularly evident in the natural image of Continuous Gray Scale owing to the correlation between the pixel in the frame.Remove the existing a lot of methods of spatial redundancy, some method has been applied in the compression of lossless image, as LOCO-I etc.Time redundancy mainly is because the time is gone up the correlation between the consecutive frame.In some video compression algorithms that diminish such as MPEG1, MPEG2, mainly be to remove temporal correlation effectively by inter-frame prediction method.This correlation does not exist only between the consecutive frame, is present between the frame close relatively on the time yet.At the present Research of present video lossless compression, press for the correlation technique of lossless video compression field is done further deep research.

Harmless video compression is similar with the video compression standard that diminishes, and number of modules has adopted new coding techniques.Wherein some algorithm is to the expansion of algorithm in the past, as supports new block matching algorithm, the more estimation of multi-reference frame and motion compensation etc.; Some then is to the improvement of algorithm in the past, reaches 1/4 pixel as the precision of estimation; Also having some then is algorithm in the complete standard different from the past, as the predictive coding of spatial domain, and integer wavelet transformation, based on contextual entropy coding etc.This shows that the raising of its code efficiency realizes by above all multimodes are common.

At present, the lossless video compression method mainly contains following several:

1.Memon a kind of incorporation time territory that proposed in 1996 Deng the people and the compression method of spatial domain, the author has also inquired into adaptive algorithm such as the document N.Memon and X.Wu. " Context-based; adaptive; lossless image coding " that is used for based on the optimum prediction of block motion compensation and frequency domain, IEEE Transactions onCommunications, 1997,45 (4), pp:437-444. is described.The CAL1C algorithm X Wu that people such as Wu proposed in 1998, W Choi, N Memon. " Lossless Interframe Image Compression via ContextModeling ", Proc of Data Compression Conf, pp:378-387,1998 and Elias carotti etc. behind the interframe neighborhood that proposed in 2002 in adaptive predictor and frame spatial domain prediction device Carotti E S G, DeMartin J C, Meo A R. " Low-complexity Iossless video coding via adaptive spatio-temporal predietionProeeedings ", 2003International Conference on Image Processing, Bareelona, 2003, pp:197-200.The compression ratio of these compression methods is unusual between 2-3 times according to video flowing.If JPEG-LS and CALIC are directly used in the compression of lossless video, do not consider video correlation in time like this, compression ratio is not high.Though the method for Wu has been considered the fusion of time and spatial prediction, do not adopt adaptive strategy, so compression ratio improves seldom.The method of Elias has been used new time domain prediction device, has reduced algorithm complex, but has also reduced the performance of prediction.Calendar year 2001, G.C.K. analyzed wavelet transformation to the later residual image of motion compensation, find that it can not reduce entropy effectively, also just can not effectively compress G CK Abhayaratne.D M Monco. " Embedded to Loeeless Coding of Compensated Prediction Residuals inLoseless Video Coding ", Proc of SPIE, 2001,4310, pp:75-85.

2003, combination such as open and propose a kind of self-adapting compressing algorithm Z.Ming-Feng based on the time domain prediction of block motion compensation and the spatial domain prediction method of in CALIC, describing, H.Jia, and Z.Li-Ming. " Lossless videocompression using comnination of temporal and spatial prediction ", in Proc.IEEE.Int.Conf.Neural Netwoks Signal Processing, Dec.2003, pp.1193-1196..Bu Luneiluo etc. have proposed a kind of new time domain prediction technology and optimum 3-D linear prediction algorithm D.Brunello based on the block motion compensation device, G.Calvagno, G.A.Mian, and R.Rinaldo. " Lossless compression of video using temporalinformation ", IEEE Transactions on Image Processing, 2003., 12 (2), pp:132-139.2004, palaces etc. have proposed the lossless video encoding algorithm based on wavelet transformation, the author proposes a kind of method Y.Gong that changes according to the motion vector situation of consecutive frame between two kinds of models, S.Pullalarevu, and S.Sheikh. " Awavelet-based lossless video coding scheme ", in Proc.Int.Conf.Signal Processing, 2004, pp:1123-1126.Thereafter the video sequence self-adaptive non-loss that proposed a kind of combined with wavelet transformed such as Park is pressed algorithm, obtain predicted value S.-G.Park by in reference frame, seeking the optimum Match target window, E.J.Delp, and H.Yu. " Adaptive lossless video compression using an integer wavelet transform ", in Proc.Int.Conf.Image Processing, 2004, pp:2251-2254..Though more than these proposed algorithm in recent years and were having some improvement aspect the lifting of performance, its algorithm complex is higher, and the selection of adaptive model is still very ineffective, the compression efficiency that causes video is not very high.

Summary of the invention

The objective of the invention is to overcome the deficiency of above-mentioned prior art, proposed a kind of adaptive video lossless compression method, further to improve lossless compress efficient by adaptive selection coding method video based on predictive coding.

The object of the present invention is achieved like this:

The present invention makes full use of the thought of block-based hybrid coding structural framing, utilizes the time, and the redundant information of space and frequency domain is carried out lossless compress to adaptive model to video by the back, reduces the transmission of boundary information simultaneously.Its

The specific implementation step is as follows:

(1) determines operation domain according to the type of different input videos,, and adopt the integer wavelet transformation method to carry out preliminary treatment if input video is that rgb format then is converted into yuv format; Then directly use spatial domain method to carry out preliminary treatment when being input as yuv format;

(2) go redundancy method or time domain to go redundancy method or Direct Model to go redundancy method that each pixel or wavelet coefficient are predicted to pretreated video data by the spatial domain;

(3) make up adaptive prediction Model Selection device e ^(S), e ^(T), e ^(N):

$e^{\left(S\right)}=\left|e_1(x,y-2)\right|+\left|e_1(x-1,y-1)\right|+\left|e_1(x,y-1)\right|+\left|e_1(x+1,y-1)\right|$

$+\left|e_1(x-21,y)\right|+\left|e_1(x-1,y)\right|+\left|e_1(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_1(x-1,y-2)|+|e_1(x+1,y-2)|+|e_1(x-2,y-1)|+|e_1(x-1,y+1)\left|\right)$

$e^{\left(T\right)}=\left|e_2(x,y-2)\right|+\left|e_2(x-1,y-1)\right|+\left|e_2(x,y-1)\right|+\left|e_2(x+1,y-1)\right|$

$+\left|e_2(x-21,y)\right|+\left|e_2(x-1,y)\right|+\left|e_2(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_2(x-1,y-2)|+|e_2(x+1,y-2)|+|e_2(x-2,y-1)|+|e_2(x-1,y+1)\left|\right)$

$e^{\left(N\right)}=\left|e_3(x,y-2)\right|+\left|e_3(x-1,y-1)\right|+\left|e_3(x,y-1)\right|+\left|e_3(x+1,y-1)\right|$

$+\left|e_3(x-21,y)\right|+\left|e_3(x-1,y)\right|+\left|e_3(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_3(x-1,y-2)|+|e_3(x+1,y-2)|+|e_3(x-2,y-1)|+|e_3(x-1,y+1)\left|\right)$

E wherein ₁The predicated error of removing redundancy method for the spatial domain,

The MED predicted value of expression sample;

e ₂The predicated error of removing redundancy method for time domain method,

Time domain prediction value for sample;

e ₃Go the predicated error of redundancy method for the Direct Model method;

Wavelet coefficient values for sample;

e ^(S), e ^(T), e ^(N)Represent current pixel or wavelet coefficient p in the step (2) respectively _i(x, e y) ₁, e ₂, e ₃Predicted value;

(4) adopt adaptive prediction Model Selection device to determine the minimum value that predicts the outcome, obtain prediction error value:

mod?e＝arg(min _{S，T，N}{e ^(S)，e ^(T)，e ^(N)})

Wherein distinguish S, T, N are spatial domain, time domain and direct predictive mode;

(5) adopt arithmetic coding method that the mapping value of predicated error is carried out entropy coding, finally obtain output bit flow.

The present invention has following effect:

1) because the present invention has adopted improved MED model, can make the predicted value in spatial domain more accurate preferably on the one hand, simultaneously obtain to a certain degree raising in the compression ratio of infra-frame prediction, MED model with respect to classics, improved MED fallout predictor can carry out more efficient compression to video sequence, improved pixel value prediction precision in frame, compression efficiency improves nearly 10%;

2) the present invention is directed to video at frequency domain, the characteristic of spatial domain and time domain, designed a kind of new adaptive prediction Model Selection device, this selector utilizes three kinds of different predictive modes that the prediction error value that obtains is compared, the model that obtains the difference minimum is optimum forecast model, can make the code word of transmission reach purpose minimum and the realization compression like this;

3) image compensation often all needs the translatory movement vector, promptly transmit boundary information, to be that data volume increases like this, compression efficiency is reduced, the present invention utilizes adaptive prediction Model Selection device, in conjunction with the residual error that forecast model produces, image is compressed, can avoid the transmitting moving vector, further reduce data volume.

Description of drawings

Fig. 1 is an adaptive video lossless compress flow chart of the present invention;

The current sample neighborhood schematic diagram that Fig. 2 makes up for the present invention;

The pixel samples time domain prediction device that Fig. 3 makes up for the present invention;

Fig. 4 is an adaptive prediction Model Selection device schematic diagram of the present invention;

Fig. 5 is interior and interframe context model figure for frame of the present invention;

Fig. 6 is the compression analogous diagram of the present invention to Akiyo (352*288) video sequence;

Fig. 7 is the compression analogous diagram of the present invention to News (720*480) video sequence.

Embodiment

Followingly describe the present invention with reference to accompanying drawing:

With reference to accompanying drawing 1, video coding step of the present invention is as follows:

Step 1:, carry out preliminary treatment according to the video input type.

If video is input as the RGB colour gamut, carry out following operation:

1-1) RGB is carried out color space conversion, its conversion formula is:

$\left\{\begin{array}{c}Y=0.299*R+0.587*G+0.114*B\\ U=-0.147*R-0.289*G+0.436*B\\ V=0.615*R-0.515*G-0.100*B\end{array}\right.$

Wherein R, G, B are respectively 3 color components of RGB colour gamut, and Y, U, V are respectively the color component of YUV colour gamut;

1-2) adopt the lifting formula of following tight support biorthogonal wavelet that the YUV sequence that obtains is carried out reversible integer (5,3) wavelet transformation:

Direct transform: $\left\{\begin{array}{c}{d}_{j-1,k}^i-=\frac{s_{j-1,k+1}^i+s_{j-1,k}^i}{2}\\ s_{j-1,k}^i+=\frac{d_{j-1,k}^i+d_{j-1,k-1}^i+2}{2}\end{array}\right.$

Inverse transformation: $\left\{\begin{array}{c}{s}_{j,2k}^i=s_{j-1,k}^i-\frac{d_{j-1,k-1}^i+d_{j-1,k}^i+2}{2}\\ d_{j,k-1}^i=d_{j-1,k}^i+\frac{s_{j,2k}^i+s_{j,2k+2}^{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000055956300031.png"\; state="\{\{state\}\}">i</figure-callout>}}}{2}\end{array}\right.$

D wherein, s is two mutually disjoint subclass that original signal sequence is divided into by even number and odd indexed, and i is the wavelet decomposition number of plies, and j, k are the coordinate vector of element;

If the YUV colour gamut in the video sequence because the integer wavelet transformation method is not suitable for the YUV sequence, thereby will not done any conversion in spatial domain to this YUV colour gamut.

Step 2:, adopt diverse ways to the sample p in the present frame according to redundancy properties different with interframe in frame _i(x y) removes redundant operation.

2-1) the redundancy method operation is gone in the spatial domain of all preliminary treatment samples: in order to reduce spatial redundancy, the present invention adopts a kind of improved spatial domain prediction device to current sample, and the expression formula of this fallout predictor is:

$\hat{x}=A+B-C$ Other

Wherein: p _i(x, the y) sample in the expression present frame, (x y) is the sample coordinate figure,

The spatial domain prediction value of expression sample, A, B, C, D represent p respectively _i(x, y) adjacent sample, as shown in Figure 2;

T ₁, T ₂Be the set threshold value of fallout predictor, threshold value T here ₁＞T ₂, and T ₁=15, T ₂=0.

Obtaining the spatial domain prediction errors table by the spatial domain prediction device is shown:

2-2) time domain of all samples is gone redundancy method operation: time domain go redundant target be in reference frame [i-1], seek and present frame in p _i(x, optimum Match sample y).

It is with p that time domain is removed redundancy method _i(x y) searches in the zone of the target window of frame [i] W*H in frame [i-1], and target window is by sample p _i(x, y) upper left adjacent sample constitutes, as shown in Figure 3.The order of search reaches minimum point for finding accumulation absolute difference CAD from left to right, from top to down.The accumulation absolute difference is expressed as:

$\mathrm{CAD}\left(T_w\right)=\underset{(m,n)\&Element;T_w}{\mathrm{\&Sigma;}}|p_i(x,y)-p_{i-1}(x+m,y+n)|$

Wherein: T _wThe target window of expression, p _i(x, y) and p _I-1(x, the y) sample value of expression present frame sample value and reference frame,

(m n) is motion vector, and (m n) is defined within the zone to this motion vector

In, so that the CAD value is minimum.

Defining the optimum movement vector that the target window of minimum CAD obtains is:

(m ₀，n ₀)＝arg _{m，n}{min?CAD(T _w)}

Obtaining the time domain prediction errors table by the time domain prediction device is shown: ${\mathrm{<figure-callout\; id="2"\; label="e"\; filenames="HSA00000055956300031.png"\; state="\{\{state\}\}">e</figure-callout>}}_2=p_i(x,y)-{\hat{p}}_i^T(x,y)$

${\hat{p}}_i^T(x,y)=p_{i-1}(x+m_0,y+n_0),$ Time domain prediction value for sample;

2-3) the Direct Model method of the sample of wavelet pretreatment is gone the redundancy method operation: to the video sequence of handling in wavelet field, because integer wavelet has the characteristic of concentration of energy, wavelet coefficient at high-frequency sub-band LH, HL, HH has smaller value usually, and directly the coding to this wavelet coefficient will further reduce the prediction redundancy.

Going redundancy to obtain predicated error by the Direct Model method can be expressed as ${\mathrm{<figure-callout\; id="3"\; label="e"\; filenames="HSA00000055956300031.png"\; state="\{\{state\}\}">e</figure-callout>}}_3=p_i(x,y)-{\hat{p}}_i^N(x,y),$

Wherein

For being the Direct Model predicted value of sample.

Step 3: make up a kind of new adaptive prediction Model Selection device as shown in Figure 4, according to the different predicated error (e of step 2 gained ₁, e ₂, e ₃), therefrom adaptive selection best model, the selection of adaptive predictor is to judge according to the predicated error sum of neighbor.

3-1) calculate spatial domain prediction error sum e ^(S):

$e^{\left(S\right)}=\left|e_1(x,y-2)\right|+\left|e_1(x-1,y-1)\right|+\left|e_1(x,y-1)\right|+\left|e_1(x+1,y-1)\right|$

$+\left|e_1(x-21,y)\right|+\left|e_1(x-1,y)\right|+\left|e_1(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_1(x-1,y-2)|+|e_1(x+1,y-2)|+|e_1(x-2,y-1)|+|e_1(x-1,y+1)\left|\right)$

3-2) calculate time domain prediction error sum e ^(T):

$e^{\left(T\right)}=\left|e_2(x,y-2)\right|+\left|e_2(x-1,y-1)\right|+\left|e_2(x,y-1)\right|+\left|e_2(x+1,y-1)\right|$

$+\left|e_2(x-21,y)\right|+\left|e_2(x-1,y)\right|+\left|e_2(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_2(x-1,y-2)|+|e_2(x+1,y-2)|+|e_2(x-2,y-1)|+|e_2(x-1,y+1)\left|\right)$

3-3) calculate Direct Model predicated error sum e ^(N):

$e^{\left(N\right)}=\left|e_3(x,y-2)\right|+\left|e_3(x-1,y-1)\right|+\left|e_3(x,y-1)\right|+\left|e_3(x+1,y-1)\right|$

$+\left|e_3(x-21,y)\right|+\left|e_3(x-1,y)\right|+\left|e_3(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_3(x-1,y-2)|+|e_3(x+1,y-2)|+|e_3(x-2,y-1)|+|e_3(x-1,y+1)\left|\right)$

Step 4: determine that according to step (3) final forecast model is:

mod?e＝arg(min _{S，T，D}{e ^(S)，e ^(T)，e ^(N)})，

With this final forecast model the predicated error of minimal mode is encoded.

Step 5: the corresponding sample p of final forecast model that step (4) is obtained _i(x, predicated error e y) _i(i=1～3) adopt context to carry out arithmetic coding.

At first, adopt two kinds of context template as shown in Figure 5 to make up p _i(wherein a kind of is to be used for frame mode for x, context y), and another is used for inter-frame mode.In frame mode, produce current sample p by 9 adjacent samples _i(x, context y); In inter-frame mode, current sample p _i(x, context y) is by p _i(x, the adjacent sample at frame [i] and frame [i-1] y) is formed jointly.

Context template can obtain by following formula:

C＝|c ₀|+|c ₁|+|c ₂|+|c ₃|+|c ₄|+|c ₅|+|c ₆|+|c ₇|+|c ₈|

C wherein _i(i=0～8) are illustrated in respective sample p among Fig. 5 _i(x, y) predicated error of Dui Ying final forecast model.

At last, use based on this contextual arithmetic coding predicated error is carried out entropy coding and output code flow.

The present invention carries out compression effects to video sequence and can further specify by experiment:

This experiment utilizes PC to realize the compression of image sequence under the VC++ programming.The video sequence of experiment employing standard detects algorithm, uses the RGB sequence of standard sequence Coastguard (176*144), Mohter (176*144), Foreman (176*144), Akiyo (352*288), Hall (352*288), Carphone (352*288), Football (720*480), News (720*480), Tennis (720*480) and Claire, Football, Moblie (360*288).

Fig. 6 and Fig. 7 have compared the simulation result of JPEG-LS algorithm, CALIC method, Park method and this method, the lossless compress efficient of method of the present invention as can be seen from Figure video in the figure of Akiyo and News remains on a relative higher level, this mainly be since under the little situation of image scene conversion aspect the inter prediction, effectively eliminated time redundancy.

Different video sequence used bit value of unit picture element when compression has been write down in this experiment simultaneously, and experimental result is with shown in the table 1.

Table 1 YUV﹠amp; RGB cycle tests unit: Bits/pixel (bpp)

From table 1, can significantly find out, the code check of the every frame of video has had tangible improvement, but for the Coastguard sequence, the efficient of video compression is improved to some extent, but it is obvious to be not so good as other sequences, this mainly is because the scene more complicated of video can not effectively be eliminated spatial redundancy and time redundancy in the video sequence.

Claims (4)

1. the video lossless compression method based on the adaptive model prediction comprises the steps:

(1) determines operation domain according to the type of different input videos,, and adopt the integer wavelet transformation method to carry out preliminary treatment if input video is that rgb format then is converted into yuv format; Then directly use spatial domain method to carry out preliminary treatment when being input as yuv format;

(2) go redundancy method or time domain to go redundancy method or Direct Model to go redundancy method that each pixel or wavelet coefficient are predicted to pretreated video data by the spatial domain;

(3) make up adaptive prediction Model Selection device e ^(S), e ^(T), e ^(N):

$e^{\left(S\right)}=\left|e_1(x,y-2)\right|+\left|e_1(x-1,y-1)\right|+\left|e_1(x,y-1)\right|+\left|e_1(x+1,y-1)\right|$

$+\left|e_1(x-21,y)\right|+\left|e_1(x-1,y)\right|+\left|e_1(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_1(x-1,y-2)|+|e_1(x+1,y-2)|+|e_1(x-2,y-1)|+|e_1(x-1,y+1)\left|\right)$

$e^{\left(T\right)}=\left|e_2(x,y-2)\right|+\left|e_2(x-1,y-1)\right|+\left|e_2(x,y-1)\right|+\left|e_2(x+1,y-1)\right|$

$+\left|e_2(x-21,y)\right|+\left|e_2(x-1,y)\right|+\left|e_2(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_2(x-1,y-2)|+|e_2(x+1,y-2)|+|e_2(x-2,y-1)|+|e_2(x-1,y+1)\left|\right)$

$e^{\left(N\right)}=\left|e_3(x,y-2)\right|+\left|e_3(x-1,y-1)\right|+\left|e_3(x,y-1)\right|+\left|e_3(x+1,y-1)\right|$

$+\left|e_3(x-21,y)\right|+\left|e_3(x-1,y)\right|+\left|e_3(x-1,y+1)\right|$

$+\frac{1}{2}\left(\right|e_3(x-1,y-2)|+|e_3(x+1,y-2)|+|e_3(x-2,y-1)|+|e_3(x-1,y+1)\left|\right)$

E wherein ₁The predicated error of removing redundancy method for the spatial domain,

The MED predicted value of expression sample;

e ₂The predicated error of removing redundancy method for time domain method,

Time domain prediction value for sample;

e ₃Go the predicated error of redundancy method for the Direct Model method;

Wavelet coefficient values for sample;

e ^(S), e ^(T), e ^(N)Represent current pixel or wavelet coefficient p in the step (2) respectively _i(x, e y) ₁, e ₂, e ₃The predicated error sum;

(4) adopt adaptive prediction Model Selection device to determine the minimum value that predicts the outcome, obtain prediction error value:

mode＝arg(min _{S，T，N}{e ^(S)，e ^(T)，e ^(N)})

Wherein distinguish S, T, N are spatial domain, time domain and direct predictive mode;

(5) adopt arithmetic coding method that the mapping value of predicated error is carried out entropy coding, finally obtain output bit flow.

2. the video lossless compression method based on the adaptive model prediction according to claim 1, wherein step (2) is described predicts pretreated video data, is the spatial redundancy amount of utilizing in the spatial domain in the improved MED spatial domain prediction device removal frame; In time domain, utilize the mode of asking minimum accumulation absolute difference to remove the correlation of interframe pixel; Utilize direct method to obtain wavelet coefficient in the high-frequency sub-band.

3. the video lossless compression method based on the adaptive model prediction according to claim 2, wherein said improved MED fallout predictor, representation is as follows:

$\hat{x}=A+B-C$ Other

Wherein The predicted value of expression sample, A, B, C, D represent the sample that x is adjacent, T respectively ₁, T ₂Be respectively set threshold value, and T ₁＞T ₂

4. the video lossless compression method based on the adaptive model prediction according to claim 2, the wherein said correlation of in time domain, utilizing the mode of asking minimum accumulation absolute difference to remove the interframe pixel, carry out as follows:

4a) this p of sampling in present frame _i(x, y) upper left adjacent sample is a target window;

4b) search for the accumulation absolute difference minimum value that obtains optimum Match (m in the region of search with this target window W*H in frame [i-1] as sample ₀, n ₀), the accumulation absolute difference is expressed as:

$\mathrm{CAD}\left(T_w\right)=\underset{(m,n)\&Element;T_w}{\mathrm{\&Sigma;}}|p_i(x,y)-p_{i-1}(x+m,y+n)|$

T wherein _wThe target window of expression, p _i(x y) is the present frame sample value, p _I-1(x y) is the sample value of reference frame, and (m n) is the interframe movement vector value.

Images