CN101232625A - High efficient multidimensional video stream encoding and decoding method - Google Patents

Abstract

The invention relates to a high-efficiency multi-dimensional video stream coding and decoding method, which is mainly applicable to moving image sequence. In the premises of all-sidedly considering of redundancy information in multi-dimensional video stream signals, and in consideration of relevance and integrity of duration, space and color, as well as the premises of guaranteeing high-quality recovery of image signals, the invention can effectively resolve the problem of improving the compression rate for multi-dimensional video stream signals. The invention includes the following steps in details: Data extraction step; partitioning step; matrix conversion step; vector quantization step and entropy coding step. The core contents of the invention include introduction of definition of multi-dimensional vector matrix and relevant algorithm definition into the compression method for multi-dimensional video stream signals, and meanwhile introduce a corresponding multi-dimensional vector matrix conversion method; besides, the invention also raises a brand new multi-dimensional vector orthogonal matrix formula in discrete cosine conversion.

Description

A kind of multidimensional video stream encoding and decoding method efficiently

Technical field:

The present invention relates to the compression method of multidimensional video stream, be mainly used in the color motion sequence.

Background technology:

Along with the mankind enter the digital video epoch, NGN, 3G and 3G evolution and NGBW etc. are to the develop rapidly demand of video, multimedia service and network application, high efficiency viewing audio data compression coding technology as video traffic and storage application core technology, more and more cause people's attention and interest, formulate standard more and more efficiently.At present the source encoding standard that can select of video industry mainly contains: MPEG-2, MEPG-4AVC (be called for short AVC, also claim JVT), H.264, AVS etc.From maker's branch, first three standard is finished by MPEG expert group, and the 4th by the autonomous formulation of China.

MPEG-2

Moving Picture Experts Group-2 (ISO/IEC 13818) was in issue in 1994, and it is the compression scheme under various application and the specified in more detail of system layer at Standard Digital Television and high definition TV.The available transfer rate of MPEG-2 is 3 ~ 10Mbps.

The backward compatible MPEG-1 of Moving Picture Experts Group-2, it has done two important expansions on the basis of MPEG-1:

(1) supports the interlacing scan of TV, be provided with " framing code " and " by a coding " two kinds of patterns specially, adopt line by line the mode of piecemeal and interlacing piecemeal to carry out DCT respectively and encode.Prediction and double-basis prediction isotype between the field in motion compensation, have been added, to improve to object prediction accuracy and the raising compression ratio faster of moving.

(2) introduced " gradability " notion, gradable and signal to noise ratio is gradable based on room and time, realize the implements spatial scalable compression coding, and can realize backward compatible to the lower resolution image.

H.264 standard

The video encoding standard of new generation of joint video team (the JVT:Joint Video Team) exploitation of the MPEG (moving picture expert group) of the VCEG of ITU-T (video coding expert group) and ISO/IEC, ITU-T with its called after H.264, ISO is with the Part-10 of its called after MPEG-4.H.264 structure is with H.263 similar, but a lot of new coding techniquess have been increased, as infra-frame prediction, the inter prediction of variable size block, many prediction reference frame, the estimation of 1/4 pixel (luminance signal) and 1/8 pixel (carrier chrominance signal) precision, 4 * 4 integer transform, remove the blocking artifact filter, unified variable-length encoding UVLC (Universal VLC) is based on the variable-length encoding CAVLC (Context Adaptive VLC) of last h literary composition with based on contextual adaptive binary arithmetic coding CABAL (Context Adaptive Binary ArithmeticCoding) etc.The employing of these technology makes compression efficiency H.264 be significantly improved than compression standard before.

The AVS standard

The AVS standard is the abbreviation of " digital audio/video encoding and decoding technique standard " series standard, comprises support standards such as four main technical standards such as system, video, audio frequency, digital copyright management and uniformity test.Compare with other similar standards, AVS has two big advantages: based on the open standard of proprietary technology and part open technique structure, properly settle the patent grant problem.

Having distinctive core technology in the middle of the AVS video comprises: 8 * 8 integer transforms, quantification, infra-frame prediction, 1/4 precision pixels interpolation, special inter prediction motion compensation, two-dimensional entropy coding, loop filtering etc.

Because present international standard and other non-standard compression method are all pressed " field ", " frame " individual processing with coloured image, for moving image, also need increase various space-time integration technologies such as " motion compensation ".

The key issue of digital video is the coding techniques of vision signal.The amount of information that video comprised is maximum in all information of human perception, but is used to represent that the data volume of video also is very large, if do not carry out compressed encoding, all there is very big problem in its storage with Network Transmission.Therefore, technology of video compressing encoding is the permanent research focus of association area always.But compression performance has reached bottleneck at present, and is all very difficult even signal to noise ratio will improve 0.5dB.Press for the breakthrough of new compression coding technology.

Summary of the invention:

The object of the present invention is to provide a kind of multidimensional video stream encoding and decoding method efficiently, under the prerequisite of the high quality resume that guarantees picture signal, improved the compression ratio of image widely.

The present invention finishes as follows the multidimensional video stream signal is compressed:

The data extract step: the brightness of corresponding colour-video signal, color difference signal extract, and arrange and storage according to frame data.

The piecemeal step: according to input signal, (being generally yuv format) carries out piecemeal to each component respectively, on time shaft, selects 2 ^mFrame number, m gets positive integer, forms a cube piece, perhaps becomes the three-dimensional matrice piece.According to computation complexity and blocking artifact and with the compatibility of existing standard, be divided into 8 * 8 * 8 piece usually.

The matrixing step: utilization multidimensional vector orthogonal transform matrix, three-dimensional submatrix is carried out the multidimensional vector orthogonal transform, calculate three-dimensional coefficient matrix;

Vector quantization step: three-dimensional coefficient matrix is carried out vector quantization, calculate vector quantization system three-dimensional matrice;

Entropy coding step: vector quantization coefficient three-dimensional matrice is carried out entropy coding, obtain the change compression code matrix of digital color image signal three-dimensional matrice, finish compression to digital color picture signal.

Above-mentioned said multidimensional vector defined matrix and algorithm thereof are: the multidimensional data permutation table is called multi-dimensional matrix, and the dimension of multi-dimensional matrix is divided into two groups, use two vector representations respectively, be called the multidimensional vector matrix, and defined the equating of multidimensional vector matrix, addition, number take advantage of, computings such as matrix is taken advantage of, transposition, and defined unit multidimensional vector matrix.

The multidimensional vector matrixing formula that is adopted in the above-mentioned said matrixing step is:

Multi-dimensional matrix for input is divided into suitable multidimensional vector matrix as required, carries out conversion then:

The direct transform of multidimensional vector matrix

B _IJ＝C _IIA _IJC _JJ ^T

The conversion of multidimensional vector matrix inversion

A _IJ＝C _II ^TB _IJC _JJ

For motion image sequence, vector I is a two-dimensional vector, and J is the one dimension scalar, and matrix A is the 3-D view piece, and B is the intact three-dimensional matrice coefficient of conversion, and T is the multidimensional vector matrix transpose.

The multidimensional vector orthogonal matrix that is adopted in the above-mentioned said matrixing step is the multidimensional vector orthogonal matrix of discrete cosine transform, and its formula is:

The concrete form of wherein four-dimensional discrete cosine transform operator C is as follows:

C _IJ＝(c _uvmm)，I＝(M，N)，J＝(M，N)

Wherein $c_{\mathrm{uvmn}}=\sqrt{\frac{2^2}{\mathrm{MN}}}c\left(u\right)c\left(v\right)\cos \frac{(2m+1)\mathrm{u\π}}{2M}\cos \frac{(2n+1)\mathrm{v\π}}{2N}$

$c\left(u\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},u=0\\ 1,u=\mathrm{others}\end{array}\right.$ $c\left(v\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},v=0\\ 1,v=\mathrm{others}\end{array}\right.$ $c\left(s\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},s=0\\ 1,s=\mathrm{others}\end{array}\right.$ $c\left(t\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},t=0\\ 1,t=\mathrm{others}\end{array}\right.$

Vector I is a two-dimensional vector, C _IIBe the four-vector quadrature cosine matrix that the present invention proposes, J is the one dimension scalar, C _JJIt is the two-dimension discrete cosine transform matrix under the common meaning.

Technique effect of the present invention: the present invention has considered the redundant information of multidimensional video stream signal comprehensively, and considered the correlation and the globality of time, space and tone, thereby under the prerequisite that guarantees the signal high quality resume, improved the compression ratio of digital color image signal.

Description of drawings:

The image sequence method of partition of Fig. 1 indication of the present invention;

The flow chart of the multidimensional video stream compression method of Fig. 2 indication of the present invention.

Embodiment:

Further specify technology contents of the present invention and embodiment thereof below in conjunction with accompanying drawing.

Core content of the present invention is to have introduced the definition of multidimensional vector matrix and the definition of algorithm thereof in the multidimensional video stream compression method, and has introduced corresponding multidimensional vector matrixing, and has completely newly proposed the multidimensional vector orthogonal matrix of discrete cosine transform.And the vector quantization of matrix and entropy coding are prior art.

If no special declaration, this paper will represent real number field with R, represent complex field with C, represent quaternion field with H, F ∈ { R; C; H}.

The multi-dimensional matrix definition

M on the F * N data arrangement (a _I1i2) being called two-dimensional matrix, its all set of forming is designated as M _{M * N}, the I on the F ₁* I ₂* ... I _nMultidimensional data permutation table (a _{I1i2 ... in}) being called multi-dimensional matrix, its all set of forming is designated as M _{I1 * I2 * ... * In}

The definition of multidimensional submatrix

The same with the matrix under the common definition, to any one multi-dimensional matrix M _{I1 * I2 * ... * In}Can be divided into the littler multi-dimensional matrix of size with some planes, so little N dimension matrix becomes N dimension matrix M _{I1 * I2 * ... * In}N dimension submatrix.

The multidimensional vector defined matrix

If the dimension of multi-dimensional matrix is divided into two groups, uses two vector representations respectively, such as M _{K1 * K2 * ... * Kr}Be expressed as M _{(I1 * I2 * ... Im) * (J1 * J2 * ... * Jn)}, be designated as M _IJ, I wherein, J is a vector, I=(I ₁, I ₂..., I _m), J=(J ₁, J ₂..., J _n), then claim multi-dimensional matrix M be dimension according to vector I, the multidimensional vector matrix that J divides is called for short the multidimensional vector matrix.An obvious multi-dimensional matrix can be expressed as a variety of multidimensional vector matrixes, and the multi-dimensional matrix that the multidimensional vector matrix is only corresponding unique.

Below we define equating of multi-dimensional matrix, addition, the multiplication of multi-dimensional matrix and number, and corresponding computing character is discussed.

If 1, the equal multi-dimensional matrix of multi-dimensional matrix $A_{I_1\×I_2\×I_n}={\left(a_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n}$ With $B_{I_1\×I_2\×I_n}={\left(b_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n}$ All be I ₁* I ₂* ... I _nThe multi-dimensional matrix on rank, and their corresponding element is equal, promptly

$a_{i_1{i}_2\·\·\·i_n}=b_{i_1{i}_2\·\·\·i_n}$ (1≤i ₁≤I ₁；1≤i ₂≤I ₂；…；1≤i _n≤I _n)

We just say that multi-dimensional matrix A equates with B, and that note is A=B.

If I ₁=I ₂=...=I _n, multi-dimensional matrix A _{I1 * I2 * ... In}Be called the multidimensional square formation.Each element all is that zero multi-dimensional matrix is called zero multi-dimensional matrix, still represents with symbol 0.

For multi-dimensional matrix A _{I1 * I2 * ... In}, no matter which kind of mode to be divided into the multidimensional vector matrix according to, its equal criterion is constant.

2, the addition of multi-dimensional matrix is established $A_{I_1\×I_2\×I_n}={\left(a_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n}$ With $B_{I_1\×I_2\×I_n}={\left(b_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n}$ All be I ₁* I ₂* ... I _nThe multi-dimensional matrix on rank,, I then ₁* I ₂* ... I _nThe rank multi-dimensional matrix

$C_{I_1\×I_2\×\·\·\·I_n}=\left(c_{i_1{i}_2\·\·\·i_n}\right)={\left(a_{i_1{i}_2\·\·\·i_n}{+b}_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n},$

Be called A and B and, be designated as C=A+B.

The addition of multi-dimensional matrix is exactly the addition of multi-dimensional matrix corresponding element.Therefore the multi-dimensional matrix of addition must have identical dimension, and the exponent number of each dimension also must equate.

For multi-dimensional matrix A _{I1 * I2 * ... In}, no matter which kind of mode to be divided into the multidimensional vector matrix according to, its addition results is constant.

3, the number of multi-dimensional matrix is taken advantage of multi-dimensional matrix (m * a _{I1i2 ... in}) _{I1 * I2 * ... In}Be called multi-dimensional matrix $A_{I_1\×I_2\×I_n}={\left(a_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n}$ Product with real number m is designated as m ^A

In other words, counting m, to take advantage of multi-dimensional matrix be exactly that each element of multi-dimensional matrix all be multiply by m.

For multi-dimensional matrix and multidimensional vector matrix, the operation result of equate, addition, number being taken advantage of all is the same, does not distinguish the dividing condition of vector, and for following computing, then must in advance the dimension of multi-dimensional matrix be divided, define the dimension of vector.

4, multidimensional vector multiplication of matrices

If multidimensional vector matrix A _IJAnd B _UV, I=(I wherein ₁, I ₂..., I _m), J=(J ₁, J ₂..., J _n), U=(U ₁, U ₂..., U _S), V=(V ₁, V ₂..., V _T), and vector J=U, then claim the multidimensional vector matrix A _IJAnd B _UVHas the property of taking advantage of.

If A is I * L matrix, B is L * J matrix, and the product that defines matrix A and B so is an I * J matrix $C=\left(c_{i_1\·\·\·i_m{j}_1\·\·\·j_n}\right),$ Wherein

$c_{i_1\·\·\·i_m{j}_1\·\·\·j_n}=\underset{L}{\mathrm{\Σ}}{a}_{i1}{b}_{1j}=\underset{l_1}{\overset{L_1}{\mathrm{\Σ}}}\·\·\·\underset{l_k}{\overset{L_k}{\mathrm{\Σ}}}{a}_{i_1\·\·\·i_m{l}_1\·\·\·l_k}{b}_{l_1\·\·\·l_k{j}_1\·\·\·j_n}$

And this note is C=AB.

Simple for explaining, use in the following formula

Expression

Use a _I1Expression a _{I1 ... iml1 ... lk}, hereinafter all use this form of presentation if not otherwise specified.

5, unit multidimensional vector matrix

If multidimensional vector matrix A _IJ, I=(I wherein ₁, I ₂..., I _m), J=(J ₁, J ₂..., J _n), and vector I=J, then claim the multidimensional vector matrix A _IJIt is the multidimensional vector square formation.Introduce mark ${\mathrm{\δ}}_{\mathrm{ij}}=\left\{\begin{array}{cc}1& i=j\\ 0& i\≠j\end{array}\right.,$ Wherein i=j represents vector i=(i ₁..., i _m), j=(j ₁..., j _n) have an identical dimension, i.e. m=n, and i ₁=j ₁..., i _m=j _nIf, A _IJ=(δ _Ij), then claim A _IJBe unit multidimensional vector matrix, note is E _II, perhaps E be in brief note.

Obviously, unit multidimensional vector matrix is the multidimensional vector square formation certainly.

6, multidimensional vector transpose of a matrix operational criterion

A _IJ ^T＝A _JI

7, multidimensional vector matrixing

Multi-dimensional matrix for input is divided into suitable multidimensional vector matrix as required, carries out conversion then.

The direct transform of multidimensional vector matrix

B _IJ＝C _IIA _IJC _JJ ^T

The conversion of multidimensional vector matrix inversion

A _IJ＝C _II ^TB _IJC _JJ

The concrete form of wherein four-dimensional discrete cosine transform operator C is as follows:

C _IJ＝(c _uvmn)I＝(M，N)，J＝(M，N)

Wherein $c_{\mathrm{uvmn}}=\sqrt{\frac{2^2}{\mathrm{MN}}}c\left(u\right)c\left(v\right)\cos \frac{(2m+1)\mathrm{u\π}}{2M}\cos \frac{(2n+1)\mathrm{v\π}}{2N}$

$c\left(u\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},u=0\\ 1,u=\mathrm{others}\end{array}\right.$ $c\left(v\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},v=0\\ 1,v=\mathrm{others}\end{array}\right.$ $c\left(s\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},s=0\\ 1,s=\mathrm{others}\end{array}\right.$ $c\left(t\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},t=0\\ 1,t=\mathrm{others}\end{array}\right.$

For motion image sequence, vector I is a two-dimensional vector, and J is the one dimension scalar, and matrix A is the 3-D view piece, and B is the intact three-dimensional matrice coefficient of conversion, C _IIBe the four-vector quadrature cosine matrix that the present invention proposes, C _JJIt is the two-dimension discrete cosine transform matrix under the common meaning.

Therefore, the N of a multidimensional video stream component is represented by N fore-and-aft plane of N dimension matrix respectively.This expression of multidimensional video stream just his relation building between each component of the correlation of the position of the position of each pixel relation, each color space relation, time orientation and different angles in same model.We just can handle multidimensional video stream with the method for multidimensional vector matrix, make full use of the correlation of multidimensional video stream between each component, thereby realize further energy compression.In JPEG and mpeg standard, take all factors into consideration the blocking effect of complexity of calculation and image, image is divided into 8 * 8 sub-piece, carries out the cosine transform conversion.For compatible with it and utilize existing technology, and fully take into account blocking artifact and computation complexity, each component of the present invention also adopts 8 submatrix dividing method.

Claims (4)

1. multidimensional video stream encoding and decoding method efficiently is characterized in that finishing according to the following steps the multidimensional video stream signal is compressed:

The data extract step: the brightness of corresponding colour-video signal, color difference signal extract, and arrange and storage according to frame data;

Piecemeal step: according to input signal, be generally yuv format, respectively each component carried out piecemeal, on time shaft, select 2 ^mBig or small frame number, m gets positive integer, forms a cube piece, perhaps becomes the three-dimensional matrice piece; According to computation complexity and blocking artifact and with the compatibility of existing standard, be divided into 8 * 8 * 8 piece usually;

The matrixing step: utilization multidimensional vector orthogonal matrix, three-dimensional submatrix is carried out the orthogonal transform of multidimensional vector matrix, calculate three-dimensional coefficient matrix;

Vector quantization step: three-dimensional coefficient matrix is carried out vector quantization, calculate vector quantization system three-dimensional matrice;

Entropy coding step: vector quantization coefficient three-dimensional matrice is carried out entropy coding, obtain the change compression code matrix of digital color image signal three-dimensional matrice, finish compression to digital color picture signal.

2. a kind of multidimensional video stream encoding and decoding method efficiently according to claim 1, it is characterized in that said multidimensional vector defined matrix and algorithm thereof are: the multidimensional data permutation table is called multi-dimensional matrix, and the dimension of multi-dimensional matrix is divided into two groups, use two vector representations respectively, be called the multidimensional vector matrix, and defined the equating of multidimensional vector matrix, addition, number take advantage of, computings such as matrix is taken advantage of, transposition, and defined unit multidimensional vector matrix.

3. a kind of multidimensional video stream encoding and decoding method efficiently according to claim 1 is characterized in that the multidimensional vector matrix orthogonal transform formula that is adopted in the said matrixing step is:

Multi-dimensional matrix for input is divided into suitable multidimensional vector matrix as required, carries out orthogonal transform then:

The direct transform of multidimensional vector matrix

B _IJ＝C _IIA _IJC _JJ ^T

The conversion of multidimensional vector matrix inversion

A _IJ＝C _II ^TB _IJC _JJ

For motion image sequence, vector I is a two-dimensional vector, and J is the one dimension scalar, and matrix A is the 3-D view piece, and B is the intact three-dimensional matrice coefficient of conversion, and T is the multidimensional vector matrix transpose.

4. a kind of multidimensional video stream encoding and decoding method efficiently according to claim 1 is characterized in that the multidimensional vector orthogonal matrix that is adopted in the said matrixing step is the multidimensional vector orthogonal matrix of discrete cosine transform, and its formula is:

The concrete form of wherein four-dimensional discrete cosine transform operator C is as follows:

C _IJ＝(c _uvmn)，I＝(M，N)，J＝(M，N)

Wherein $c_{\mathrm{uvmn}}=\sqrt{\frac{2^2}{\mathrm{MN}}}c\left(u\right)c\left(v\right)\cos \frac{(2m+1)\mathrm{u\π}}{2M}\cos \frac{(2n+1)\mathrm{v\π}}{2N}$

$c\left(u\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},u=0\\ 1,u=\mathrm{others}\end{array}\right.$ $c\left(v\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},v=0\\ 1,v=\mathrm{others}\end{array}\right.$ $c\left(s\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},s=0\\ 1,s=\mathrm{others}\end{array}\right.$ $c\left(t\right)=\left\{\begin{array}{c}\frac{1}{\sqrt{2}},t=0\\ 1,t=\mathrm{others}\end{array}\right.$

Vector I is a two-dimensional vector, C _IIBe the four-vector quadrature cosine matrix that the present invention proposes, J is the one dimension scalar, C _JJIt is the two-dimension discrete cosine transform matrix under the common meaning.

Images