CN101257629A - Encoding and decoding method of image sequence nondestructive compression based on interested area - Google Patents

Abstract

The invention relates to an interesting area based nondestructive compression method, which is mainly used for motion image sequence. The coding/decoding method of nondestructive compression based on motion image sequence of interesting area improves greatly the compression ratio of image on the premise of guaranteeing non-distortion recovery of interesting area image signal. The specific step comprises: step of extracting interesting area; step of dividing blocks; step of transforming matrix; step of coding entropy; step of generating four-dimensional n-order orthogonal transformation Hadamard matrix. The core content of the invention is that the definition of the four-dimensional n-order matrix and the definition of its algorithm are introduced in the image sequence signal compression method, and the corresponding four-dimensional n-order orthogonal transformation are introduced, and Hadamard four-dimensional n-order orthogonal matrix and its Kronic multiplication generation method are provided.

Description

Decoding method based on the compression of the image sequence nondestructive of area-of-interest

Technical field:

The present invention relates to lossless compression method, be mainly used in motion image sequence based on the image sequence of area-of-interest.

Background technology:

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.One of subject matter that the application of multimedia technology and Internet and development are faced is exactly expression, transmission and the storage that solves huge image information, and the basic method that addresses this problem is exactly an image compression, image compression is exactly not have distortion or do not having under the prerequisite of obvious distortion, and the message bit pattern of image is transformed into another can be with the representation of image data amount reduction.Image compression is divided into lossless compress and lossy compression method two big classes, and lossless compress can undistorted ground reconstructing image information.Some imaging device prices are extremely expensive, image to obtain cost very big, this compression of some application requirements can't harm in the reality, but the compression ratio of its compressed image is very low, along with the growth of image data amount, entire image is all carried out this method of lossless compress can not satisfy in the practical application requirement high compression ratio.Lossy compression method can obtain higher compression ratio, but it can not guarantee that the useful information of image do not lose, and the image reconstruction quality is also relatively poor relatively.

In actual applications, the partial information in the piece image is more important, and this zone is called area-of-interest (ROI).In medical image, the doctor is only interested in the lesion portion, in remote sensing images, only to target exist regional interested.Image sequence nondestructive compression method based on area-of-interest (ROI), it is exactly area-of-interest employing lossless compress at image, thereby guaranteed that important information do not lose, and in other zones in order to improve the compression ratio of image, then adopt lossy compression method, taken into account the requirement of picture quality and compression ratio.

JPEG-IS is ISO/ITU one a newly approved Lossless Image Compression Algorithm standard.In JPEG2000, just put forward the international standard of Lossless Image Compression Algorithm in fact.

And it is less at the lossless compression method standard of video, a kind of is separately each frame to be adopted lossless compression method, give no thought to the correlation on the time shaft, mainly be in addition on existing lossy compression method standard base, MPEG-2 for example, H.264, AVS etc., the subalgorithm that wherein diminishes is removed, realized the lossless compress function of video.And present international standard and other non-standard compression method all are to adopt technology such as motion prediction to remove correlation on the time shaft.

Summary of the invention:

The object of the present invention is to provide decoding method, under the prerequisite that guarantees the undistorted recovery of area-of-interest picture signal, improved the compression ratio of image widely based on the lossless compress of the motion image sequence of area-of-interest.

The present invention finishes as follows area-of-interest image sequence signal is compressed:

Area-of-interest extraction step: manually select as required or with the result of image recognition as a reference, in video, determine area-of-interest.

The piecemeal step: according to input signal, to image sequence, just each component of multidimensional data signal carries out piecemeal, on row, column and time shaft, selects the size of suitable piece, forms a cube piece, perhaps is called 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.

Matrixing step: three-dimensional submatrix is carried out the Four-dimensional N Exponent orthogonal transform, calculate the transform coefficient matrix of the sub-piece of three-dimensional image sequence;

The entropy coding step: the three-dimensional matrice intact to orthogonal transform scans, entropy coding, obtains the encoder matrix of image sequence signal three-dimensional matrice, finishes the compression to vision signal.

Four-dimensional N Exponent orthogonal transform hadamard matrix generates step: according to the four-dimensional Hadamard orthogonal matrix of the low order that provides, utilize the method for multidimensional Crow internal medicine product, produce the four-dimensional quadrature hadamard matrix of high-order.

Said Four-dimensional N Exponent Matrix orthogonal transform comprises:

The Four-dimensional N Exponent Matrix direct transform

B _III，n＝H _IV，nA _III，nH _II，n

The Four-dimensional N Exponent Matrix inverse transformation

A _III，n＝H _IV，nB _III，nH _II，n

Said Four-dimensional N Exponent quadrature hadamard matrix, its four-dimensional second order quadrature hadamard matrix is expressed as follows:

$\left[\begin{array}{cc}1& 1\\ -1& -1\end{array}\right]$ $\left[\begin{array}{cc}1& -1\\ 1& -1\end{array}\right]$

$\left[\begin{array}{cc}1& 1\\ 1& 1\end{array}\right]$ $\left[\begin{array}{cc}1& -1\\ -1& 1\end{array}\right]$

The generation way of said multidimensional Crow internal medicine product:

A and B are with the multi-dimensional matrix of dimension, then claim following matrix in block form $A\⊗B={\left(a_{i_1{i}_2\·\·\·i_n}B\right)}_{I_1{J}_1\×I_2{J}_2\×\·\·\·\×I_n{J}_n}$ For the multidimensional Crow internal medicine of A and B is amassed: direct product and tensor product;

Hadamard orthogonal matrix and multidimensional Crow internal medicine product method according to four-dimensional second order can produce Hadamard orthogonal matrixes such as four-dimensional quadravalence, four-dimensional eight rank.

Technique effect is: the present invention has considered between the row and row of image sequence signal comprehensively, between row and the row, redundant information on the time shaft between frame and the frame, and considered the integral transformation in time, space, thereby under the prerequisite that guarantees the undistorted recovery of signal, improved the compression ratio of digital video signal.

Description of drawings:

The flow chart of the decoding method that the image sequence nondestructive based on area-of-interest of Fig. 1 indication of the present invention compresses.

Embodiment:

Further specify particular content of the present invention and execution mode thereof below in conjunction with accompanying drawing.

Core content of the present invention is to have introduced the definition of Four-dimensional N Exponent Matrix and the definition of algorithm thereof in the image sequence compression method, and introduced corresponding Four-dimensional N Exponent Matrix orthogonal transform, and completely newly proposed Hadamard Four-dimensional N Exponent orthogonal matrix, and its Crow internal medicine product produces way.And the extraction of area-of-interest 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 four-matrix definition

Four-dimensional data arrangement table [a on the F _Ijst] be called I * J * S * T rank four-matrix.

The Four-dimensional N Exponent Matrix definition

For I * J * S * T rank four-matrix A arbitrarily, if its exponent number satisfies: I=J=S=T=n, i.e. A _{I * J * S * T}=[a _Ijst] _{N * n * n * n}The time, claim that then A is a Four-dimensional N Exponent Matrix, is designated as A _{IV, n}=[a _Ijst] _{IV, n}Obviously, Four-dimensional N Exponent Matrix is exactly the Four-dimensional N Exponent square formation, is the special four-matrix of a class.

Below use capitalization A, B, C wait and represent a Four-dimensional N Exponent Matrix.When needs illustrate the exponent number n of four-matrix, available symbols A _{IV, n}Represent.

Four-dimensional N Exponent Matrix character:

(1) works as A _{IV, n}=[a _Ijst] _IVnFour variable i, j, s, any one among the t is taken as constant, and remaining three variable has just constituted a three-dimensional cube matrix data table when minimum value changes to maximum, is called the cube vector;

(2) when four variable i, j, s, any two among the t are taken as constant, and remaining two variable has just constituted a planar square tables of data when minimum value changes to maximum, is called planar square vector (being similar to the two-dimentional square formation on the ordinary meaning);

The Four-dimensional N Exponent Matrix operational criterion

1, equates

If for Four-dimensional N Exponent Matrix A arbitrarily _{IV, n}=[a _Ijst] _{IV, n}And B _{IV, n}=[b _Ijst] _{IV, n}, if their corresponding element is equal, promptly

a _ijst＝b _ijst(1≤i≤n；i≤j≤n；1≤s≤n；1≤t≤n)

We just say that Four-dimensional N Exponent Matrix A equates with B, and that note is A=B.

Each element all is that zero Four-dimensional N Exponent Matrix is called zero Four-dimensional N Exponent Matrix, still represents with symbol 0.

2, addition

If Four-dimensional N Exponent Matrix A _{IV, n}=[a _Ijst] _{IV, n}And B _{IV, n}=[b _Ijst] _{IV, n}, then the addition of Four-dimensional N Exponent Matrix is defined as:

C _IV，n＝[c _ijst] _IV，n＝[a _ijst+b _ijst] _IV，n

We following formula call A and B and, be designated as C=A+B.

3, number is taken advantage of

If m ∈ is F, n ∈ F; A, B are any Four-dimensional N Exponent Matrix, and then its number is taken advantage of and is defined as: [m * a _Ijst] _IVn, be designated as mA, promptly each element of Four-dimensional N Exponent Matrix all multiply by several m.

4, Four-dimensional N Exponent Matrix multiplication

For being without loss of generality, be provided with Four-dimensional N Exponent Matrix A _{IV, n}=[a _Ijpq] _{IV, n}, B _{IV, n}=[b _Xyst] _{IV, n}, C _{IV, n}=[c _Ijst] _{IV, n}, then the Four-dimensional N Exponent Matrix multiplication is defined as following equation establishment:

$c_{\mathrm{ijst}}=\underset{p=1}{\overset{n}{\mathrm{\Σ}}}\underset{q=1}{\overset{n}{\mathrm{\Σ}}}{a}_{\mathrm{ijpq}}\·b_{\mathrm{pqst}}=\underset{x=1}{\overset{n}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{a}_{\mathrm{ijxy}}\·b_{\mathrm{xyst}}$

Can be designated as C=AB.

5, Four-dimensional N Exponent unit matrix

Known Four-dimensional N Exponent Matrix A _{IV, n}=[a _Ijst] _{IV, n}

If $a_{\mathrm{ijpg}}=\left\{\begin{array}{cc}1& (i,j)=(s,t)\\ 0& (i,j)\≠(s,t)\end{array}\right.$ (1≤i≤n; I≤j≤n; 1≤s≤n; 1≤t≤n), then claim A _{IV, n}For the Four-dimensional N Exponent unit matrix, be designated as E _{IV, n}

Easily demonstrate,prove, for any Four-dimensional N Exponent Matrix A _{IV, n}, perseverance has:

A _IV，nE _IV，n＝E _IV，nA _IV，n＝A _IV，n

6, the transposition operational criterion of Four-dimensional N Exponent Matrix

If Four-dimensional N Exponent Matrix A _{IV, n}=[a _Ijst] _{IV, n}, B _{IV, n}=[b _Ijst] _{IV, n}

If satisfy b _Ijst=a _Stij(1≤i≤n; I≤j≤n; 1≤s≤n; 1≤t≤n), then claim B _{IV, n}Be A _{IV, n}Transposed matrix is designated as B _{IV, n}=(A _{IV, n}) ^T

If A _{IV, n}=(A _{IV, n}) ^T, then claim A _{IV, n}Be the Four-dimensional N Exponent symmetrical matrix.

7, the Crow internal medicine of multi-dimensional matrix is long-pending

If $A_{I_1\×I_2\×\·\·\·I_n}={\left(a_{i_1{i}_2\·\·\·i_n}\right)}_{I_1\×I_2\×\·\·\·I_n},$ $B_{J_1\×J_2\×\·\·\·J_n}={\left(b_{j_1{j}_2\·\·\·j_n}\right)}_{J_1\×J_2\×\·\·\·J_n},$ A and B are with the multi-dimensional matrix of dimension, then claim following matrix in block form

$A\⊗B={\left(a_{i_1{i}_2\·\·\·i_n}B\right)}_{I_1{J}_1\×I_2{J}_2\×\·\·\·\×I_n{J}_n}$

Crow internal medicine long-pending (direct product, tensor product) for A and B.

Hadamard matrix is the special quadrature square formation of a class, and its element constitutes by+1 and-1, and any two row of matrix are mutually orthogonal.Utilize the Silvester structured approach, we can generate exponent number is 2 ^KHadamard matrix:

$H_{\mathrm{IV},2^K}=\left[\begin{array}{cc}{H}_{2^{K-1}}& H_{2^{K-1}}\\ H_{2^{K-1}}& -H_{2^{K-1}}\end{array}\right]=H_{\mathrm{IV},2}\⊗H_{\mathrm{IV},2^{K-1}}$

$\left[\begin{array}{cc}1& 1\\ -1& -1\end{array}\right]$ $\left[\begin{array}{cc}1& -1\\ 1& -1\end{array}\right]$

H wherein _{IV, 1}=[1], $H_{\mathrm{IV},2}=\left[\begin{array}{cc}1& 1\\ 1& 1\end{array}\right]$ $\left[\begin{array}{cc}1& -1\\ -1& 1\end{array}\right]$

2≤K ∈ N, symbol "

" expression multidimensional Kronecker takes advantage of.

8, Four-dimensional N Exponent Matrix conversion

The Four-dimensional N Exponent Matrix direct transform

B _III，n＝H _IV，nA _III，nH _II，n

The Four-dimensional N Exponent Matrix inverse transformation

A _III，n＝H _IV，nB _III，nH _II，n

A wherein _{III, n}For previous step rapid-divided the three-dimensional sub-piece of image sequence of piece, B in the piecemeal step _{III, n}Be the three-dimensional coefficient after the conversion, H _{II, n}Be the two-dimentional Hadamard orthogonal transform matrix under the ordinary meaning, H _{IV, n}The Four-dimensional N Exponent Hadamard orthogonal transform matrix that provides for the present invention.

Therefore, three of image sequence components are represented by three fore-and-aft planes of three-dimensional matrice respectively.This expression of image sequence is just built the correlation of the position of his each pixel relation, time orientation in same model.We just can make full use of the correlation between each component with the method processing moving sequence of multi-dimensional matrix, thereby realize further energy compression.

Claims (4)

1, a kind of decoding method of the image sequence nondestructive compression based on area-of-interest is characterized in that finishing as follows area-of-interest image sequence signal is compressed:

Area-of-interest extraction step: manually select as required or with the result of image recognition as a reference, in video, determine area-of-interest;

The piecemeal step: according to input signal, to image sequence, promptly each component of multidimensional data signal carries out piecemeal, on row, column and time shaft, selects 2 ^mThe size of piece, m gets positive integer, forms a cube piece, is the three-dimensional matrice piece; According to computation complexity and blocking artifact and with the compatibility of existing standard, get m=3 usually, i.e. 8 * 8 * 8 piece;

Matrixing step: three-dimensional submatrix is carried out the Four-dimensional N Exponent Matrix orthogonal transform, calculate the transform coefficient matrix of the sub-piece of three-dimensional image sequence;

The entropy coding step: the three-dimensional matrice intact to orthogonal transform scans, entropy coding, obtains the encoder matrix of image sequence signal three-dimensional matrice, finishes the compression to vision signal;

Four-dimensional N Exponent orthogonal transform hadamard matrix generates step: according to the four-dimensional Hadamard orthogonal matrix of the low order that provides, utilize the method for multidimensional Crow internal medicine product, produce the four-dimensional quadrature hadamard matrix of high-order.

2, the decoding method that compresses based on the image sequence nondestructive of area-of-interest according to claim 1 is characterized in that said Four-dimensional N Exponent Matrix orthogonal transform comprises:

Four-dimensional N Exponent Matrix direct transform B _{III, n}=H _{IV, n}A _{III, n}H _{II, n}

Four-dimensional N Exponent Matrix inverse transformation A _{III, n}=H _{IV, n}B _{III, n}H _{II, n}

3, according to the decoding method that compresses based on the image sequence nondestructive of area-of-interest according to claim 1, it is characterized in that said Four-dimensional N Exponent quadrature hadamard matrix, its four-dimensional second order quadrature hadamard matrix is expressed as follows:

$\left[\begin{array}{cc}1& 1\\ -1& -1\end{array}\right]$ $\left[\begin{array}{cc}1& -1\\ 1& -1\end{array}\right]$

$\left[\begin{array}{cc}1& 1\\ 1& 1\end{array}\right]$ $\left[\begin{array}{cc}1& -1\\ -1& 1\end{array}\right].$

4, the decoding method of the image sequence nondestructive compression based on area-of-interest according to claim 1 is characterized in that the generation way of said multidimensional Crow internal medicine product:

A and B are with the multi-dimensional matrix of dimension, then claim following matrix in block form $A\⊗B={\left(a_{i_1{i}_2\·\·\·i_n}B\right)}_{I_1{J}_1\×I_2{J}_2\×\·\·\·\×I_n{J}_n}$ For A with

The multidimensional Crow internal medicine of B is long-pending: direct product and tensor product;

Hadamard orthogonal matrix and multidimensional Crow internal medicine product method according to four-dimensional second order can produce Hadamard orthogonal matrixes such as four-dimensional quadravalence, four-dimensional eight rank.

Images