CN101833743A - Image based three-dimensional grid model blind watermarking method - Google Patents

Abstract

The invention relates to an image based three-dimensional grid model blind watermarking method. The existing method has poor stability. The method comprises a method for embedding a watermark and a method for extracting the watermark, wherein the method for embedding the watermark utilizes Fourier addition property to indirectly embed the watermark in the coordinate of the vertex of a grid model, namely firstly the watermark is embedded in a stable image, secondly the image embedded with the watermark is compared with the original image to obtain watermark information and finally the watermark information is embedded on the coordinate of the vertex of the grid model to obtain the three-dimensional grid model of the watermark. The method for extracting the watermark does not require the original grid and pretreatment, and can resist various attacks such as noise-adding, smoothness, enhancement, rotation, translation, scaling, resampling, clipping and some combined attacks.

Description

Three-dimensional grid model blind watermarking method based on image

Technical field

The present invention relates to a kind of three-dimensional grid model blind watermarking method based on image.

Background technology

Current many 3D mesh watermarking algorithm right and wrong blind and also when extracting watermark, need pre-service.Benedens proposes a privately owned mesh watermarking algorithm [Benedens 1999a].This method can be resisted the summit random alignment effectively, grid is changed and lattice simplified etc. attacked or operation.Yin Kangkang proposes the non-blind mesh watermarking algorithm [Yin2001a] at mesh transformations territory embed watermark based on the research work of [Guskov1999], but when grid to be detected was subjected to the operation of cutting or affined transformation, system need carry out reorientation and resample returning to original position, direction, size, topological structure and resolution level to it.Recently, Ohbuchi is based on their previous research work [Ohbuchi2001] and propose an effective mesh watermarking algorithm [Ohbuchi2002] that can embed high capacity information.Although this method has robustness to a lot of attacks, as change annexation, cutting, lattice simplified and grid is smooth, but still existing, it needs improved place, need a lot of information (information of original mesh, key and decision original mesh subdivision) as watermark extracting, before extracting watermark, need grid to be detected is carried out grid alignment and gridding again.

Kanai is applied to the method for wavelet transformed domain the 3D grid and has proposed a blind watermarking algorithm of strong grid [Kanai1998].This method can be resisted attacks such as affined transformation, part excision and random noise, but it to require original mesh be the 1-to-4 sub-structure.

Praun proposes the non-blind watermark method of a kind of grid [Praun1999] based on transform domain, and original mesh can be randomly topologically structured polygonal mesh.This method is that the nuclear by a spatial domain embeds information on the low frequency component coefficient of grid geological information.This algorithm can resist affined transformation, smooth, add attack such as make an uproar.In addition, if grid to be detected waits maintenances shape and changes the attack of topological structure through lattice simplified, before the detection watermark, need utilize original mesh to carry out grid and calibrate and resampling.

No matter can draw such conclusion from above analysis, if watermark information is embedded on the geometry signals of grid, be with spatial domain means or frequency domain means, can express with following formula:

v′＝v+B(v，w)

V ' is the apex coordinate of the grid of embed watermark, and v is the apex coordinate of original mesh, and w is watermark.B is a matrix or a function by the algorithm decision, and function can be an implicit expression, also can show.

If watermark directly is embedded into the matrix of being made up of the original mesh apex coordinate, grid can produce bigger unacceptable distortion so, and because the cumulative errors that LPM and FFT bring, watermark is difficult to be extracted out.The work of ORuanaidh is based on the image watermark algorithm of Fourier-Mellin conversion, and this algorithm has very strong robustness, especially to geometric transformation [ORuanaidh1998].But, because the error that computing such as interpolation and too many conversion bring makes algorithm be difficult to realize.This algorithm has stability preferably to general pattern, will regard piece image as by the matrix that the X-coordinate on original mesh summit is formed, and this algorithm has instability to this class image so, also just is difficult to watermark extracting is come out.

Summary of the invention

The objective of the invention is at the deficiencies in the prior art, a kind of three-dimensional grid model blind watermarking method based on image is provided.

The inventive method utilizes the Fourier addition property that watermark is embedded on the grid model apex coordinate indirectly, promptly earlier watermark is embedded in the stabilized image, then the image behind the embed watermark and original image are compared and obtain watermark information, at last watermark information is embedded the three-dimensional grid model that obtains embed watermark on the grid model apex coordinate.Experimental result proves that this algorithm has stronger robustness, and realizes extracting watermark without any need for supplementary and pre-service.

The inventive method comprises embed watermark method and extraction water mark method, wherein

The concrete steps of embed watermark method are:

Step (1) watermark encoder

As watermark information, character string comprises four kinds of characters with character string w, and every kind of character is by 50 and 11 sequence of forming, and the bit stream of character string w is expressed as

w＝(w ₁，w ₂，...，w ₂₄)，w _i∈{0.1}，i＝1，2，...，24。

Step (2) is resequenced to original mesh X-coordinate, makes it become the grid that ascending rising arrangement mode obtains a new X-coordinate;

Step (3) will be embedded into the noise W that obtains containing watermark information in the original image as the character string w of watermark information ^*(x, y);

Original image is a gray level image, be expressed as I (x y), is of a size of 64 * 64, x=1,2 ..., 64, y=1,2 ..., 64.This step is utilized based on Fourier conversion (FFT) and Log-Polar and is shone upon the geometrical invariants (RST) that (LPM) produces this original image, and concrete steps are as follows:

A. to original image I (x, y) carry out the Fourier conversion obtain the amplitude spectrum matrix B (x, y), x=1,2 ..., 64, y=1,2 ..., 64;

B. (x, (x, y), (x is that (x, center y) is the center of circle, external diameter r with original image I y) to intermediate frequency coefficient C to intermediate frequency coefficient C y) to select B ₁Be 52 internal diameter r ₂It is the coefficient in 42 the annulus scope;

C. then by bilinear interpolation with C (x y) is mapped to the log-polar coordinate system, with the C of Cartesian coordinate system (x, y) convert to the log-polar coordinate system C (ρ, θ),

$\left\{\begin{array}{c}x=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\cos (\mathrm{\θ}\×\mathrm{\Δ\θ})\\ y=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\sin (\mathrm{\θ}\×\mathrm{\Δ\θ})\end{array}\right.---\left(1\right)$

Wherein, Δ ρ=15/64, Δ θ=2 π/64; ρ=1,2 ..., 64, θ=1,2 ..., 64.

D. to C (ρ, θ) carry out the Fourier conversion obtain the amplitude spectrum matrix D (i, j), i=1,2 ..., 64, j=1,2 ..., 64;

E. with w _i(i j) obtains new amplitude spectrum matrix D to embed D ^*(i, j),

D ^*((k+32)，(12+l))＝D((k+32)，(12+l))+d (2)

Wherein, k=5,7 ,-5 ,-7, l=1,2 ..., 6, w _i=1 o'clock d=10 ⁸, w _i=0 o'clock d=0;

F. to D ^*(i j) carries out the image I that anti-Fourier conversion, anti-LPM conversion and anti-Fourier conversion obtain embed watermark respectively ^*(x, y).

G. use formula (3) to obtain watermark information W ^*

$W^{*}(i,j)=(I^{*}(i,j)-I(i,j))\×\left(\frac{1}{500}\right)---\left(3\right)$

Step (4) is with watermark information W ^*Be added to the new X-coordinate after the ordering, concrete steps are as follows:

H. select a point every a point the X-coordinate of the grid M after ordering, choose 4096 numerical value altogether, and the matrix A of composition 64 * 64 (i, j);

I. with watermark information W ^*(i, j) (i, j) addition obtain new matrix A with matrix A ^*(i, j), (i j) replaces to A with the A among the grid M ^*(i j), has obtained the grid M ' of embed watermark;

Be changed to method ripe in the image processing techniques, no longer explanation here based on Fourier conversion (FFT) and Log-Polar mapping (LPM) and corresponding contravariant in the embed watermark method.

The concrete steps of extracting water mark method are:

Step (1) is with grid M to be detected " the X-coordinate handle according to following formula:

The original X-coordinate that wherein, f (x) is grid M to be detected " original X-coordinate, g (x) is the grid M to be detected after handling ", T is a preset threshold, f (m) be f (x) and each three point of front and back with

f(m)＝f(x-3)+f(x-2)+f(x-1)+f(x)+f(x+1)+f(x+2)+f(x+3)

G (x) is carried out processing and amplifying obtain amplifying coordinate f ' (x), magnification function is defined as follows:

f′(x)＝g(x)+(f(x)-g(x))×β (5)

Wherein β is an amplification coefficient.

Step (2) is selected grid M to be detected " select a point every a point in the corresponding X-coordinate, choose 4096 numerical value altogether, the matrix A of composition 64 * 64 ' (i, j).

Step (3) to A ' (i, j) carry out the Fourier conversion obtain the amplitude spectrum matrix B ' (x, y).

Step (4) is selected B ', and (x, (x, y), (x is to be the center of circle, external diameter r ' with B ' (32,32) y) to intermediate frequency coefficient C ' to intermediate frequency coefficient C ' y) ₁Be 52 internal diameter r ' ₂It is the coefficient in 42 the annulus scope;

Then by bilinear interpolation with C ' (x y) is mapped to the log-polar coordinate system, with the C ' of Cartesian coordinate system (x, y) convert to the log-polar coordinate system C ' (ρ, θ),

$\left\{\begin{array}{c}x=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\cos (\mathrm{\θ}\×\mathrm{\Δ\θ})\\ y=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\sin (\mathrm{\θ}\×\mathrm{\Δ\θ})\end{array}\right.---\left(6\right)$

Wherein, Δ ρ=15/64, Δ θ=2 π/64; ρ=1,2 ..., 64, θ=1,2 ..., 64.

Step (5) to C ' (ρ, θ) carry out the Fourier conversion obtain the amplitude spectrum matrix D ' (i, j), i=1,2 ..., 64, j=1,2 ..., 64.

Step (6) is with D ' (i, j) sequence of part value D ' ((k+32), (12+l)) taking-up composition in, this sequence is made up of 24 numerical value, and this sequence average is divided into 4 groups, and maximal value is decided to be 1 in every group, remaining value is decided to be 0, method according to watermark encoder recovers corresponding character, k=5 wherein, 7,-5,-7, l=1,2,, 6.

In the extraction water mark method is method ripe in the image processing techniques based on Fourier conversion (FFT) and Log-Polar mapping (LPM), no longer explanation here.

The inventive method has good robustness to convergent-divergent and translation.We are the multiplication by constants respectively of the apex coordinate of the grid of embed watermark, just obtain the grid behind the convergent-divergent.If add the grid after constant just obtains translation respectively.According to addition and the multiplicative property of Fourier, the coefficient in the frequency domain all is exaggerated or adds a constant successively, and the strategy according to the extraction watermark of our algorithm does not influence us watermark recovery is come out.This method is extracted watermark does not need original mesh, does not need pre-service.Can resist a lot of attacks, attack comprise add make an uproar, smooth, enhancing, rotation, translation, convergent-divergent, resampling, cutting and some combination attacks.

Embodiment

Three-dimensional grid model blind watermarking method based on image comprises the embed watermark method and extracts water mark method that wherein the concrete steps of embed watermark method are:

Step (1) watermark encoder

As watermark information, character string comprises four kinds of characters with character string w, and every kind of character is by 50 and 11 sequence of forming, and the bit stream of character string w is expressed as

w＝(w ₁，w ₂，...，w ₂₄)，w _i∈{0.1}，i＝1，2，...，24。

Step (2) is resequenced to original mesh X-coordinate, makes it become the grid M that ascending rising arrangement mode obtains a new X-coordinate, and new X-coordinate has slickness preferably.Therefore, watermark can be regarded a kind of noise of new grid as.We can regulate these noises as required in the process that slips into watermark and extraction watermark.To the resequence key of the effect in whole algorithm of original mesh X-coordinate is to guarantee to obtain of original X-coordinate by smooth means to approach before extracting watermark.When we obtain this and approach, just can obtain containing the noise of watermark information by the difference of the X-coordinate behind relatively it and the embed watermark, then these noises are amplified, the noise after at last these being amplified carries out watermark extracting.

Step (3) will be embedded into the noise W that obtains containing watermark information in the original image as the character string w of watermark information ^*(x, y).In this section, we utilize the image that obtains embed watermark information based on the image watermark algorithm of Fourier conversion and Log-Polar mapping earlier.Then, image and the original image by embed watermark information in the spatial domain subtracts each other the noise that obtains containing watermark information.Watermark information is further dwindled to embed in the grid to reach the requirement of the transparency.

Original image is a gray level image, be expressed as I (x y), is of a size of 64 * 64, x=1,2 ..., 64, y=1,2 ..., 64.This step is utilized based on Fourier conversion (FFT) and Log-Polar and is shone upon the geometrical invariants (RST) that (LPM) produces this original image, and concrete steps are as follows:

A. to original image I (x, y) carry out the Fourier conversion obtain the amplitude spectrum matrix B (x, y), x=1,2 ..., 64, y=1,2 ..., 64.

B. (x, (x, y), (x is that (x, center y) is the center of circle, external diameter r with original image I y) to intermediate frequency coefficient C to intermediate frequency coefficient C y) to select B ₁Be 52 internal diameter r ₂It is the coefficient in 42 the annulus scope;

C. then by bilinear interpolation with C (x y) is mapped to the log-polar coordinate system, with the C of Cartesian coordinate system (x, y) convert to the log-polar coordinate system C (ρ, θ),

$\left\{\begin{array}{c}x=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\cos (\mathrm{\θ}\×\mathrm{\Δ\θ})\\ y=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\sin (\mathrm{\θ}\×\mathrm{\Δ\θ})\end{array}\right.---\left(1\right)$

Wherein, Δ ρ=15/64, Δ θ=2 π/64; ρ=1,2 ..., 64, θ=1,2 ..., 64.

D. to C (ρ, θ) carry out the Fourier conversion obtain the amplitude spectrum matrix D (i, j), i=1,2 ..., 64, j=1,2 ..., 64.

E. with w _i(i j) obtains new amplitude spectrum matrix D to embed D ^*(i, j),

D ^*((k+32)，(12+l))＝D((k+32)，(12+l))+d (2)

Wherein, k=5,7 ,-5 ,-7, l=1,2 ..., 6, w _i=1 o'clock d=10 ⁸, w _i=0 o'clock d=0.

F. to D ^*(i j) carries out the image I that anti-Fourier conversion, anti-LPM conversion and anti-Fourier conversion obtain embed watermark respectively ^*(x, y).

G. use formula (3) to obtain watermark information W ^*W ^*In the spatial domain, has the transparency preferably.

$W^{*}(i,j)=(I^{*}(i,j)-I(i,j))\×\left(\frac{1}{500}\right)---\left(3\right)$

Step (4) is with watermark information W ^*Be added to the new X-coordinate after the ordering: a part of choosing new X-coordinate is formed 64 * 64 matrixes.From the X-coordinate, select a point, choose 4096 numerical value altogether every a point.Then adjusted watermark information is added to the grid that just obtains embed watermark on this matrix.According to the addition property of Fourier conversion, the addition in spatial domain equals the addition of frequency domain, so, can from the grid of embed watermark, extract watermark.Concrete steps are as follows:

H. select a point every a point the X-coordinate of the grid M after ordering, choose 4096 numerical value altogether, and the matrix A of composition 64 * 64 (i, j).Matrix A can be regarded as the function that is defined on [64,64].F represents Discrete Fourier Transform, and the addition property of Fourier conversion is defined as follows:

F[f(x ₁，x ₂)+g(x ₁，x ₂)]＝F[f(x ₁，x ₂)]+F[g(x ₁，x ₂)]

I. with watermark information W ^*(i, j) (i, j) addition obtain new matrix A with matrix A ^*(i, j), (i j) replaces to A with the A among the grid M ^*(i j), has obtained the grid M ' of embed watermark.A ^*Frequency domain be subjected to W ^*The influence of frequency domain.

$F\left[A^{*}\right]=F[W^{*}(i,j)+A(i,j)]=F\left[W^{*}(i,j)\right]+F\left[A(i,j)\right]$

$=\left(F\right[I^{*}(i,j)]-F[I(i,j)\left]\right)\×\left(\frac{1}{500}\right)+F\left[A(i,j)\right]$

The concrete steps of extracting water mark method are:

Step (1) is with grid M to be detected " the X-coordinate handle according to following formula:

The original X-coordinate that wherein, f (x) is grid M to be detected " original X-coordinate, g (x) is the grid M to be detected after handling ", T is a preset threshold, f (m) be f (x) and each three point of front and back with

f(m)＝f(x-3)+f(x-2)+f(x-1)+f(x)+f(x+1)+f(x+2)+f(x+3)

G (x) is carried out processing and amplifying obtain amplifying coordinate f ' (x), magnification function is defined as follows:

f′(x)＝g(x)+(f(x)-g(x))×β (5)

Wherein β is an amplification coefficient.

Step (2) is selected grid M to be detected " select a point every a point in the corresponding X-coordinate, choose 4096 numerical value altogether, the matrix A of composition 64 * 64 ' (i, j).

Step (3) to A ' (i, j) carry out the Fourier conversion obtain the amplitude spectrum matrix B ' (x, y).

Step (4) is selected B ', and (x, (x, y), (x is to be the center of circle, external diameter r ' with B ' (32,32) y) to intermediate frequency coefficient C ' to intermediate frequency coefficient C ' y) ₁Be 52 internal diameter r ' ₂It is the coefficient in 42 the annulus scope;

Then by bilinear interpolation with C ' (x y) is mapped to the log-polar coordinate system, with the C ' of Cartesian coordinate system (x, y) convert to the log-polar coordinate system C ' (ρ, θ),

$\left\{\begin{array}{c}x=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\cos (\mathrm{\θ}\×\mathrm{\Δ\θ})\\ y=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\sin (\mathrm{\θ}\×\mathrm{\Δ\θ})\end{array}\right.---\left(6\right)$

Wherein, Δ ρ=15/64, Δ θ=2 π/64; ρ=1,2 ..., 64, θ=1,2 ..., 64.

Step (5) to C ' (ρ, θ) carry out Fouri er conversion obtain the amplitude spectrum matrix D ' (i, j), i=1,2 ..., 64, j=1,2 ..., 64.

Step (6) is with D ' (i, j) sequence of part value D ' ((k+32), (12+l)) taking-up composition in, this sequence is made up of 24 numerical value, and this sequence average is divided into 4 groups, and maximal value is decided to be 1 in every group, remaining value is decided to be 0, method according to watermark encoder recovers corresponding character, k=5 wherein, 7 ,-5,-7, l=1,2 ... 6.

Claims (1)

1. based on the three-dimensional grid model blind watermarking method of image, comprise the embed watermark method and extract water mark method, it is characterized in that:

The concrete steps of embed watermark method are:

Step (1) watermark encoder

As watermark information, character string comprises four kinds of characters with character string w, and every kind of character is by 50 and 11 sequence of forming, and the bit stream of character string w is expressed as

w＝(w ₁，w ₂，...，w ₂₄)，w _i∈{0.1}，i＝1，2，...，24；

Step (2) is resequenced to original mesh X-coordinate, makes it become the grid that ascending rising arrangement mode obtains a new X-coordinate;

Step (3) will be embedded into the noise W that obtains containing watermark information in the original image as the character string w of watermark information ^*(x, y);

Original image is a gray level image, be expressed as I (x y), is of a size of 64 * 64, x=1,2 ..., 64, y=1,2 ..., 64; This step is utilized based on Fourier conversion (FFT) and Log-Polar and is shone upon the geometrical invariants (RST) that (LPM) produces this original image, and concrete steps are as follows:

A. to original image I (x, y) carry out the Fourier conversion obtain the amplitude spectrum matrix B (x, y), x=1,2 ..., 64, y=1,2 ..., 64;

B. (x, (x, y), (x is that (x, center y) is the center of circle, external diameter r with original image I y) to intermediate frequency coefficient C to intermediate frequency coefficient C y) to select B ₁Be 52 internal diameter r ₂It is the coefficient in 42 the annulus scope;

C. then by bilinear interpolation with C (x y) is mapped to the log-polar coordinate system, with the C of Cartesian coordinate system (x, y) convert to the log-polar coordinate system C (ρ, θ),

$\left\{\begin{array}{c}x=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\cos (\mathrm{\θ}\×\mathrm{\Δ\θ})\\ {y=e}^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\sin (\mathrm{\θ}\×\mathrm{\Δ\θ})\end{array}\right.---\left(1\right)$

Wherein, Δ ρ=15/64, Δ θ=2 π/64; ρ=1,2 ..., 64, θ=1,2 ..., 64;

D. to C (ρ, θ) carry out the Fourier conversion obtain the amplitude spectrum matrix D (i, j), i=1,2 ..., 64, j=1,2 ..., 64;

E. with w _i(i j) obtains new amplitude spectrum matrix D to embed D ^*(i, j),

D ^*((k+32)，(12+l))＝D((k+32)，(12+l))+d (2)

Wherein, k=5,7 ,-5 ,-7, l=1,2 ..., 6, w _i=1 o'clock d=10 ⁸, w _i=0 o'clock d=0;

F. to D ^*(i j) carries out the image I that anti-Fourier conversion, anti-LPM conversion and anti-Fourier conversion obtain embed watermark respectively ^*(x, y);

G. use formula (3) to obtain watermark information W ^*

$W^{*}(i,j)=(I^{*}(i,j)-I(i,j))\×\left(\frac{1}{500}\right)---\left(3\right)$

Step (4) is with watermark information W ^*Be added to the new X-coordinate after the ordering, concrete steps are as follows:

H. select a point every a point the X-coordinate of the grid M after ordering, choose 4096 numerical value altogether, and the matrix A of composition 64 * 64 (i, j);

I. with watermark information W ^*(i, j) (i, j) addition obtain new matrix A with matrix A ^*(i, j), (i j) replaces to A with the A among the grid M ^*(i j), has obtained the grid M ' of embed watermark;

The concrete steps of extracting water mark method are:

Step (1) is with grid M to be detected " the X-coordinate handle according to following formula:

The original X-coordinate that wherein, f (x) is grid M to be detected " original X-coordinate, g (x) is the grid M to be detected after handling ", T is a preset threshold, f (m) be f (x) and each three point of front and back with

f(m)＝f(x-3)+f(x-2)+f(x-1)+f(x)+f(x+1)+f(x+2)+f(x+3)

G (x) is carried out processing and amplifying obtain amplifying coordinate f ' (x), magnification function is defined as follows:

f′(x)＝g(x)+(f(x)-g(x))×β (5)

Wherein β is an amplification coefficient;

Step (2) is selected grid M to be detected " select a point every a point in the corresponding X-coordinate, choose 4096 numerical value altogether, the matrix A of composition 64 * 64 ' (i, j);

Step (3) to A ' (i, j) carry out the Fourier conversion obtain the amplitude spectrum matrix B ' (x, y);

Step (4) is selected B ', and (x, (x, y), (x is to be the center of circle, external diameter r with B ' (32,32) y) to intermediate frequency coefficient C ' to intermediate frequency coefficient C ' y) ₁' be 52 internal diameter r ₂' be the coefficient in 42 the annulus scope;

Then by bilinear interpolation with C ' (x y) is mapped to the log-polar coordinate system, with the C ' of Cartesian coordinate system (x, y) convert to the log-polar coordinate system C ' (ρ, θ),

$\left\{\begin{array}{c}x=e^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\cos (\mathrm{\θ}\×\mathrm{\Δ\θ})\\ {y=e}^{\mathrm{\ρ}\×\mathrm{\Δ\ρ}}\sin (\mathrm{\θ}\×\mathrm{\Δ\θ})\end{array}\right.---\left(6\right)$

Wherein, Δ ρ=15/64, Δ θ=2 π/64; ρ=1,2 ..., 64, θ=1,2 ..., 64;

Step (5) to C ' (ρ, θ) carry out the Fourier conversion obtain the amplitude spectrum matrix D ' (i, j), i=1,2 ..., 64, j=1,2 ..., 64;

Step (6) is with D ' (i, j) sequence of part value D ' ((k+32), (12+l)) taking-up composition in, this sequence is made up of 24 numerical value, and this sequence average is divided into 4 groups, and maximal value is decided to be 1 in every group, remaining value is decided to be 0, method according to watermark encoder recovers corresponding character, k=5 wherein, 7,-5,-7, l=1,2,, 6.