CN103854250A - Self-synchronizing image watermarking method applied to wireless multimedia sensor network - Google Patents

Abstract

The invention discloses a self-synchronizing image watermarking method applied to a wireless multimedia sensor network. Watermarking information and synchronizing signals are embedded into an observation domain formed after original image sparse treatment is carried out to make watermarks have the self-synchronizing ability. Local characteristics of observation values are utilized, the efficiency of searching the observation domain for the synchronizing signals is effectively improved, and the contradiction between synchronizing signal robustness and the search range is well solved. Furthermore, because original images are not needed when the watermarks need extracting, the method is a blind watermark method. In this way, the algorithm has strong practicability and has certain application value.

Description

A kind of motor synchronizing image watermark method that is applied to wireless multimedia sensor network

Technical field

The invention belongs to digital image watermarking technology field, relate to a kind of motor synchronizing image watermark method that is applied to wireless multimedia sensor network.

Background technology

In recent years, along with a kind of proposition and development of new data acquisition technology-compressed sensing theory, it has obtained a series of important achievements in research in analog information conversion, the field such as picture, bio-sensing that is compressed into.Compressed sensing theory is utilized the sparse property priori of signal, by constructing suitable measurement matrix, signal is carried out to linear random projection, obtains the compression measured value of minute quantity.Because measured value has retained the prototype structure of signal, therefore can be by applicable optimized algorithm Exact Reconstruction original signal.

In digital figure watermark field, compressed sensing theory has also obtained preliminary application.Someone has proposed a kind of watermarking algorithm based on compressed sensing and LPDC code.Algorithm is by carrying out linear random projection to original image, and the compression measured value obtaining, through LDPC coding, is embedded in original image in watermark mode; When watermark detection, tampered image is carried out to linear random projection equally, the measured value obtaining is as side information and carry out LDPC decoding.Distort detection by the compression measured value realization of comparing between original image and tampered image, experimental result shows that this algorithm has stronger anti-attack ability and distorts preferably detectability.Afterwards, all swallows propose again a kind of compressed sensing watermarking algorithm based on hash message authentication code.Someone proposes fragile zero watermarking algorithm of a kind of image half based on piecemeal compressed sensing again.Algorithm is first divided into image some piecemeals, and point block size can be adjusted according to watermark data amount and tampering location precision.According to compressive sensing theory, each image block is observed again, and observed reading is preserved as zero watermark information registration.Experimental result shows, this algorithm can accurately be located and illegally distorted and recover the region being tampered by watermark information.Someone proposed a kind of video based on compressed sensing and distorted detection watermarking algorithm afterwards.Algorithm adopts compressed sensing generate half fragile authenticating water-mark of I two field picture and be embedded in the medium-high frequency coefficient of I two field picture, adopt Hash operation generate the Watermarking for Integrity of P frame number and be embedded in the motion vector of P frame, experimental result show this algorithm between frame of video, distort with frame in distort and all there is good detectability, can be accurate to the sub-block of picture frame.

Summary of the invention

In order to overcome the defect existing in prior art, the invention provides a kind of motor synchronizing image watermark method that is applied to wireless multimedia sensor network, algorithm embeds watermark information and synchronizing signal in the observation territory after original image rarefaction, makes watermark have self-synchronization.Utilize observed reading local characteristics simultaneously, effectively improved the efficiency of searching for synchronizing signal in observation territory, solved preferably the contradiction between synchronizing signal robustness and volumes of searches.Its technical scheme is as follows:

Be applied to a motor synchronizing image watermark method for wireless multimedia sensor network,

Comprise the following steps:

A watermark embeds

Original image I is carried out wavelet decomposition by a1, obtains sparse coefficient of wavelet decomposition matrix A ₁;

A2 is to the coefficient of wavelet decomposition A after sparse ₁be multiplied by the observing matrix that user key generates, obtain observing territory compressed signal A ₂;

A3 will observe territory compressed signal A ₂of one-dimensional, carries out watermark embedding according to following formula;

$A_i=\left\{\begin{array}{c}{A}_i-\mathrm{mod}(A_i,S)+3S/4m_i=1\\ A_i-\mathrm{mod}(A_i,S)+S/4m_i=-1\end{array}\right.$

mod(A _i,S)＝

Wherein

for downward bracket function, S is watermark embed strength;

A4) be reconstructed into 2 dimension matrixes by what obtain containing watermark sequence, adopt OMP algorithm to recover to obtain sparse signal A with watermarked information ₃;

A5) to A ₃carry out wavelet inverse transformation, obtain containing watermarking images I _w;

B. watermark extracting

Algorithm of the present invention does not need original image in the time carrying out watermark extracting, is a kind of blind watermarking algorithm, and watermark extraction process is as described below in detail:

B1) will be containing watermarking images I _wcarry out wavelet decomposition, obtain coefficient of wavelet decomposition sparse matrix B ₁;

B2) to the coefficient of wavelet decomposition B after sparse ₁be multiplied by observing matrix, obtain observing territory compressed signal B ₂;

B3) will observe territory compressed signal B ₂of one-dimensional, carries out watermark extracting according to following formula;

$w_i=\left\{\begin{array}{c}1,\mathrm{mod}(A_i^{*},S)\&GreaterEqual;S/2\\ -1,\mathrm{mod}(A_i^{*},S)<S/2\end{array}\right.$

B4) sequence { w from extracting _iin determine after synchronizing signal, then carry out the extraction of watermark signal.

Can obtain from above formula, at A _iwhile changing d,

as long as d ∈ (nS-S/4, nS+S/4) interval, can both be from correctly extract m _i.

Further preferably, this method adopts m sequence as synchronizing signal.If { a _nand { b _ntwo m sequences with same period T, a _n, b _n∈ { 1,1}, sequence { a _nand { b _nbetween cross correlation function be defined as

${\mathrm{\&rho;}}_{a,b}\left(\mathrm{\&tau;}\right)=\frac{1}{T}\underset{n=1}{\overset{T}{\mathrm{\&Sigma;}}}{a}_n{b}_{n-\mathrm{\&tau;}}$

M sequence { a _nautocorrelation function there is following character:

${\mathrm{\&rho;}}_{a,a}\left(\mathrm{\&tau;}\right)=\left(\begin{array}{c}1,\mathrm{\&tau;}=0\\ -\frac{1}{T},\mathrm{\&tau;}\&NotEqual;0\end{array}\right.$

If τ=0 o'clock, m sequence { a _nand { b _ncross-correlation coefficient

ρ _a,b(0)≥h/T

Wherein h is threshold value, gets odd number.The implication of threshold value h is as infructescence { b _nonly have at most (T-h)/2 bit and { a _nnot identical, think { b _nit is a synchronizing signal.

Compared with prior art, beneficial effect of the present invention:

The present invention embeds watermark information and synchronizing signal in the observation territory after original image rarefaction, makes watermark have self-synchronization; The present invention utilizes observed reading local characteristics, has effectively improved the efficiency of searching for synchronizing signal in observation territory, has solved preferably the contradiction between synchronizing signal robustness and volumes of searches.In addition, this algorithm without original image, is a kind of blind watermarking algorithm in the time extracting watermark, and therefore algorithm of the present invention has stronger practicality, has certain using value.

Accompanying drawing explanation

Fig. 1 is hiding data structure;

Fig. 2 is that watermark embeds process flow diagram;

Fig. 3 is the test of the algorithm transparency, wherein, and Fig. 3 (a) original image Lena; Fig. 3 (b) original image Baboon; Fig. 3 (c) original image Peppers; Fig. 3 (d) original image Barbara; Fig. 3 (e) is containing watermarking images Lena, PSNR=34.09; Fig. 3 (f) is containing watermarking images Baboon, PSNR=32.76; Fig. 3 (g) is containing watermarking images Peppers, PSNR=33.55; Fig. 3 (h) is containing watermarking images Barbara, PSNR=32.95;

Fig. 4 is safety analysis test result.

Embodiment

Further illustrate technical scheme of the present invention below in conjunction with the drawings and specific embodiments.

Compressed sensing (CompressiveSensing, CS) core concept is a signal class to sparse prior, obtain fraction observed reading through nonlinear sampling, as long as observed reading comprises enough good approximation signals, original signal can be by the linearity of certain type or non-linear decoded mechanism high probability Exact Reconstruction.CS theory is made up of the sparse conversion of signal, three elements such as incoherent measurement and the reconstruction of sparse signal of sparse signal.Wherein the reconstruction algorithm of fast and stable is the main research of CS, is also that CS moves towards practical key.

If x ∈ is R ⁿfor original signal, obtain accidental projection signal y ∈ R by y=Φ x ^m, wherein Φ ∈ R ^{m × n}be called measurement matrix, meet the appearance conditions (RestrictedIsometryProperty, RIP) such as restriction.If original signal x is k-sparse (having n nonzero element of k < <), the target of compressed sensing is the measured value y Exact Reconstruction original sparse signal x by accidental projection.This target can be equivalent to following optimization problem:

x＝argmin||x|| ₀s.t.y＝Φx (1)

Be that signal x is the solution of formula (1) minimization problem.

Conventionally, the demoder based on linear programming solves needs cK projection, wherein c ≈ log ₂(1+N/K), rebuilding complexity is O (N ³).Cands and Tao prove to have obtained the sparse recovery theorem under the appearance conditions such as restriction: have parameter (2n if measure matrix Φ, 0.2) restrictive condition is held in wait, and the unique solution that each n-sparse vector x can be served as protruding optimization problem is measured Φ x by Exact Reconstruction from it.

Watermarking algorithm

Synchronizing signal

The present invention adopts m sequence as synchronizing signal.If { a _nand { b _ntwo m sequences with same period T, a _n, b _n∈ { 1,1}, sequence { a _nand { b _nbetween cross correlation function be defined as

${\mathrm{\&rho;}}_{a,b}\left(\mathrm{\&tau;}\right)=\frac{1}{T}\underset{n=1}{\overset{T}{\mathrm{\&Sigma;}}}{a}_n{b}_{n-\mathrm{\&tau;}}---\left(2\right)$

M sequence the autocorrelation function of an} has following character:

${\mathrm{\&rho;}}_{a,a}\left(\mathrm{\&tau;}\right)=\left(\begin{array}{c}1,\mathrm{\&tau;}=0\\ -\frac{1}{T},\mathrm{\&tau;}\&NotEqual;0\end{array}\right.---\left(3\right)$

If τ=0 o'clock, m sequence { a _nand { b _ncross-correlation coefficient

ρ _a,b(0)≥h/T (4)

Wherein h is threshold value, gets odd number.The implication of threshold value h is as infructescence { b _nonly have at most (T-h)/2 bit and { a _nnot identical, think { b _nit is a synchronizing signal.

Watermark embeds

Watermark sequence and synchronous m sequence are all converted to 1,1} sequence, and according to Fig. 1 mode tectonic sequence { m _i| m _i{ 1,1}}, wherein synchronous m sequence is placed on { m to ∈ _ifront portion, watermark sequence is placed on { m _irear portion.

Algorithm watermark of the present invention embeds process flow diagram as shown in Figure 2.

Watermark embed process is as described below in detail:

1) original image I is carried out to wavelet decomposition, obtain sparse coefficient of wavelet decomposition matrix A ₁;

2) to the coefficient of wavelet decomposition A after sparse ₁be multiplied by the observing matrix that user key generates, obtain observing territory compressed signal A ₂;

3) will observe territory compressed signal A ₂of one-dimensional, carries out watermark embedding according to formula (5);

$A_i=\left\{\begin{array}{c}{A}_i-\mathrm{mod}(A_i,S)+3S/4m_i=1\\ A_i-\mathrm{mod}(A_i,S)+S/4m_i=-1\end{array}\right.---\left(5\right)$

mod(A _i,S)＝

Wherein

for downward bracket function, S is watermark embed strength.

4) be reconstructed into 2 dimension matrixes by what obtain containing watermark sequence, adopt OMP algorithm to recover to obtain sparse signal A with watermarked information ₃;

5) to A ₃carry out wavelet inverse transformation, obtain containing watermarking images I _w.

Watermark extracting

Algorithm of the present invention does not need original image in the time carrying out watermark extracting, is a kind of blind watermarking algorithm, and watermark extraction process is as described below in detail:

1) will be containing watermarking images I _wcarry out wavelet decomposition, obtain coefficient of wavelet decomposition sparse matrix B ₁;

2) to the coefficient of wavelet decomposition B after sparse ₁be multiplied by observing matrix, obtain observing territory compressed signal B ₂;

3) will observe territory compressed signal B ₂of one-dimensional, carries out watermark extracting according to formula (7) ^[12];

$w_i=\left\{\begin{array}{c}1,\mathrm{mod}(A_i^{*},S)\&GreaterEqual;S/2\\ -1,\mathrm{mod}(A_i^{*},S)<S/2\end{array}\right.---\left(7\right)$

4) sequence { w from extracting _iin determine after synchronizing signal, then carry out the extraction of watermark signal.

Can obtain from above formula, at A _iwhile changing d,

as long as d ∈ (nS-S/4, nS+S/4) interval, can both be from

correctly extract m _i.

Experiment simulation

In experiment, use WindowsXP operating system and MATLAB7.0 as experiment simulation platform, select Lena, Baboon, Peppers etc. as test pattern, synchronizing sequence is to be 63 m sequence in the cycle, watermark sequence is that length is 1024 random series, in detail experimental result and be analyzed as follows described in:

Transparency test

Algorithm of the present invention uses the transparency of PSNR measure algorithm, is defined as:

$\mathrm{PSNR}=10\log \left[\frac{255^2}{\frac{1}{M}\underset{i=0}{\overset{M-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{[X(i,j)-X^{\&prime;}(i,j)]}^2}\right]---\left(8\right)$

Wherein, X is original image, and X' is the image after embed watermark.

Fig. 3 (a～d) is primary standard test pattern, Fig. 3 (e～h) under algorithm of the present invention containing watermarking images.From the experimental result of Fig. 3, algorithm of the present invention is 33.21dB containing the mean P SNR of watermarking images, meets invisibility requirement.

Robustness test

Table 1 has provided algorithm of the present invention and has processed the robustness under attacking at common image, and in order to compare the performance of algorithm of the present invention, table 1 has listed file names with the experimental result of prior art under same experimental conditions.As seen from the table, in algorithm of the present invention and prior art, algorithm is processed under attack and is all had stronger robustness common image, and the performance of algorithm of the present invention under the attacks such as Gaussian noise, JPEG compression, gaussian filtering is better than prior art simultaneously.

Table 1

Table 2 has been listed the ability of algorithm opposing geometric attack of the present invention, and for the ease of comparing, table 2 has listed file names with the experimental result of prior art under same experimental conditions.As seen from Table 2, no matter be single geometric attack or associating geometric attack, the performance of algorithm of the present invention is all better than documents algorithm.Because algorithm of the present invention is carrying out when watermark embeds having used synchronous code m sequence, effectively improve the performance that algorithm is resisted various geometric attacks.

Table 2

Extract test

The observing matrix that user key generates is depended in the security of algorithm of the present invention, and different user keys produces different gaussian random matrixes, and the watermark sequence extracting is not identical yet.For the security of verification algorithm, generate at random 1000 groups of user's detection key and carry out watermark extracting, and to arrange the 500th group be original user key, user key and extract related coefficient NC relation between watermark as shown in Figure 4.

Related coefficient NC is defined as follows:

$\mathrm{NC}=\frac{\underset{i=1}{\overset{L}{\mathrm{\&Sigma;}}}x\left(i\right)x^{\&prime;}\left(i\right)}{\underset{i=1}{\overset{L}{\mathrm{\&Sigma;}}}{\left[x\left(i\right)\right]}^2}---\left(9\right)$

In formula, x (i) and x'(i) be respectively the watermark sequence of original watermark sequence and extraction.Algorithm of the present invention, to user key sensitivity, is a kind of safe digital watermarking algorithm as seen from Figure 3.

The above, be only best mode for carrying out the invention, any be familiar with those skilled in the art the present invention disclose technical scope in, the simple change of the technical scheme that can obtain apparently or equivalence replace all fall within the scope of protection of the present invention.

Claims (2)

1. a motor synchronizing image watermark method that is applied to wireless multimedia sensor network, is characterized in that,

Comprise the following steps:

A watermark embeds

Original image I is carried out wavelet decomposition by a1, obtains sparse coefficient of wavelet decomposition matrix A ₁;

A2 is to the coefficient of wavelet decomposition A after sparse ₁be multiplied by the observing matrix that user key generates, obtain observing territory compressed signal A ₂;

A3 will observe territory compressed signal A ₂of one-dimensional, carries out watermark embedding according to following formula;

$A_i=\left\{\begin{array}{c}{A}_i-\mathrm{mod}(A_i,S)+3S/4m_i=1\\ A_i-\mathrm{mod}(A_i,S)+S/4m_i=-1\end{array}\right.$

mod(A _i,S)＝

Wherein

for downward bracket function, S is watermark embed strength;

A4) be reconstructed into 2 dimension matrixes by what obtain containing watermark sequence, adopt OMP algorithm to recover to obtain sparse signal A with watermarked information ₃;

A5) to A ₃carry out wavelet inverse transformation, obtain containing watermarking images I _w;

B. watermark extracting

Algorithm of the present invention does not need original image in the time carrying out watermark extracting, is a kind of blind watermarking algorithm, and watermark extraction process is as described below in detail:

B1) will be containing watermarking images I _wcarry out wavelet decomposition, obtain coefficient of wavelet decomposition sparse matrix B ₁;

B2) to the coefficient of wavelet decomposition B after sparse ₁be multiplied by observing matrix, obtain observing territory compressed signal B ₂;

B3) will observe territory compressed signal B ₂of one-dimensional, carries out watermark extracting according to following formula;

$w_i=\left\{\begin{array}{c}1,\mathrm{mod}(A_i^{*},S)\&GreaterEqual;S/2\\ -1,\mathrm{mod}(A_i^{*},S)<S/2\end{array}\right.$

B4) sequence { w from extracting _iin determine after synchronizing signal, then carry out the extraction of watermark signal,

Can obtain from above formula, in the time that Ai changes d,

d ∈ (nS-S/4, nS+S/4) interval, from

correctly extract m _i.

2. the motor synchronizing image watermark method that is applied to wireless multimedia sensor network according to claim 1, is characterized in that, adopts m sequence as synchronizing signal, establishes { a _nand { b _ntwo m sequences with same period T, a _n, b _n∈ { 1,1}, sequence { a _nand { b _nbetween cross correlation function be defined as

${\mathrm{\&rho;}}_{a,b}\left(\mathrm{\&tau;}\right)=\frac{1}{T}\underset{n=1}{\overset{T}{\mathrm{\&Sigma;}}}{a}_n{b}_{n-\mathrm{\&tau;}}$

M sequence { a _nautocorrelation function there is following character:

${\mathrm{\&rho;}}_{a,a}\left(\mathrm{\&tau;}\right)=\left(\begin{array}{c}1,\mathrm{\&tau;}=0\\ -\frac{1}{T},\mathrm{\&tau;}\&NotEqual;0\end{array}\right.$

If τ=0 o'clock, m sequence { a _nand { b _ncross-correlation coefficient

ρ _a,b(0)≥h/T

Wherein h is threshold value, gets odd number, and the implication of threshold value h is as infructescence { b _nonly have at most (T-h)/2 bit and { a _nnot identical, think { b _nit is a synchronizing signal.

Images