CN103247018A - Expanding conversion shaking watermark modulating method based on logarithm domain visual model - Google Patents

Abstract

The invention provides an expanding conversion shaking watermark modulating method based on a logarithm domain visual model. The method comprises a watermark embedding step and a watermark detecting step, and comprises the concrete steps of firstly performing block dividing operation on a carrier image, extracting a coefficient vector of each to-be-embedded watermark from each sub block, performing logarithm domain conversion on the projection of each vector in random direction, computing the quantified step size corresponding to each coefficient vector according to the visual model of the logarithm domain at the same time, utilizing shaking modulating watermark to embed watermark information into each sub block, and finally reestablishing to obtain the image with the embedded watermark. During watermark detecting, the watermark information is extracted by the logarithm domain conversion and quantified step size according to the same method. The expanding conversion shaking watermark modulating method based on the logarithm domain visual model can greatly improve the robustness and the invisibility of the watermark.

Description

Method for expanding and transforming jitter and modulating watermark based on the log-domain vision mode

Technical field

The present invention relates to a kind of method for expanding and transforming jitter and modulating watermark based on the log-domain vision mode, belong to the multimedia signal processing technique field.

Background technology

Fast development along with multimedia messages treatment technology and computer networking technology, the storage of digital medium informations such as voice, image and video, copy and propagation becomes more and more easier, the legitimate rights and interests that this has greatly endangered the copyright owner, the problem of piracy that causes thus and dispute over copyright problem become the factor that influences social stability.Under this background, digital watermark technology arises at the historic moment, and as a kind of emerging copyright protection technology, is widely used in a plurality of fields such as copyright protection, issue tracking and authentication of digital product.

Digital watermark technology has been obtained very big development in recent years, Chen etc. have proposed quantization index modulation (QIM:Quantization Index Modulation) water mark method, as the blind watermarking algorithm of a quasi-representative, the significant advantage on algorithm application and computation complexity has obtained paying close attention to widely.Yet owing to adopted the embedding criterion of uniform quantization, the assailant can be by averaging operation at watermark signal, thereby be easy to estimate the watermark signal of embedding, therefore causes this method to the fragility of collusion attack.

P.Gonzalez etc. are subjected to the inspiration of Weber's law, propose watermark signal is embedded into the log-domain of carrier signal, and to passing through the embedding of the signal of number conversion being carried out watermark, the method has effectively solved the weakness of QIM at the fragility of collusion attack then.But at log-domain, along with reducing of carrier signal amplitude, quantization step reduces thereupon, has seriously reduced the robustness of the method.

Nima K.K etc. has proposed a kind of improved logarithm quantization function, and this method efficiently solves the problems referred to above, has improved the robustness of algorithm.But the method adopts fixed quantization step, therefore greatly reduces the method robustness that multiplication is attacked for brightness of image.And more seriously, improved logarithm transfer function can cause that the signal of embed watermark may fall between minus zone in the method, thereby causes the mistake that sign symbol is put upside down, and when not receiving malicious attack, the extraction mistake of watermark signal also can occur thus.

Summary of the invention

At the defective that exists in the prior art, the invention provides a kind of method for expanding and transforming jitter and modulating watermark based on the log-domain vision mode, this method proposes a kind of new logarithm quantization function, can when guaranteeing the watermark invisibility, significantly improve the robustness of watermark according to the quantization step of the adaptive adjustment watermark embedding of log-domain signal model.

The method for expanding and transforming jitter and modulating watermark of log-domain vision mode of the present invention, comprise to the projection of burst to be embedded and to number conversion the processing stage and by log-domain vision mode self-adaptation calculate quantization step the processing stage, the specific implementation step is:

(1) to the projection of burst to be embedded and to number conversion the processing stage, comprising:

A. from carrier, extract the coefficient of the watermark to be embedded of carrier transform domain, composition of vector x, the compute vector x projection x on random vector u ^TU.

B. construct new logarithm transfer function, the formula of employing is as follows

$y=F\left(x^{T}u\right)=\frac{\ln (1+\mathrm{\μ}\frac{x^{T}u}{C_0})}{\ln (1+\mathrm{\μ})}---\left(1\right)$

Wherein, y is the numerical value after changing, x ^TU is the projection of vector x on random vector u, and μ is key parameter, C ₀Mean flow rate for carrier image.

(2) by log-domain vision mode self-adaptation calculate quantization step the processing stage, comprising:

A. at the maximum invisible changes delta of log-domain vision mode structure vector x on random direction, adopt formula as follows:

$\mathrm{\&Delta;}=2s^{T}u\&times;\frac{x+({\mathrm{<figure-callout\; id="0"\; label="C"\; filenames="HDA00003217511300012.png"\; state="\{\{state\}\}">C</figure-callout>}}_0/\mathrm{\&mu;})}{({\mathrm{<figure-callout\; id="0"\; label="C"\; filenames="HDA00003217511300012.png"\; state="\{\{state\}\}">C</figure-callout>}}_0/\mathrm{\&mu;})}---\left(2\right)$

Wherein, s ^TU is the projection of visual redundancy vector s on random vector u.

B. construct the maximum invisible changing value Δ to the signal after the number conversion _y, the formula of employing is as follows:

${\mathrm{\&Delta;}}_y=\frac{\ln (1+2s^{T}u\&times;\frac{\mathrm{\&mu;}}{C_0})}{\ln (1+\mathrm{\&mu;})}---\left(3\right)$

Wherein, with Δ _yQuantization step as the watermark embedding.

Among the substep a of described step (1), what transform domain adopted is discrete cosine transform.

Among the substep a of described step (1), u is the random vector that is generated by key K, and length is identical with x.

Among the substep a of described step (1), the vector of pending log-domain conversion is the projection of DCT coefficient vector on the random direction that certain key generates of watermark to be embedded.

Among the substep b of described step (1), parameter μ need satisfy condition:

Guarantee that the numerical value of log-transformation is in positive interval.

Among the substep a of described step (2), visual redundancy vector s shelters structure according to the brightness in the Watson vision mode of revising, and obtains the maximum invisible changes delta of vector x on random direction at the log-domain correction with this.

Among the substep b of described step (2), the Δ and the formula that obtain among the substep a according to step (2)

) obtain the maximum invisible changes delta of signal y _y

Description of drawings

Fig. 1 is the schematic block diagram of watermark embedding of the present invention and detection method.

Fig. 2 the present invention is directed to the simulation comparison experiment that the brightness of image multiplication is attacked.

Fig. 3 the present invention is directed to the simulation comparison experiment that Gaussian noise is attacked.

Fig. 4 the present invention is directed to the simulation comparison experiment that salt-pepper noise is attacked.

Fig. 5 is the simulation comparison experiment that the present invention is directed to the JPEG compression attack.

Embodiment

Fig. 1 has provided the expanding and transforming jitter and modulating watermark embedding of log-domain vision mode of the present invention and the principle steps that detects.Following l-G simulation test is an object lesson of the inventive method, can verify the performance of the inventive method by this example.One 256 * 256 image is used in emulation, and it is embedded one 32 * 32 binary bitmap.Concrete watermark embed process is as follows:

(1) image is carried out 8 * 8 piecemeal DCT(discrete cosine transforms) conversion, each piecemeal DCT coefficient is carried out zig-zag scanning, and choose the AC(interchange of 3-10 position) coefficient composition vector x, the corresponding vector of each 8 * 8 piecemeal amounts to 1024.By the random vector u of key K generation with the vector x equal length.

(2) shelter the visual redundancy vector s that formula calculates each piecemeal coefficient correspondence to be embedded, the projection x of compute vector x and visual redundancy vector s on random vector u respectively according to the brightness in the Watson vision mode of revising ^TU and s ^TU.

(3) calculate the mean flow rate C of carrier image ₀, choose qualified parameter μ according to formula (4), wherein parameter μ saves as key.By formula (1), with the projection x of each piecemeal correspondence ^TU is converted to y.

(4) calculate watermark according to formula (3) and embed required quantization step λ * Δ _y, wherein parameter lambda is as the intensity of intensity factor control watermark embedding, and wherein parameter lambda saves as key.

(5) according to calculating quantization step λ * Δ _y, each piece calculated among the y by jitter modulation embeds a watermark information, treat all piecemeals embed finish after, replace original coefficient with the DCT coefficient after changing, and by inverse dct transform, obtain the image of embed watermark.

Image behind the embed watermark in the process of transmission, may experience the operation that some signal is handled, such as brightness enhancing, Gaussian noise, salt-pepper noise and JPEG compression etc., experiment shows that the watermark of embedding still can quite good detecting be come out, shown in Fig. 2 to 5.Concrete testing process is as follows:

(1) treat detected image and carry out 8 * 8 piecemeal dct transforms, each piecemeal DCT coefficient is carried out zig-zag scanning, and choose the AC(interchange of 3-10 position) the coefficient composition of vector

The corresponding vector of each 8 * 8 piecemeal amounts to 1024.Generated and vector by key K

The random vector of equal length

(2) shelter the visual redundancy vector that formula calculates each piecemeal coefficient correspondence to be embedded according to the brightness in the Watson vision mode of revising

The difference compute vector

With the visual redundancy vector

Projection on random vector u

With

(3) calculate the mean flow rate of carrier image

The key parameter μ that the input watermark embedding stage preserves is by formula (1), with the projection of each piecemeal correspondence

Be converted to

(4) the key parameter λ of input watermark embedding stage preservation calculates the required quantization step of watermark extracting according to formula (3)

In each piecemeal, go out watermark information according to dithering modulation of watermarking detecting device independent detection.After treating that all piecemeals extractions are finished, detected watermark sequence is reassembled into 32 * 32 watermarking images.

In order to show the robustness of water mark method among the present invention, adopt common image processing method formula that watermarking images is handled, comprise the brightness of image multiplication attack under the various different brightness multiplication factors, average is that the different Gaussian noise of 0 standard deviation is attacked, salt-pepper noise is attacked under the different densities, the JPEG compression attack under the different JPEG compression qualities.Simultaneously for purpose relatively, provided the test result of the present invention's (abbreviating LSTDM-WM as) with several correlation techniques, comprise LQIM method (list of references Nima Khademi Kalantari and Seyed Mohammad Ahadi, " A logarithmic quantization index modulation for perceptually better data hiding; " IEEE Transaction on Image Processing, vol.19, no.6, pp.1504 – 1517, Jun.2010.), STDM-WM method (list of references Li Q and Cox I.J, " Improved spread transform dither modulation by using a perceptual model:robustness to amplitude scaling and jpeg compression; " Proceedings of IEEEICASSP., pp.185 – 188, Apr.2007.), and STDM-AdpWM (list of references Liong Ma, " Adaptive spread-transform modulation using a new perceptual model for color image; " IEICE Trans.Info.Sys., vol.E93-D, no.4, pp.843 – 856, Apr.2010.).In all emulation, by adjusting watermark embed strength with the unified 42dB that is taken as of PSNR, accompanying drawing 2 to 5 has provided above method is extracted the bit error rate of watermark under various attack test result contrast.

Claims (7)

1. An extended transformation dither modulation watermarking method for a logarithmic domain visual model, including the processing stages of projection and logarithmic conversion of the signal sequence to be embedded and the processing stage of adaptively calculating the quantization step size through the logarithmic domain visual model, specifically implemented The steps are: (1) The processing stage of projection and logarithmic transformation of the signal sequence to be embedded, including: a. Extract the coefficients of the watermark to be embedded in the transform domain of the carrier from the carrier, form a vector x, and calculate the projection x ^T u of the vector x on the random vector u; b. Construct a new logarithmic conversion function, the formula used is: $\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; ln</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; u</span>}}{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; ln</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span>}$ Among them, y is the converted value, x ^T u is the projection of the vector x on the random vector u, μ is the key parameter, and C ₀ is the average brightness of the carrier image; (2) The processing stage of adaptively calculating the quantization step size through the logarithmic domain visual model, including: a. Construct the maximum invisible change Δ of the vector x in a random direction for the logarithmic domain visual model, using the following formula: $\mathrm{<span\; class="notranslate">\; \&Delta;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; \&times;</span>\frac{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span>}$ Among them, s ^T u is the projection of the visual redundancy vector s on the random vector u; b. Construct the maximum invisible change value Δ _y of the logarithmically transformed signal, using the following formula: ${\mathrm{<span\; class="notranslate">\; \&Delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; ln</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; \&times;</span>\frac{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; ln</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span>}$ Among them, _Δy is used as the quantization step of watermark embedding. 2 . The extended transform dither modulation watermarking method for a logarithmic domain visual model according to claim 1 , wherein in sub-step a of the step (1), discrete cosine transform is used in the transform domain. 3 . 3. The extended transformation dithering modulation watermarking method of the logarithmic domain visual model as claimed in claim 1, characterized in that: in the sub-step a of the step (1), u is a random vector generated by the key K, and the length Same as x. 4. The extended transformation dithering modulation watermarking method of the logarithmic domain visual model as claimed in claim 1, characterized in that: in the sub-step a of the step (1), the vector to be transformed into the logarithmic domain is the watermark to be embedded The projection of a vector of DCT coefficients in a random direction generated by a certain key. 5. The extended transformation dithering modulation watermarking method of the logarithmic domain visual model as claimed in claim 1, characterized in that: in the sub-step b of the step (1), the parameter μ needs to meet the conditions:

Ensure that the values of the logarithmic transformation are in the positive interval. 6. The extended transformation dithering modulation watermarking method of the logarithmic domain visual model as claimed in claim 1, characterized in that: in the sub-step a of the step (2), the visual redundancy vector s is based on the revised Watson visual model Constructed by the luminance mask in , to obtain the maximum invisible change Δ in a random direction of the vector x corrected for the logarithmic domain. 7. The extended transformation dithering modulation watermarking method of the logarithmic domain visual model as claimed in claim 1, characterized in that: in the sub-step b of the step (2), according to the obtained in the sub-step a of the step (2) Δ and formula

get) the maximum invisible change Δ _y to the signal y.

Images