CN103366337A - Digital watermarking method capable of being applied to 2D-3D (Two Dimensional-Three Dimensional) conversion - Google Patents

Abstract

The invention provides a Contourlet-conversion-based digital watermarking method capable of being applied to 2D-3D (Two Dimensional-Three Dimensional) conversion. The digital watermarking method comprises the following steps: firstly, carrying out the Contourlet conversion on a central image to obtain a converted low-frequency coefficient; then embedding a watermark into the low-frequency coefficient, wherein the embedding of watermark information is carried out by adopting a blind watermark embedding algorithm so that damages of the DIBR (Depth Image Based Rendering) process which is similar to horizontal movement to the watermark information can be effectively reduced, and a pick-up algorithm corresponding to the blind watermark method in the embedding process can be used for realizing the blind pick-up of the watermark.

Description

The digital watermark method that can be used for the 2D-3D conversion

Technical field

The present invention relates to a kind of digital watermark method of the 2D-3D of can be used for conversion, belong to the multimedia signal processing technique field.

Background technology

Along with the develop rapidly of infotech, 3D rendering and video technique are causing increasing concern.At present, 3D rendering and video resource are few, become this technology of restriction and realize widespread use and a key factor of popularizing fast.The obtain manner of 3D image mainly contains two kinds: the one, directly take the 3D image, and the 2nd, utilize 2D-3D switch technology generating three-dimensional image on the basis of bidimensional image.When directly taking the 3D image, need to use two video cameras to make up shooting, not only equipment is huge and expensive, and complicated operation, need photographer to possess specialty superb camera work and rich experience, so the cost that 3D takes is very high.And by the 2D-3D switch technology, need not again to take, just can obtain the 3D image by the 2D image by technical finesse.At present, the progressive highly significant of 2D-3D switch technology, progressively reaching in real time, automatic full HD video conversion with 2D TV programme or the shooting of 2D video camera is the level of the 3D image of nature.The 2D-3D switch technology significantly reduces the cost of manufacture of 3D image, can take full advantage of again existing abundant 2D image resource, solves rapidly not enough this problem of 3D content, therefore becomes the at present main mode that adopts.

When 3D image display technology is occupying the increasing market share, the Copyright Protection of 3D image resource also will become and become more and more important.At present, the research of this respect just just begins.Research and the application of digital watermark technology aspect the copyright protection of two dimensional image and video is gradually ripe; but because 3D rendering/video is based on that the binocular parallax principle is taken or and obtain through a series of variations; compare very large different of existence from common two dimensional image/video, the digital watermark technology of therefore using at present comparatively general two dimensional image can not be applied directly to wherein.

The 2D-3D switch technology is to utilize the binocular parallax principle, on the basis of original two dimensional image and depth map, by certain rendering technique (such as, based on the rendering technique (DIBR) of depth image), obtain left-eye view and right-eye view.When beholder's right and left eyes received respectively the right and left eyes view, people's brain will be fused into 3-D view with these views, thereby formed stereoscopic sensation.For the needs of copyright protection, we wish to work out a kind of digital watermark method behind the embed watermark, can both extract watermark in the left and right sides view that obtains and the centre view after through the 2D-3D conversion in original two dimensional image (being Centered Graphs).That is to say that the method has robustness for the conversion process of 2D-3D conversion, namely watermark information is not damaged in conversion process.

Summary of the invention

Change for 2D-3D in the process of the digital watermark method with robustness in searching, we find that the Contourlet conversion is a kind of translation invariant conversion.Because the right and left eyes view can regard that Centered Graphs moves horizontally as and obtain that therefore, we have proposed a kind of digital watermark method that can be used for the 2D-3D conversion based on the Contourlet conversion.

The technical solution used in the present invention is:

A kind of digital watermark method for the 2D-3D conversion, it is characterized in that: the method is based on the Contourlet conversion, and concrete steps are:

At the watermark built-in end:

(1) original center image is carried out the Contourlet conversion, obtain corresponding low frequency coefficient matrix;

(2) character string information that will input arbitrarily is as watermark information, and by given specific key K watermark information carried out scramble, obtains the watermark information w behind the scramble _i

(3) with w _iAs a column information, and this column information is listed as to copy obtains N row watermark information, N is the columns of step (1) medium and low frequency matrix of coefficients;

(4) adopt blind watermark embedding method, N row watermark information obtained in the previous step is embedded in the low frequency coefficient matrix;

The low frequency coefficient matrix of the embedding low frequency watermark information that (5) step (4) is obtained carries out the Contourlet inverse transformation, obtains the center image with watermark information;

(6) center image that step (5) is obtained is carried out the DIBR conversion, obtains having the right and left eyes virtual image of certain parallax;

At the watermark detection end:

(1) at first obtains image at receiving end, then image is carried out the Contourlet conversion, obtain low frequency coefficient matrix corresponding to this image;

The corresponding extracting method of blind watermark method when (2) adopting with embedding carries out the extraction of watermark information;

(3) watermark information that extracts is carried out anti-string processing, obtain corresponding character string information, thus the existence of checking watermark information.

Because the translation invariance of Contourlet conversion, so that it can effectively tackle the impact of the transfer pair watermark of the DIBR that level of approximation moves, the corresponding extracting method of blind watermark method when adopting simultaneously with embedding can be realized the blind extraction of watermark.Experimental result shows that the method that the present invention proposes has preferably detectability and stronger robustness.

Description of drawings

Fig. 1 is the schematic block diagram that the present invention can be used for the digital watermark method of 2D-3D conversion;

(a) of Fig. 2 and (b) be respectively original center image and corresponding depth map;

Fig. 3 is the center image after the embed watermark;

(a) of Fig. 4 and the left-and right-eye images with watermark information of image through obtaining after the DIBR conversion that (b) is respectively embed watermark;

Fig. 5 is the PSNR value with the center image of watermark information corresponding to different quantization step;

(a) of Fig. 6 and (b) are respectively through the PSNR value with the left-and right-eye images of watermark information corresponding to the different quantization step after the DIBR conversion;

(a) of Fig. 7 and (b) be respectively the normalized correlation coefficient NC(normalized correlation coefficient of the extraction watermark of left-and right-eye images) value;

(a) of Fig. 8 is respectively with (b) that left-and right-eye images extracts respectively the NC value of watermark under the different parallax conditions;

(a) of Fig. 9 and (b) be respectively by the left-eye image conversion and obtain eye image and obtained the NC value of watermark that left-eye image is extracted by the eye image conversion;

(a) of Figure 10 and (b) carry out respectively the NC value extracted after ten kinds of attacks for left-and right-eye images, the corresponding ten kinds of attacks of numeral 1 to 10 among the figure on the transverse axis are respectively: 1, Gaussian noise, 2, salt-pepper noise, 3, gaussian filtering, 4, medium filtering, 5, compressibility factor is 80 JPEG compression, 6, compressibility factor is 50 JPEG compression, 7, compressibility factor is 10 JPEG compression, 8, histogram equalization, 9, amplify 4 times of operations, 10, dwindle 1/4 operation.

Embodiment

The digital watermark method that can be used for the 2D-3D conversion that the present invention proposes may further comprise the steps:

The telescopiny of watermark:

(1) original image Y carries out the Contourlet conversion that turriform progression is (0,2,3,4), obtains the low frequency coefficient Matrix C of its conversion, and matrix size is M * N;

(2) with length be the character string information of m as watermark information, m＜M wherein, given key K is carried out scramble to w and is obtained character string information w _i

(3) in order to tackle the DIBR process, with w _iAs a column information, and this column information is listed as to copy obtains N row watermark information.These N row watermark informations are embedded in the low frequency coefficient Matrix C, and then obtain having the low frequency coefficient Matrix C of the Contourlet conversion of watermark information _w, the telescopiny of watermark adopts blind watermarking algorithm.

(4) with C _wCarry out the Contourlet inverse transformation, obtain having the original image of watermark information.

The telescopiny of watermark adopts blind watermarking algorithm, at first definition, and coefficient corresponding to low frequency Matrix C is p, the watermark information that this coefficient is corresponding is w(0 or 1), the low frequency coefficient after the embed watermark is p ', it is as follows to embed concrete steps: definition d ₀=-step/2+step * randnum is then according to d ₀Size definition d ₁, wherein step is the step-length of projection, randnum is [0,1] upper equally distributed random number.Definition d ₁Method as follows:

if?d ₀＜0

d ₁＝d ₀+step/2;

else

d ₁＝d ₀-step/2;

end

p′＝round(p+d(w+1)/step)×step-d(w+1)；

D (1)=d wherein ₀; D (2)=d ₁

The DIBR conversion process is mainly used 3D warping equation

$x_l=x_c+\frac{t_x}{2}\×\frac{f}{Z};x_r=x_c-\frac{t_x}{2}\×\frac{f}{Z}$

Wherein, x _cx _lx _rThe pixel value that represents respectively center image and DIBR conversion right and left eyes image representative afterwards; t _xThe expression baseline, 5% of general Selection Center picture traverse; F represents focal length; Z represents the pixel value of the depth map that center image is corresponding.

The leaching process of watermark:

To can obtain having afterwards with the center image of watermark information the right and left eyes virtual image of watermark information through the DIBR conversion, because the leaching process of the watermark of carrying out at the right and left eyes virtual image is the same, so only specifically describes the process steps of left-eye image being carried out watermark extracting below us.

(1) image of left eye carried out the Contourlet conversion that turriform progression is (0,2,3,4), obtain the low frequency coefficient Matrix C of its conversion _Wol, matrix size is M * N;

(2) to Matrix C _WolCarry out the extraction of watermark;

In the concrete leaching process of watermark, definition p " represents C _WolInterior low frequency coefficient, the watermark information that w ' representative is extracted, concrete steps are as follows:

Definition d ₀=-step/2+step * randnum, wherein randnum is [0,1] upper equally distributed random number, step is quantization step, then according to d ₀Size definition d ₁d ₁Define method as follows:

if?d ₀＜0

d ₁＝d ₀+step/2;

else

d ₁＝d ₀-step/2;

end

m ₀＝(p″-s ₀) ²;m ₁＝(p″-s ₁) ²;

Wherein, s ₀=round (p "+d ₀/ step) * step-d ₀;

s ₁＝round(p″+d ₁/step)×step-d ₁;

According to the m that obtains ₀And m ₁, the method for trying to achieve w ' is:

if?m ₀＜m ₁

w′＝0;

else

w′＝1;

end

For check the inventive method, adopt as a comparison method of block-based non-blind Detecting digital watermark method, and the present invention's realization is the blind extraction of digital watermarking.The matlabR2008a programmed environment is adopted in experiment, original image is the image of 720*576, watermark is input character arbitrarily, algorithm adopts the image PSNR of embed watermark to judge its picture quality, if the PSNR value is higher, image and original image that embed watermark then is described are more approaching, can determine the invisibility of watermark.Adopt normalization coefficient NC that quantitative evaluation is carried out in the watermark of extraction and the similarity of original watermark, but it has characterized the detectability of watermark, establishing host image is f (x, y), and under fire image is g (x, y), and establishing former watermark is w _i, the extraction watermark is w _i', shown in then PSNR and NC are defined as follows:

$\mathrm{PSNR}=10\lg \frac{{f(x,y)}_{\mathrm{<figure-callout\; id="2"\; label="max"\; filenames="HDA00003627661600031.png,HDA00003627661600033.png"\; state="\{\{state\}\}">max</figure-callout>}}^2}{\mathrm{\&Sigma;\&Sigma;}{(f(x,y)-g(x,y))}^2};\mathrm{NC}=\frac{\underset{i}{\mathrm{\&Sigma;}}{w}_i{w_i}^{\&prime;}}{\underset{i}{\mathrm{\&Sigma;}}{\left[w_i\right]}^2}$

We have carried out the calculating of PSNR value of the center image of embed watermark in the experiment, Fig. 1 we utilized center image and corresponding depth map thereof to carry out the DIBR conversion, with obtain with the left-and right-eye images of watermark information with not with the left-and right-eye images of watermark information and calculate respectively their PSNR value, and on left-and right-eye images, carry out respectively the extraction of watermark, we have changed parallax such as Fig. 8 simultaneously, obtained different parallaxes to the impact of watermarking inset and distill, and ten kinds of attacks such as Figure 10 impact for watermark information is provided.

In addition, except being generated by center image the right and left eyes image, we have also carried out being obtained eye image and being carried out the extraction of watermark and obtained left-eye image and carried out the extraction of watermark by the eye image conversion by the left-eye image conversion such as Fig. 9.Can be found out by experimental result Fig. 7,9, increase along with step-length, although it is more and more higher to extract the accuracy of watermark, but watermark also can become large to the impact of carrier image thereupon, and then may cause the visually variation of picture quality, therefore rational step-length has larger impact to invisibility and the detectability of watermark.The Transformation Graphs 8 of parallax also can produce certain impact to the extraction of watermark, the quality that generates the right and left eyes virtual image by center image when parallax becomes large is variation thereupon, and therefore reasonably parallax value is conducive to the process of 2D-3D conversion and watermarking is produced preferably impact.Simultaneously, we have also carried out the experiment of various attacks, can be found out by experimental result, and the method that the present invention proposes has preferably robustness and detectability, are a kind of digital watermark methods that can be used for preferably the 2D-3D conversion.

Claims (1)

1. one kind can be used for the digital watermark method that 2D-3D changes, and it is characterized in that: the method is based on the Contourlet conversion, and concrete steps are:

At the watermark built-in end:

(1) original center image is carried out the Contourlet conversion, obtain corresponding low frequency coefficient matrix;

(2) character string information that will input arbitrarily is as watermark information, and by given specific key K watermark information carried out scramble, obtains the watermark information behind the scramble;

(3) with w _iAs a column information, and this column information is listed as to copy obtains N row watermark information, N is the columns of step (1) medium and low frequency matrix of coefficients;

(4) adopt blind watermark embedding method, N row watermark information obtained in the previous step is embedded in the low frequency coefficient matrix;

The low frequency coefficient matrix of the embedding low frequency watermark information that (5) step (4) is obtained carries out the Contourlet inverse transformation, obtains the center image with watermark information;

(6) center image that step (5) is obtained is carried out the DIBR conversion, obtains having the right and left eyes virtual image of certain parallax;

At the watermark detection end:

(1) at first obtains image at receiving end, then image is carried out the Contourlet conversion, obtain low frequency coefficient matrix corresponding to this image;

The corresponding extracting method of blind watermark method when (2) adopting with embedding carries out the extraction of watermark information;

(3) watermark information that extracts is carried out anti-string processing, obtain corresponding character string information, thus the existence of checking watermark information.

Images