CN112767227A - Image watermarking method capable of resisting screen shooting - Google Patents

Abstract

The present invention provides an image watermarking method that is resistant to screen shooting, and solves the problem that the existing image watermarking method is not conducive to effective watermark extraction. The method includes a watermark embedding process and a watermark extraction process, and the watermark embedding process embeds the watermark into the host. Image, the watermark extraction process is based on the watermark embedding process. After the perspective correction is performed on the host image that has been embedded in the watermark and captured by the screen, the watermark extraction operation is performed to extract the embedded watermark information of the host image, which ensures the validity of the watermark extraction information. .

Description

An Image Watermarking Method Resistant to Screen Shooting

technical field

The present invention relates to the technical field of image authentication, and more particularly, to an image watermarking method that is resistant to screen shooting.

Background technique

Digital watermarking is a technology that embeds specific information into an information carrier, the embedded information is called a watermark, and the information carrier embedded with the watermark is called a watermark carrier. Copyright protection is achieved by embedding watermark information identifying the copyright owner in works such as digital images and electronic documents. When a copyright dispute occurs in a work in an open environment, the copyright owner can extract the watermark information from the pirated work as a certificate to identify the pirated copyright infringement. protect its own interests. Usually, for the extraction and detection of synchronization information watermark, a pseudo-random sequence for synchronization is also embedded, which is called template watermark.

Although the current image watermarking technology can achieve good results in the process of digital processing, in practical application scenarios, the watermarked image may not be a digital electronic image that can be directly obtained, but a photo obtained by using a smartphone to shoot the watermarked image. This is mainly because smartphones have the advantages of being easy to carry, fast in taking pictures, and flexible in taking pictures. However, because the watermark images captured by mobile phones are usually deformed, it is difficult to directly perform watermark detection on pictures captured by mobile phones. In practical applications, it is difficult to ensure that the angle taken by the mobile phone is parallel to the shooting plane, so perspective projection deformation will be introduced in the imaging process, which hinders the detection and extraction of watermarks, and even makes the detection and extraction of watermarks fail. Therefore, the embedded digital watermark must be confronted. Screen shots as well as robustness to geometric transformations.

等人在Signal ImageVideo Process中发表了“Toward an interactive poster using digital watermarkingand a mobile phone camera”的文章，文章中提出在纸质印刷品中嵌入水印，并利用手机拍摄提取水印信息的方法，嵌入算法中主要分为载体图像处理和秘密信息处理两个过程，其中，载体图像处理包括了维纳滤波、对图像和JND图像进行分块处理；秘密信息处理过程包括对信息进行汉明编码，将信息以4个比特长度进行拆分，4个比特表示一个角度，按照角度对原始伪随机图进行旋转，根据设计的水印嵌入公式将旋转后的伪随机图嵌入到图像块中，最终得到水印图像。在提取水印信息时，左上角的图像块作为同步块，对每个图像块计算自相关函数并检验自相关峰值的排列，估计角度提取水印信息。该文献提出的水印方法在打印拍摄场景下具有一定的可行性，但是对拍摄的距离有较大的限制，不利于水印的有效提取。2019年，H.Fang,W.Zhang,H.Zhou,H.Cui and N.Yu等人在IEEETrans.Inf.Forensics Security中发表“Screen-Shooting Resilient Watermarking”的文章，该文章中提出了一种抗屏摄水印算法，利用强度的尺度不变特征变换算法来定位特征区域，并在相关区域的离散余弦系数(DCT)中进行水印的嵌入和提取。该方法在应用的过程中给水印图像加上可见的特制边框，便于在屏幕拍摄时准确地定位到目标图像的位置信息，进行透视矫正，虽然具有较好的抗屏摄性能，但是该方法采用的加特制边框操作，在一定程度上影响了图像的美观而限制了算法的应用，并且边框容易遭受破坏，使得后续的水印提取失败。In 2012, A. Pramila, A. Keskinarkaus and T.

et al. published the article "Toward an interactive poster using digital watermarking and a mobile phone camera" in Signal ImageVideo Process. The article proposes to embed watermarks in paper prints and use mobile phone photography to extract watermark information. In the embedding algorithm, the main It is divided into two processes: carrier image processing and secret information processing. Among them, the carrier image processing includes Wiener filtering and block processing of images and JND images; The original pseudo-random image is rotated according to the angle, and the rotated pseudo-random image is embedded into the image block according to the designed watermark embedding formula, and finally the watermark image is obtained. When extracting the watermark information, the image block in the upper left corner is used as the synchronization block, and the autocorrelation function is calculated for each image block and the arrangement of the autocorrelation peak value is checked, and the watermark information is extracted by estimating the angle. The watermarking method proposed in this document has certain feasibility in printing and shooting scenarios, but has a large limitation on the shooting distance, which is not conducive to the effective extraction of watermarks. In 2019, H.Fang, W.Zhang, H.Zhou, H.Cui and N.Yu et al. published the article "Screen-Shooting Resilient Watermarking" in IEEETrans.Inf.Forensics Security, which proposed a The anti-screening watermarking algorithm uses the intensity scale-invariant feature transformation algorithm to locate the feature area, and embeds and extracts the watermark in the discrete cosine coefficient (DCT) of the relevant area. This method adds a visible special frame to the watermark image in the process of application, which is convenient for accurately locating the position information of the target image and performing perspective correction when shooting on the screen. Although it has good anti-screen shooting performance, this method adopts The operation of adding a special border to a certain extent affects the beauty of the image and limits the application of the algorithm, and the border is easily damaged, which makes the subsequent watermark extraction fail.

SUMMARY OF THE INVENTION

In order to solve the problem that the existing image watermarking method is not conducive to effective watermark extraction, the present invention proposes an image watermarking method that is resistant to screen shooting, has high reliability, and ensures the effectiveness of watermark extraction information.

In order to achieve above-mentioned technical effect, technical scheme of the present invention is as follows:

An image watermarking method that is resistant to screen shooting, the method includes a watermark embedding process and a watermark extraction process, the watermark embedding process embeds the watermark into a host image, and the watermark extraction process is based on the watermark embedding process. After the perspective correction of the host image captured on the screen is performed, a watermark extraction operation is performed to extract the embedded watermark information of the host image.

In this technical solution, an image watermarking method that is resistant to screen shooting is proposed. Aiming at the problem that the existing image watermarking method is not conducive to the effective extraction of watermarks, a matching watermark embedding process and watermark extraction process are proposed. When extracting the later watermark of the host image captured on the screen, perspective correction is performed on the host image, so that the watermark can still be detected after the watermark image is captured on the screen, which ensures the validity of the subsequent watermark extraction information.

Preferably, the watermark embedding process at least includes:

S1. use the key to generate a pseudo-random sequence and a template sequence, and combine the watermark bit information to generate a two-dimensional watermark matrix to be embedded;

S2. Determine the host image, perform two-dimensional discrete Fourier transform on the host image, translate the transformed DC component to the center of the discrete Fourier amplitude spectrum, embed a watermark in the upper half plane of the Fourier coefficient amplitude spectrum, and embed the points in the region The coordinates are mapped to log polar coordinates;

S3. The log-polar coordinates are used as the coordinates of the two-dimensional watermark matrix, and the same watermark bit information is embedded in the Fourier coefficients with the same log-polar coordinates;

S4. Since the Fourier amplitude spectrum has center symmetry, the corresponding coefficients of the upper half plane are copied to the lower half plane, and then the inverse Fourier transform is performed on the Fourier coefficients after embedding the watermark bit information to obtain the watermarked image.

Preferably, the watermark extraction process at least includes:

SA. Perform perspective correction on the host image captured by the screen and embedded with the watermark, and crop out the target image area as the image to be measured;

SB. Do a two-dimensional discrete Fourier transform on the image to be measured, translate the transformed DC component to the center of the discrete Fourier amplitude spectrum, take the center of the discrete Fourier amplitude spectrum as the origin of the rectangular coordinate system, and extract it from the upper half plane of the Fourier coefficient amplitude spectrum. Watermark, which maps the rectangular coordinates of the amplitude coefficients in the extraction range to the log polar coordinates;

SC. Calculate a two-dimensional Fourier coefficient magnitude matrix: average the magnitude coefficients that have the same log-polar coordinates after mapping, and use it as an element of the two-dimensional Fourier coefficient magnitude matrix;

SD. According to the principle of phase correlation, the original template and the Fourier coefficient amplitude matrix are quickly matched and calculated, and the position of the watermark matrix in the two-dimensional Fourier coefficient amplitude matrix is determined according to the coordinate position of the maximum correlation value, and the amplitude matrix synchronized with the embedded watermark is obtained. ;

SE. The amplitude matrix is despread and modulated with a pseudo-random sequence, and the watermark information is extracted.

Preferably, the process of generating the two-dimensional watermark matrix to be embedded in combination with the watermark bit information in step S1 is as follows:

S101. Let the key be key, and use the key key to generate a bipolar pseudo-random sequence g={g _i with length S _g ; 1≤i≤S _g , g _i ∈ {-1,1}} and the length is The template sequence of S _z Z={Z _i ; 1≤i≤S _z , Z _i ∈{-1,1}}, g _i and Z _i represent the i-th element of the bipolar pseudo-random sequence, the template sequence, respectively the ith element of ;

S102. Set the sequence of watermark bit information as b={b _i ; 1≤i≤S _b , b _i ∈ {0,1}}, perform spread spectrum modulation on each bit in sequence b: if b _i is "1", the spread spectrum modulation is the in-phase sequence of g, and the spread spectrum sequence W _i ={w _ij ; w _ij ∈{-1,+1}, 1≤j≤S _g }=+1×g; if b _i is "0", then the spread spectrum modulation is the inverse sequence of g, that is, W _i =-1×g, and the watermark information array W={W _i ; 1≤i≤S _b };

S103. Arrange the watermark information array W and the template sequence Z into a two-dimensional watermark matrix WZ={wz(h,c); 0≤h<H, 0≤c<C} with H rows and C columns in column order.

Preferably, the element wz(h,c) of the two-dimensional watermark matrix WZ={wz(h,c); 0≤h<H, 0≤c<C} consists of bipolar bits "1" and "-1" After the two-dimensional watermark matrix is generated, the template sequence Z is stored in the adjacent columns of the two-dimensional watermark matrix WZ in column order to form the original template Z _b , so that the spread spectrum sequence generated by the modulation of each meaningful watermark bit information occupies Adjacent columns of the two-dimensional watermark matrix WZ.

Preferably, in step S2, the process of determining the host image includes: when the original image is a color image, the original image is converted to the YCbCr color space, and the Y channel image is selected as the host image; when the original image is a grayscale image, no conversion is performed, directly as the host image.

Preferably, in step S2, the host image is subjected to l×l two-dimensional discrete Fourier transform, l=max(l _y , l _x ), _ly and l _x represent the length and width of the host image, respectively, and the transformed DC component Translate to the center of the discrete Fourier amplitude spectrum, the center of the discrete Fourier amplitude spectrum is used as the origin of the rectangular coordinate system, and the embedding area is located near the intermediate frequency where the normalized frequency value of the Fourier coefficient amplitude spectrum is f _n . The process of mapping coordinates to log-polar coordinates is: the rectangular coordinates (u, v) or polar coordinates (ρ, θ) of the Fourier coefficients of the embedded region are transformed into discrete log-polar coordinates (α, β), and the transformation formula is:

D= _fn ×l

a ^-H/2 ×D≤ρ<a ^H/2 ×D

θ表示傅立叶系数在极坐标中的角度，θ＝arctan(u/v)；在傅立叶系数的中低频区域嵌入水印，D取为傅立叶变换归一化频率值为f_n的中频附近，f_n的取值在0.15～0.25之间；Q表示保证α≥0的偏移常数；floor()函数表示向下取整函数，平衡水印的鲁棒性和不可见性。Among them, a represents a constant greater than 1 and close to 1, a=2 ^1/H , ρ represents the polar diameter, which represents the distance from the amplitude point to the center point corresponding to the Fourier amplitude spectrum,

θ represents the angle of the Fourier coefficient in polar coordinates, θ=arctan(u/v); the watermark is embedded in the mid-low frequency region of the Fourier coefficient, D is taken as the Fourier transform normalized frequency value near the intermediate frequency of f _n , f _n The value is between 0.15 and 0.25; Q represents the offset constant that guarantees α≥0; the floor() function represents the round-down function to balance the robustness and invisibility of the watermark.

Preferably, the polar diameter ρ satisfies:

a ^-H/2 ×D≤ρ<a ^H/2 ×D

The watermark embedding area in the corresponding rectangular coordinate system is a ring, and the coordinate range of discrete log polar coordinates (α, β) satisfies:

0≤α<H, 0≤β<C;

Embed the watermark according to the additive embedding formula or the multiplicative embedding formula, where the additive embedding formula is:

c(u,v)=c(u,v)+σ×wz(α,β)

The multiplicative embedding formula is:

c(u,v)=c(u,v)×(I+σ×wz(α,β));

When performing inverse Fourier transform on the Fourier coefficients after embedding the watermark bit information in step S4, when the original image is a color image, it is necessary to replace the Y channel of the YCbCr space with an image embedded with a watermark, and transform it into the RGB color space.

Here, embedding the watermark through the additive embedding formula or the multiplicative embedding formula does not require interpolation of the Fourier coefficients of the box, which eliminates the defect of image interpolation distortion. The same watermark can be embedded symmetrically about the symmetry of the DC component point in the center of the amplitude spectrum.

Preferably, the process of performing perspective correction on the host image captured by the screen and embedded with the watermark described in step SA is:

SA01. Set the host image captured by the screen and embedded with the watermark as I ₀ , convert I ₀ into a grayscale image, and perform Gaussian filtering on the grayscale image to obtain an image I ₁ ;

SA02. Adopt the Sobel algorithm to detect the edge of the image I ₁ to obtain the edge point image I ₂ ;

SA03. Carry out morphological closing operation on image I ₂ to obtain image I ₃ , delete connected components less than P from image I ₃ , generate binary image I ₄ , and P represents a pixel evaluation threshold of deleted connected components;

SA04. Set the target image as I', track the outer boundary area of the object in the binary image I ₄ , take the largest connected area as the contour of the target image I', calculate the minimum polygonal convex hull containing the contour of the image I', and obtain the polygonal convexity package's vertex collection points;

SA05. Calculate the edge straight line of the target image I': traverse the vertex set points of the polygonal convex hull in a counterclockwise order, determine a straight line for every two points, select the longest length of the two line segments with similar angles to calculate its parameter expression, and obtain A series of straight lines; when the number of straight lines is greater than 4, delete the lines with too large angle with the two adjacent straight lines;

SA06. Calculate the four vertices of the target image I': confirm the intersection of the four straight lines, and obtain the vertex set vecs of the target image I'; when the vertex coordinates are beyond the range of the host image I ₀ , the watermark image is incomplete through the screen shot, and the The content outside the range of the host image I ₀ is filled with 0;

SA07. Set the minimum side length of the target image I' as L _m , perform perspective correction on the host image I ₀ according to the vertex sets vecs and L _m , and cut out the transformed target image to obtain the image I _new .

Here, in step SA01, performing Gaussian filtering on the grayscale image can weaken the influence of moiré noise on subsequent operations. In step SA03, performing morphological closing operation on the image I ₂ can increase the continuity of the image edge, and P represents a For relatively small values, in step SA04, computing the minimum polygonal convex hull containing the contour of the image I' can reduce the dependence on the contour being complete and continuous.

Preferably, the image to be tested in step SB is the image I _new , and the image to be tested is subjected to a two-dimensional discrete Fourier transform of l′×l′, where l′=max( _{ly ′,l x ′} ), _ly ′ and l _x ′ are the length and width of the image to be tested, respectively, and the extraction area is located near the intermediate frequency with the normalized frequency f _n on the upper half plane of the Fourier coefficient amplitude spectrum. The extraction formula is:

a ^-λH ×D′≤ρ<a ^λH ×D′

D′=f _n ×l′

Among them, ρ is the radial range of the amplitude coefficients in the extraction area, λ is a parameter controlling the size of the extraction area, and the formula for mapping the rectangular coordinates of the amplitude coefficients in the extraction range to the log polar coordinates is:

where Q' represents an offset constant that guarantees α≥0.

Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

The invention proposes an image watermarking method that is resistant to screen shooting. Aiming at the problem that the existing image watermarking method is not conducive to the effective extraction of watermarks, a matching watermark embedding process and watermark extraction process are proposed. During the later watermark extraction of the host image, the perspective correction is performed on the host image, so that the watermark can still be detected after the watermark image is shot on the screen, which ensures the validity of the watermark extraction information.

Description of drawings

FIG. 1 shows a schematic diagram of an overall framework of an image watermarking method that is resistant to screen shooting proposed in an embodiment of the present invention;

Fig. 2 shows the flow chart of the watermark embedding process in the image watermarking method for resisting screen shooting proposed in the embodiment of the present invention;

3 shows a schematic diagram of the arrangement of the generated two-dimensional watermark matrix WZ proposed in the embodiment of the present invention;

FIG. 4 shows a flowchart of a watermark extraction process in a method for image watermarking resistant to screen shooting proposed in an embodiment of the present invention.

Detailed ways

The accompanying drawings are for illustrative purposes only, and should not be construed as limitations on this patent;

In order to better illustrate this embodiment, some parts of the drawings are omitted, enlarged or reduced, which do not represent the actual size;

For those skilled in the art, it is understandable that descriptions of certain well-known contents in the accompanying drawings may be omitted.

The positional relationship described in the accompanying drawings is only for exemplary illustration, and should not be construed as a limitation on this patent;

The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

Example 1

The present invention proposes an image watermarking method that is resistant to screen shooting, as shown in FIG. 1 as a whole. Referring to FIG. 1, the method includes a watermark embedding process and a watermark extraction process. The watermark embedding process embeds the watermark into the host image, and the watermark extraction process is based on the In the watermark embedding process described above, after perspective correction is performed on the host image that has been embedded in the watermark and captured by the screen, a watermark extraction operation is performed to extract the watermark information that has been embedded in the host image. During watermark extraction, perspective correction is performed on the host image, so that the watermark can still be detected after the watermark image is captured on the screen, which ensures the validity of the subsequent watermark extraction information.

Specifically, referring to FIG. 2, in this embodiment, the watermark embedding process includes:

S1. use the key to generate pseudo-random sequence and template sequence, and combine the watermark bit information to generate the two-dimensional watermark matrix to be embedded; wherein, the process of generating the two-dimensional watermark matrix to be embedded in combination with the watermark bit information is:

S101. Let the key be key, and use the key key to generate a bipolar pseudo-random sequence g={g _i with length S _g ; 1≤i≤S _g , g _i ∈ {-1,1}} and the length is The template sequence of S _z Z={Z _i ; 1≤i≤S _z , Z _i ∈{-1,1}}, g _i and Z _i represent the i-th element of the bipolar pseudo-random sequence, the template sequence, respectively The i-th element of , the value of the pseudo-random sequence g is ±1;

S102. Set the sequence of watermark bit information as b={b _i ; 1≤i≤S _b , b _i ∈ {0,1}}, perform spread spectrum modulation on each bit in sequence b: if b _i is "1", the spread spectrum modulation is the in-phase sequence of g, and the spread spectrum sequence W _i ={w _ij ; w _ij ∈{-1,+1}, 1≤j≤S _g }=+1×g; if b _i is "0", then the spread spectrum modulation is the inverse sequence of g, that is, W _i =-1×g, and the watermark information array W={W _i ; 1≤i≤S _b } is obtained, and each watermark bit information The S _g bit sequence obtained by spread spectrum modulation is arranged in two columns with the same length;

S103. Arrange the watermark information array W and the template sequence Z into a two-dimensional watermark matrix with H rows and C columns in column order WZ={wz(h,c); 0≤h<H, 0≤c<C}, template The length of the sequence Z satisfies: S _z =H×CS _b ×S _g , the template sequence Z is sorted by column, and the sequence modulated by the watermark bit information forms an H×C watermark matrix WZ, in the two-dimensional watermark matrix WZ= {wz(h,c); 0≤h<H,0≤c<C} element wz(h,c) consists of bipolar bits "1" and "-1", after the two-dimensional watermark matrix WZ is generated , the template sequence Z is stored in the adjacent columns of the two-dimensional watermark matrix WZ in column order to form the original template Z _b , so that the spread spectrum sequence generated by the modulation of each meaningful watermark bit information occupies several adjacent columns of the two-dimensional watermark matrix WZ column, see Figure 3 for details.

S2. Determine the host image, perform two-dimensional discrete Fourier transform on the host image, translate the transformed DC component to the center of the discrete Fourier amplitude spectrum, embed a watermark in the upper half plane of the Fourier coefficient amplitude spectrum, and embed the points in the region The coordinates are mapped to log polar coordinates; in practice, it is possible to be approximately uniform.

In this embodiment, the process of determining the host image includes: when the original image is a color image, converting the original image to the YCbCr color space, and selecting a Y channel image as the host image; when the original image is a grayscale image, no conversion is performed, and the as the host image.

The host image is subjected to l×l two-dimensional discrete Fourier transform, l=max(l _y ,l _x ), _ly and l _x represent the length and width of the host image, respectively, and the transformed DC component is translated into the discrete Fourier amplitude spectrum. The center, the center of the discrete Fourier amplitude spectrum is taken as the origin of the rectangular coordinate system, and the embedded area is located near the intermediate frequency where the normalized frequency value of the Fourier coefficient amplitude spectrum is f _n , in this embodiment, f _n =0.25, the The process of mapping the coordinates of the points in the embedded region to log polar coordinates is: transform the rectangular coordinates (u, v) or polar coordinates (ρ, θ) of the Fourier coefficients of the embedded region into discrete log polar coordinates (α, β) , the transformation formula is:

D= _fn ×l

a ^-H/2 ×D≤ρ<a ^H/2 ×D

θ表示傅立叶系数在极坐标中的角度，θ＝arctan(u/v)；在傅立叶系数的中低频区域嵌入水印，D取为傅立叶变换归一化频率值为f_n的中频附近，f_n的取值在0.15～0.25之间；Q表示保证α≥0的偏移常数；floor()函数表示向下取整函数，平衡水印的鲁棒性和不可见性。极径ρ满足：Among them, a represents a constant greater than 1 and close to 1, a=2 ^1/H , ρ represents the polar diameter, which represents the distance from the amplitude point to the center point corresponding to the Fourier amplitude spectrum,

θ represents the angle of the Fourier coefficient in polar coordinates, θ=arctan(u/v); the watermark is embedded in the mid-low frequency region of the Fourier coefficient, D is taken as the Fourier transform normalized frequency value near the intermediate frequency of f _n , f _n The value is between 0.15 and 0.25; Q represents the offset constant that guarantees α≥0; the floor() function represents the round-down function to balance the robustness and invisibility of the watermark. The polar diameter ρ satisfies:

a ^-H/2 ×D≤ρ<a ^H/2 ×D

The watermark embedding area in the corresponding rectangular coordinate system is a ring, and the coordinate range of discrete log polar coordinates (α, β) satisfies:

0≤α<H, 0≤β<C;

Embed the watermark according to the additive embedding formula or the multiplicative embedding formula, where the additive embedding formula is:

c(u,v)=c(u,v)+σ×wz(α,β)

The multiplicative embedding formula is:

c(u,v)=c(u,v)×(1+σ×wz(α,β));

Embedding the watermark by the additive embedding formula or the multiplicative embedding formula does not need to interpolate the Fourier coefficients of the box, which eliminates the defect of image interpolation and interpolation distortion. The same watermark can be embedded symmetrically at the symmetrical position of the central DC component point. In this embodiment, the watermark information bits wz(α, β) are embedded according to the multiplicative embedding formula.

S3. The log-polar coordinates are used as the coordinates of the two-dimensional watermark matrix, and the same watermark bit information is embedded in the Fourier coefficients with the same log-polar coordinates;

S4. Perform inverse Fourier transform on the Fourier coefficients after embedding the watermark bit information to obtain a watermarked image.

Referring to Figure 4, in this embodiment, the watermark extraction process includes:

SA. Perform perspective correction on the host image captured by the screen and embedded with the watermark, and crop out the target image area as the image to be measured; the specific perspective correction process includes:

SA01. Set the host image captured by the screen and embedded with the watermark as I ₀ , convert I ₀ into a grayscale image, and perform Gaussian filtering with a standard deviation of 5 on the grayscale image to obtain an image I ₁ , which can weaken the moire noise and affect the subsequent the impact of the operation;

SA02. Adopt the Sobel algorithm to detect the edge of the image I ₁ , and obtain the edge point image I ₂ ; During the specific implementation, the default threshold value threshold, adopt the Sobel algorithm again to detect the edge of the image I ₁ , and the threshold value can be taken as the threshold × 0.5, and obtain the edge point image I ₂ ;

SA03. Perform morphological closing operation on the image I ₂ , which can increase the continuity of the edge of the image to obtain the image I ₃ , delete the connected components smaller than P from the image I ₃ , and generate a binary image I ₄ , where P represents a deleted connected component The pixel evaluation threshold of the component; during the specific implementation, the structuring element object se required for the morphological closing operation can select a flat disc structuring element with a radius r of 5, and delete all connected components less than P pixels from the image _I3 (object), P can be taken as 100, and another binary image I ₄ is generated;

SA04. Set the target image as I', track the outer boundary area of the object in the binary image _I4 , take the largest connected area as the contour of the target image I', and calculate the minimum polygonal convex hull containing the contour of the image I', which can reduce the number of The complete and continuous dependency of the contour, the vertex set points of the polygonal convex hull are obtained;

SA05. Calculate the edge straight line of the target image I': traverse the vertex set points of the polygonal convex hull in counterclockwise order, determine a straight line for every two points, and select the longest length of the two line segments with similar angles to calculate its parameter expression, specifically , according to the angle A _g between the straight lines and the straight line length L _g to filter the straight lines, the two line segments with the included angle A _g less than M _g select the one with the largest length L _g to calculate its parameter expression, M _g can be taken as 10, and then get A series of straight lines; when the number of straight lines is greater than 4, delete the lines with too large angle with the two adjacent straight lines;

SA06. Calculate the four vertices of the target image I': confirm the intersection of the four straight lines, and obtain the vertex set vecs of the target image I'; when the vertex coordinates are beyond the range of the host image I ₀ , the watermark image is incomplete through the screen shot, and the The content outside the range of the host image I ₀ is filled with 0;

SA07. Set the minimum side length of the target image I' as L _m , perform perspective correction on the host image I ₀ according to the vertex sets vecs and L _m , and cut out the transformed target image to obtain the image I _new .

SB. Do a two-dimensional discrete Fourier transform on the image to be measured, translate the transformed DC component to the center of the discrete Fourier amplitude spectrum, take the center of the discrete Fourier amplitude spectrum as the origin of the rectangular coordinate system, and extract it from the upper half plane of the Fourier coefficient amplitude spectrum. Watermark, which maps the rectangular coordinates of the amplitude coefficients in the extraction range to the log polar coordinates;

The image to be tested is the image I _new , and the image to be tested is subjected to a two-dimensional discrete Fourier transform of l′×l′, where l′=max( _{ly ′, l x ′), and ly ′ and l x ′ are respectively} The length and width of the image to be measured, the extraction area is located near the intermediate frequency with the normalized frequency value f _n of the upper half plane of the Fourier coefficient amplitude spectrum, and the extraction formula is:

a ^-λH ×D′≤ρ<a ^λH ×D′

D′=f _n ×l′

Among them, ρ is the radial range of the amplitude coefficient of the extraction area, and λ is a parameter that controls the size of the extraction area. Considering the impact of image scaling attacks on the image frequency domain, the extraction area should be larger than the embedded area, and the value of λ is generally greater than 1/2. Here λ=1 is taken, and the formula for mapping the rectangular coordinates of the amplitude coefficients in the extraction range to the log polar coordinates is:

Among them, Q' represents an offset constant that guarantees α≥0, such as H'/2, generally H'>H, such as H'=2H; at this time, 0≤α<H', 0≤β< C;

SC. Calculate a two-dimensional Fourier coefficient magnitude matrix: average the magnitude coefficients that have the same log-polar coordinates after mapping, and use it as an element of the two-dimensional Fourier coefficient magnitude matrix;

Specifically, a two-dimensional Fourier coefficient magnitude matrix amp is established, and the Fourier coefficients with the same discrete log polar coordinates (α, β) (these coefficients are located in a sector in a rectangular coordinate system) are averaged to obtain amp(α, β), taking it as an element of amp, the two-dimensional Fourier coefficient amplitude matrix amp={amp(α,β)|0≤α<H′, 0≤β<C};

SD. According to the principle of phase correlation, the original template and the Fourier coefficient amplitude matrix are quickly matched and calculated, and the position of the watermark matrix in the two-dimensional Fourier coefficient amplitude matrix is determined according to the coordinate position of the maximum correlation value, and the amplitude matrix synchronized with the embedded watermark is obtained. ;

Since the rotation transformation of the image in the log polar coordinate domain of the Fourier amplitude spectrum of the image is represented as a circular translation in the angular (β direction) direction, the scaling transformation of the image is represented by a translation in the log polar diameter (ie the α direction) direction; therefore The original template Z _b and the amplitude matrix amp are used for fast matching calculation of phase correlation according to the correlation theorem. According to the coordinate position of the maximum correlation value, the position of the watermark matrix WZ in the DFT amplitude coefficient mean matrix amp is determined, and the amplitude synchronized with the embedded watermark WZ is obtained. matrix:

The original template Z _b is filled with 0 to form a matrix g(h, c) of the same size as the magnitude matrix amp, and the translation correlation value between them is:

0≤k＜H′, 0≤l＜C

According to the two-dimensional correlation theorem, that is, the correlation of two two-dimensional functions in the space domain and the product of their two-dimensional DFT frequency domain functions are mutual Fourier transform and inverse transform pairs, so there are:

Among them, "*" represents complex conjugate, AMP(h,c)=DFT(amp(h,c)), G(h,c)=DFT(g(h,c)), namely AMP(h,c) ) and G(h,c) are the two-dimensional Fourier transform coefficients of amp(h,c) and g(h,c) respectively, so the following formula can be used to quickly calculate the correlation value r(k,l):

r(k,l)=IDFT[AMP(h,c)G ^* (h,c)]

Shift phase correlation values can also be calculated

是AMP(h,c)的相角；in,

is the phase angle of AMP(h,c);

的最大值可以确定嵌入水印位置，得到与嵌入水印WZ同步的幅度矩阵

Since the original template matrix has a correlation with the magnitude matrix of the embedded template, the correlation value matrix r(k,l) or

The maximum value of can determine the embedded watermark position, and obtain the amplitude matrix synchronized with the embedded watermark WZ

SE. The amplitude matrix is despread and modulated with a pseudo-random sequence, and the watermark information is extracted.

矩阵进行解扩频调制，得到有意义的多比特水印信息；根据嵌入W_i时的位置和顺序从

中取出与对应的S_g个DFT系数，并组成一段长为S_g的序列C_i。按照解扩频原理，把C_i与长为S_g的原始伪随机序列g进行点乘运算得到C′_i，如果C′_i的正数平均值与负数平均值之和大于0，则判定提取的水印比特b′_i为“1”，否则判定提取的水印比特b′_i为“0”，解扩频之后得到二进制的有意义水印信息序列b′。Using the original pseudo-random sequence g pair

The matrix is despread and modulated to obtain meaningful multi-bit watermark information; according to the position and order of _embedding Wi from

Take out the corresponding S _g DFT coefficients from the , and form a sequence C _i with a length of S _g . According to the principle of despreading, the point product of C _i and the original pseudo-random sequence g of length S _g is performed to obtain C' _i . If the sum of the positive and negative average values of C' _i is greater than 0, it is determined that The watermark bit b' _i of , is "1", otherwise it is determined that the extracted watermark bit b' _i is "0", and a binary meaningful watermark information sequence b' is obtained after despreading.

The positional relationship described in the accompanying drawings is only for exemplary illustration, and should not be construed as a limitation on this patent;

Obviously, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those of ordinary skill in the art, changes or modifications in other different forms can also be made on the basis of the above description. There is no need and cannot be exhaustive of all implementations here. Any modification, equivalent replacement and improvement made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (10)

1. A watermark method of an image capable of resisting screen shooting is characterized by comprising a watermark embedding process and a watermark extracting process, wherein the watermark embedding process embeds a watermark into a host image, and the watermark extracting process is based on the watermark embedding process, and after perspective correction is carried out on the host image which is embedded with the watermark and shot by a screen, watermark extracting operation is carried out to extract watermark information embedded in the host image.

2. The screen-shot resistant image watermarking method of claim 1, wherein the watermark embedding process comprises at least:

s1, generating a pseudo-random sequence and a template sequence by using a secret key, and generating a two-dimensional watermark matrix to be embedded by combining watermark bit information;

s2, determining a host image, performing two-dimensional discrete Fourier transform on the host image, translating the transformed direct current component to the center of a discrete Fourier magnitude spectrum, embedding a watermark in the upper half plane of the Fourier coefficient magnitude spectrum, and mapping the coordinates of points in an embedding area to log-polar coordinates;

s3, taking the log-polar coordinates as the coordinates of the two-dimensional watermark matrix, and embedding the same watermark bit information into Fourier coefficients with the same log-polar coordinates;

and S4, because the Fourier magnitude spectrum has central symmetry, copying the corresponding coefficient of the upper half plane to the lower half plane, and then carrying out inverse Fourier transform on the Fourier coefficient embedded with the watermark bit information to obtain the image containing the watermark.

3. The screen-shot resistant image watermarking method of claim 2, wherein the watermark extraction process comprises at least:

SA, performing perspective correction on the host image which is shot by the screen and embedded with the watermark, and cutting out a target image area as an image to be detected;

SB. performing two-dimensional discrete Fourier transform on the image to be measured, translating the transformed DC component to the center of the discrete Fourier magnitude spectrum, taking the center of the discrete Fourier magnitude spectrum as the origin of a rectangular coordinate system, extracting a watermark in the upper half plane of the Fourier coefficient magnitude spectrum, and mapping the rectangular coordinate of the magnitude coefficient in the extraction range to a log polar coordinate;

SC. a two-dimensional fourier coefficient magnitude matrix is calculated: averaging the amplitude coefficients which have the same log-polar coordinates after mapping, and taking the average as an element of a two-dimensional Fourier coefficient amplitude matrix;

SD., according to the phase correlation principle, the original template and the Fourier coefficient amplitude matrix are subjected to fast matching calculation, and according to the coordinate position of the maximum correlation value, the position of the watermark matrix in the two-dimensional Fourier coefficient amplitude matrix is determined, so as to obtain the amplitude matrix synchronous with the embedded watermark;

SE. the amplitude matrix is despread and modulated by the pseudo-random sequence to extract the watermark information.

4. The screen-shot resistant image watermarking method according to claim 3, wherein the process of generating the two-dimensional watermark matrix to be embedded in combination with the watermark bit information in step S1 is:

s101, setting the key as a key, and generating the key with the length S_gIs given as a bipolar pseudo-random sequence g ═ g { (g) }_i；1≤i≤S_g,g_iE { -1,1} } and length S_zTemplate sequence of (Z) { Z ═ Z_i；1≤i≤S_z,Z_i∈{-1,1}}，g_i、Z_iRespectively representing the ith element of the bipolar pseudo-random sequence and the ith element of the template sequence;

s102, setting the sequence of the watermark bit information as b ═ b { (b)_i；1≤i≤S_b,b_iE {0,1} }, and performing spread spectrum modulation on each bit in the sequence b if b_iIs 1, the spread spectrum is modulated into the same phase sequence of g to obtain a spread spectrum sequence W_i＝{w_ij；w_ij∈{－1,+1},1≤j≤S_g+1 × g; if b is_iTo "0", the spread spectrum modulation is the inverse sequence of g, i.e., W_iGet watermark information array W ═ 1 Xg_i；1≤i≤S_b}；

S103, arranging the watermark information array W and the template sequence Z into a two-dimensional watermark matrix WZ (H, C) with H rows and C columns according to the column sequence; h is more than or equal to 0 and less than H, and C is more than or equal to 0 and less than C.

5. The screen-shot resistant image watermarking method according to claim 4, wherein the watermark is generated in a two-dimensional watermark matrix WZ ═ { WZ (h, c); h is not less than 0<H,0≤c<C is composed of bipolar bits '1' and '-1', after the two-dimensional watermark matrix is generated, the template sequence Z is stored in the adjacent column of the two-dimensional watermark matrix WZ according to the column sequence to form the original template Z_b。

6. The screen-shot resistant image watermarking method of claim 5, wherein in the step S2, the determining process of the host image comprises: when the original image is a color image, converting the original image into a YCbCr color space, and selecting a Y-channel image as a host image; when the original image is a gray image, the original image is directly used as a host image without conversion.

7. The screen shot resistant image watermarking method of claim 6, wherein in step S2, the host image is subjected to l x l two-dimensional discrete Fourier transform, l ═ max (l ═ max)_y,l_x)，l_yAnd l_xRespectively representing the length and width of a host image, translating the converted direct current component to the center of a discrete Fourier amplitude spectrum, taking the center of the discrete Fourier amplitude spectrum as the origin of a rectangular coordinate system, and setting the normalized frequency value f of the Fourier coefficient amplitude spectrum in an embedded area_nThe process of mapping the coordinates of the points in the embedding region onto the log-polar coordinates is as follows: the rectangular (u, v) or polar (ρ, θ) coordinates of the fourier coefficients of the embedding region are transformed into discrete log polar coordinates (α, β) by the transformation formula:

D＝f_n×l

a^-H/2×D≤ρ＜a^H/2×D

wherein a represents a constant greater than 1 and tending to 1, and a is 2^1/HAnd p represents the polar diameter, the distance from the amplitude point to the center point of the spectrum corresponding to the Fourier amplitude,

θ represents an angle of a fourier coefficient in polar coordinates, θ ═ arctan (u/v); in FuEmbedding watermark in the middle and low frequency region of the vertical leaf coefficient, taking D as Fourier transform normalized frequency value f_nNear the intermediate frequency of f_nThe value of (a) is between 0.15 and 0.25; q represents an offset constant that ensures α ≧ 0; the floor () function represents a floor function.

8. The screen-shoot-resistant image watermarking method as claimed in claim 7, wherein the polar diameter p satisfies:

a^-H/2×D≤ρ＜a^H/2×D

the watermark embedding area in the rectangular coordinate system is an annular shape, and the coordinate range of discrete logarithm polar coordinates (alpha, beta) meets the following requirements:

0≤α＜H，0≤β＜C；

embedding the watermark according to an additive embedding formula or a multiplicative embedding formula, wherein the additive embedding formula is as follows:

c(u，v)＝c(u，v)+σ×wz(α，β)

the multiplicative embedding formula is:

c(u，v)＝c(u，v)×(1+σ×wz(α，β))；

when the fourier coefficient after embedding the watermark bit information is subjected to inverse fourier transform in step S4, when the original image is a color image, the Y channel of the YCbCr space needs to be replaced with an image embedded with a watermark, and the image is converted into the RGB color space.

9. The screen-shot resistant image watermarking method according to claim 8, wherein the process of perspective correction of the screen-shot and watermarked host image in step SA is:

SA01. let the host image shot by the screen and embedded with the watermark be I₀Is shown by₀Converting into gray image, and performing Gaussian filtering on the gray image to obtain image I₁；

SA02. detecting the image I by adopting Sobel algorithm₁Edge point image I is obtained₂；

SA03. for image I₂Performing morphological closed operation to obtain image I₃From picture I₃Deleting connected components smaller than P to generate binary image I₄P represents a pixel evaluation threshold for a removed connected component;

SA04, setting the target image as I', and tracking the binary image I₄In the outer boundary region of the middle object, taking the maximum connected region as the contour of the target image I ', and calculating the minimum polygonal convex hull containing the contour of the image I' to obtain the vertex set points of the polygonal convex hull;

SA05, calculating edge straight lines of the target image I', namely traversing vertex sets points of a polygonal convex hull in a counterclockwise sequence, determining a straight line at every two points, and selecting two line segments with similar angles and calculating a parameter expression of the two line segments with the largest length to obtain a series of straight lines; when the number of the straight lines is more than 4, deleting the straight lines with overlarge included angles with the two adjacent straight lines;

SA06. calculating four vertices of the target image I': confirming the intersection point of the four straight lines to obtain a vertex set vecs of the target image I'; when the vertex coordinates exceed the host image I₀When the watermark image is not complete through screen shooting, the host image I is obtained₀Filling 0 in the content outside the range;

SA07, the minimum side length of the target image I' is set as L_mFrom vertex sets vecs and L_mFor host image I₀Performing perspective correction, and cutting out the transformed target image to obtain an image I_new。

10. The screen shot resistant image watermarking method of claim 9, wherein the image to be tested in step SB is image I_newThe image to be measured is subjected to a two-dimensional discrete fourier transform of l ' × l ', where l ' ═ max (l)_y′,l_x′)，l_y' and l_x' the length and the width of the image to be measured are respectively, the normalized frequency value of the upper half plane of the Fourier coefficient magnitude spectrum of the extraction area is f_nNear the intermediate frequency, the extraction formula is:

a^-λH×D′≤ρ<a^λH×D′

D′＝f_n×l′

wherein ρ is the radial range of the amplitude coefficient of the extraction region, λ is a parameter for controlling the size of the extraction region, and the formula for mapping the rectangular coordinate of the amplitude coefficient in the extraction region to the log-polar coordinate is as follows:

wherein Q' represents an offset constant that ensures α ≧ 0.

Images