CN113658030A

CN113658030A - Low false alarm zero watermark algorithm based on regional XOR

Info

Publication number: CN113658030A
Application number: CN202110948460.7A
Authority: CN
Inventors: 胡森; 吴德阳; 王苗苗; 胡超; 曲长波
Original assignee: Liaoning Technical University
Current assignee: Liaoning Technical University
Priority date: 2021-08-18
Filing date: 2021-08-18
Publication date: 2021-11-16

Abstract

The invention discloses a low false alarm zero watermark algorithm based on regional XOR, which enlarges the difference between similar characteristics by redundantly constructing the characteristics and watermarks, quantizing three values and indexing the characteristics to perform regional XOR so as to solve the false alarm problem of image zero watermarks and further realize the copyright protection of digital information. The method comprises the steps of extracting frequency domain characteristics and space domain characteristics of an image by using two characteristic extraction methods, constructing the frequency domain characteristics into an index matrix, dividing a copyright watermark of a redundant structure into two regions according to the index matrix, performing exclusive OR with the space domain characteristics and the chaotic matrix of the redundant structure respectively, combining the regions to obtain a zero watermark, obtaining a copyright watermark image of the redundant structure by using the same characteristic extraction and region exclusive OR method in the copyright authentication process, performing redundancy removal operation, and quantizing the image into a three-value matrix to realize copyright authentication of distinguishable similar images.

Description

Low false alarm zero watermark algorithm based on regional XOR

Technical Field

The invention belongs to the technical field of digital image processing, information security and copyright protection, and particularly relates to a low false alarm zero watermark algorithm based on regional exclusive OR.

Background

With the development of the information age, digital works in various forms can be easily acquired in a network, and meanwhile, the copyright problem also occurs. The method is characterized in that information is embedded in an image to inevitably affect the image quality, and the contradiction between robustness and invisibility exists, the problem is solved by the proposal of zero watermark [ hot spring, Sun 38188;,. Feng, King Tree, zero watermark concept and application [ J ]. electronic bulletin 2003,31(02):214-216 ], a characteristic sequence with uniqueness and robustness is extracted from the image and directly used as the zero watermark or used as the zero watermark after being linked with a copyright identifier, and is stored in an authentication center to finish the registration of the image copyright. The zero watermark generation process can be mainly divided into two parts: 1) image feature extraction, 2) establishing connection between the features and the copyright (for example, generating zero watermark by XOR between the features and the watermark). Evaluation indexes such as robustness, operation efficiency, false alarm rate and the like of the zero-watermark algorithm are mainly influenced by the image feature extraction part; after extracting features from the image, establishing a connection with the copyright identification to generate a zero watermark, and obtaining the copyright identification from the zero watermark and the image features in the copyright authentication process to realize copyright visualization; in the process of establishing the relationship between the features and the copyright, an encryption algorithm is added to improve the security of the zero watermark.

In the current image zero watermark research, image feature extraction methods can be mainly classified into the following methods. Based on frequency domain transformations, such as Shift-Invariant shear wave transformation (SIST) [ Shi S, Luo T, Huang J, et al. A Novel HDR Image Zero-wavelet Based on Shift-inverse shear wave transformation [ J ]. Securability and Communication networks.2021,2021(8655):1-12 ], Gabor transformation [ Fan D, Li Y, Gao S, et al. A Novel Zero wavelet transformation Based on wavelet transformation and distribution co-Transform [ J ]. Concurrentand Computation, n/a transformation and experiment, n/a transformation, III/C.89, growing of wavelet Transform, C.70. C.C., DTCP [ Jing L, Jingbingg L, Jixin M, et al. A routing Multi-watermark Based on DTCP-DCT and Henon Map [ J ]. Applied sciences.2019,9(4) ], contour wave transformation (conjugate Transform, CT) [ Xiaoqi W, Jingbingg L, Rong T, et al. Contourlet-DCT Based on Multi-texture mapping [ J ]. Multimedia contacts and applications.2019,78 (7.), non-subsampled contour wave transformation [ Ayersha S, V.M. zero-watermark in Transform Doxon Quosys and 2019, U.S. watermark-D. watermark in parallel with D.E. watermark, U.S. watermark, J.S. watermark, D.E.S. watermark, J.S. watermark, 12. Transform, J.S. watermark, 12. watermark, D.S. watermark, J.S. watermark, D.S. watermark, J. watermark, modified optical Watermarking [ 20, U.E.S. watermark, D.S. watermark, D. watermark, E. watermark, D. watermark, No. 20, No. watermark, No. 20, No. watermark, No. 12, No. 20, No. 12, No. watermark, No. 12, No. watermark, No. 7, No. watermark, No. 7, No. 12, No. watermark, No. 12, No, the image feature can be directly constructed, or the principal component construction feature can be extracted by further performing operations such as Discrete Cosine Transform (DCT), Singular Value Decomposition (SVD), improved singular value decomposition and the like on the coefficient. Computing invariant moment features of images using polar harmonic transforms such as decimal-order polar harmonic transforms (DoPHTs) [ Xia Z, Wang X, Han B, et al, color image triple zero-water polar transforms using a decimal-order polar harmonic transform and a quaternary system [ J ] Signal Processing ], Quaternary Polar Harmonic Transforms (QPHTs) [ Zhoqiu X, Xingyuan W, Wenjie Z, et al, color image local harmonic transform and an acquisition polar transform [ J ] Signal Processing ], quaternary polar harmonic transforms [ Zhi Q, Xingyuan W, Wenjie Z, et al, color image linear harmonic transform and an acquisition polar transform [ J ] Signal Processing, 2019,157 Radial polar coefficients, Quaterial polar harmonics, Quaterial coefficients, Q-square polar transforms [ Q ] Q, E, Q, E, Q, E, Q, E, QPHFMs) [ Zhiqiu X, Xingyuan W, Xiaoxiao L, et al, effective copy protection for the same CT images based on quantitative polar harmonic feeder implementation [ J ]. Signal processing.2019,164 ], and the like. The method comprises the steps of extracting features In a space domain, constructing a stable relation between subblock means and integral means into Image features [ auspicious light bear, a strong space domain Robust Zero watermarking scheme [ J ] automation report 2018,44(01): 160-.

In the above zero watermark research, attention is mostly paid to improving the stability of image features, and when the number of images to be protected is large or the similarity between the images is high, a false alarm problem occurs, and an unprotected image is mistakenly judged as a protected image. The existing research mostly solves the false alarm problem of similar Images from the aspect of feature extraction, extracts features with robustness and higher uniqueness from the Images for generating Zero watermarks, and researches the similarity distinguishing problem of the Zero watermarks of a plurality of Medical Images in documents [ Wang W, Li Y, Liu S.A Polar Complex explicit Transform-Based Zero-watermark for Multiple Medical Images with High Discrimination [ J ]. Security and Communication networks.2021, 2021-13 ], constructs features according to the stability of the same order amplitude and the same repetition amplitude relation, and assigns a reference value for all amplitudes so as to reduce the similarity of the features of the similar Images. In the false alarm experiments of 10 cancer medical images, the average bit error rate and the minimum bit error rate were 0.3631 and 0.2629, respectively. Document [ Xiaoobing K, Fan Z, Yajun C, et al. combining polar harmonic transforms and 2D compound magnetic map for distinguishing and generating color Image zero-water marking algorithm [ J ]. Journal of Visual Communication and Image reproduction. 2020,70(C) ]. And simultaneously calculating three Polar Harmonic Transforms (PHTs) of the image, selecting a precise moment to enhance robustness, constructing a binary characteristic sequence according to the relation between the amplitudes of adjacent moments to generate a zero watermark, and performing a false alarm test on the babon image in the USC-SIPI data set and other similar babon images, wherein the error rate between the zero watermarks generated by the similar images is about 0.5. In the document [ Beiji Z, Jingyu D, Xiyao L, et al, discrete quantized zero-water marking scheme with similarity-based retrieval for digital rights Management of background Image [ J ]. Multimedia Tools and applications.2018,77(21) ], a circular region of a Fundus medical Image is divided into a plurality of sector-shaped sub-blocks, the difference between adjacent sub-blocks is calculated, and a four-valued feature vector is quantized as an Image feature, and a feature library is established, and in the copyright authentication process, a similarity search is performed in the feature library to distinguish a protected Image from a similar Image. The literature [ Liu X, Lou J, Wang Y, et al, dispersive and robust zero-watermarking scheme based on complete local binary pattern for authentication and copyrightness identification of the dimensional images [ C ]. In: Imaging information for Healthcare, Research, & application.2018 ] first constructs three CLBP value maps using Complete Local Binary Pattern (CLBP), divides them into a plurality of concentric circular ring regions, calculates the mean, variance, skewness, kurtosis of the regions, binarizes them as image features, and In a similar image false alarm test, the mean and minimum error rates are 0.4636 and 0.1111, respectively. In the document [ Liu X, Sun Y, Wang J, et al. A novel zero-water marking scheme with enhanced rendering and robust for volumetric information imaging [ J ]. Signal Processing: Image communication.2021,92(prepublish) ]. the Image is divided into a plurality of concentric circular areas, the mean, variance, skewness and kurtosis information residual features of each area are calculated, and the four features in the area are constructed into the features of one area after binarization, and in a similar Image false alarm test, the mean and minimum error rate values are about 0.4 and about 0.2 respectively. The documents [ Wang W, Li Y, Liu S.apolar compact explicit Transform [ J ]. secure and Communication networks.2021,2021(2):1-13 ] and documents [ Beiji Z, Jingyu D, Xiyao L, et al.Distinguishable Zero-watermark scheme with direct Image Management of rights Image [ J ]. Multimedia objects and applications.2018,77(21) ] solve the problem of Zero-warning, although the problem of Zero-warning, which can be generated for similar Images is a fixed watermark, and the problem of Zero-watermark extraction is still a fixed spatial discriminability, which is Based on a certain type of similarity, and the problem of Zero-watermark extraction is still indistinguishable, although it is a fixed watermark, Based on a certain level of similarity of spatial scalability.

After the image features are extracted, the image features and the watermark image are linked through methods such as logic operation, visual passwords, neural network training and the like to obtain a zero watermark, and the zero watermark is stored in an authentication center. Although watermark information is not embedded into a carrier, the generated zero watermark is used as an expression form for establishing a link between the carrier and a copyright identifier, if the generated zero watermark is low in security, under the condition that the zero watermark generation algorithm is disclosed, an attacker can analyze the corresponding relation between a protected electronic carrier and a copyright owner, and in special application environments such as military, medical treatment and the like, the corresponding relation is used as a piece of information and has a value which is difficult to estimate, so that the zero watermark is required to have certain security. Usually, both copyright watermark and Image features are binary matrices, so that logical operations (e.g. XOR) have the advantages of low complexity and reversibility, which are widely used in Zero-watermark algorithms, such as [ Shi S, Luo T, Huang J, et al. A Novel HDR Image Zero-watermark basic on Shift-initialization Shearlet Transform [ J ]. Securability and Communication networks.2021,2021(8655):1-12, Fan D, Li Y, Gao S, et al. A Novel watermark optimization algorithm on gain Transform and discrete cosine Transform [ J ]. Concurability and Computation: reaction and experiment, n/a (n/a): 5689.5-13, bear-light enhancement [ J ]. 160, null-watermark enhancement [ 18, 120J ],. 8, 18, 8, 18, 8-watermark, the method comprises the steps of (1) providing AROBustine Image Zero-water marking using conditional Neural Networks [ C ]. In: 20197 th International working on Biometrics and Forensics (IWBF).2019 ], and adding an encryption algorithm In the step mostly to improve the safety; the visual code is a secret shareable encryption scheme and is also used to construct zero watermarks such as the documents [ Beiji Z, Jingyu D, Xiyao L, et al, Distinguishable zero-watermark with spatial-based correlation Management of funus Image [ J ]. Multimedia Tools and applications.2018,77(21), Liu X, Lou J, Wang Y, et al, Distinguishing and robust zero-watermark based on complex local encoding for use In the authentication and correlation of the visual code [ C ] In: Imaging for the health, coding for the graphics, coding of the graphics, for the graphics, 365-; a method based on a Neural Network, such as documents [ Wang X, Ma D, Hu K, et al. mapping based basal Neural Network for Non-embedding and bland Image Watermarking [ J ]. Journal of Information Security and application.2021, 59 ], extracts features from an Image, then takes the features as input of a Residual Convolutional Neural Network (RCNN), takes a copyright watermark Image as output, trains a Network model, and stores trained model parameters as zero watermarks to an authentication center. In the three methods, an encryption algorithm needs to be additionally added in the logic operation, so that the algorithm efficiency can be reduced while the safety is improved; the visual password expands copyright watermark pixels and increases the data volume of the zero watermark; when the neural network method generates the zero watermark of each carrier image, network parameters need to be trained, so that the method has the limitation of the complexity of a parameter quantity network model.

Disclosure of Invention

Based on the defects of the prior art, the technical problem to be solved by the invention is to provide a low false alarm zero watermark algorithm based on regional exclusive OR, which can solve the false alarm problem of similar images on the premise of ensuring certain robustness and can be applied to zero watermark algorithms of various feature extraction methods.

The invention discloses a low false alarm zero watermark algorithm based on regional XOR, which comprises copyright watermark embedding and copyright authentication:

the zero watermark construction comprises the following steps:

1.1, extracting an image characteristic sequence, namely dividing the brightness component of a color image with the size of 512 multiplied by 3 into blocks without overlapping, making each sub-block with the size of 16 multiplied by 16 for increasing the characteristic stability, constructing the characteristic according to the size relation of the sub-block mean value and the brightness component overall mean value, and obtaining the space domain characteristic f with the size of 32 multiplied by 32^SPExtracting the frequency domain characteristic of the image, reducing the dimension of the color image characteristic, extracting the non-overlapping blocks of brightness components, wherein the size of each sub-block is 2 multiplied by 2, calculating the mean value of the sub-blocks to construct a matrix with the size of 256 multiplied by 256, extracting the frequency domain characteristic of the image by using discrete cosine transform, selecting a coefficient matrix with the size of 4 multiplied by 8 at the low frequency part, and converting the coefficient matrix into a 32-bit binary sequence f according to the positive and negative of the coefficient^FRAs a frequency domain feature;

1.2 redundant construction vehicle characteristics: feature f of size 32X 32^SPRedundant construction of features F of size 128X 128^SPAccording to a 32-bit signature sequence f^FRRedundant construction F^FR；

1.3 redundant construction of copyright watermark: dividing the copyright identification w of 32 multiplied by 32 into non-overlapping blocks, wherein the size of each sub-block is 32/p multiplied by 32/q, i belongs to {1,2, 3.. once, 128}, a random number matrix L with the size of (4 multiplied by p) x (4 multiplied by q) is generated, the numerical values are quantized into p multiplied by q, the number of each numerical value is the same, the numerical values are stored as a sub-block arrangement mode key K1, and the sub-blocks are arranged in a redundant mode;

1.4 regional XOR of feature indices: frequency domain feature F after redundant construction^FRDividing the redundant copyright mark W into two areas as an index matrix, and respectively connecting with a spatial domain characteristic F of a redundant structure^SPAnd a randomly generated binary matrix key K₂Performing XOR to obtain a zero watermark;

the copyright authentication comprises the following steps:

2.1 extracting space domain feature F 'with the size of 128 x 128 from the image needing copyright verification'^SPAnd frequency domain feature F'^FRSelecting pixel points in the zero watermark Z according to the frequency domain characteristics and respectively comparing the pixel points with the space domain characteristics F'^SPAnd a secret key K₂And performing exclusive or to obtain the redundant copyright watermark.

Optionally, the method further comprises watermark redundancy removal and ternary quantization;

and (3) removing redundancy of the watermark: the redundant copyright watermark does not overlap the sub-blocks, the sub-block size is 32/p multiplied by 32/q, the sub-blocks are marked as b'_m,nWherein m is in the range of {1,2, 3., 4 × p }, n is in the range of {1,2, 3., 4 × q }, and the key K is arranged according to the arrangement mode₁And (5) redundancy removal.

Further, when the quantization threshold is s₁＝1、s₂When the watermark is 15, the obtained redundant copyright watermark W' is subjected to redundancy removal ternary quantization to obtain a final ternary matrix W^final。

Therefore, on the basis of the existing feature extraction algorithm, the invention enlarges the difference between similar features by redundantly constructing the features and the watermarks, quantizing the three values and dividing the regional XOR of the feature indexes so as to solve the false alarm problem of the image zero watermark, thereby realizing the copyright protection of the digital information. The method comprises the steps of extracting frequency domain characteristics and space domain characteristics of an image by using two characteristic extraction methods, constructing the frequency domain characteristics into an index matrix, dividing a copyright watermark of a redundant structure into two regions according to the index matrix, performing exclusive OR with the space domain characteristics and the chaotic matrix of the redundant structure respectively, combining the regions to obtain a zero watermark, obtaining a copyright watermark image of the redundant structure by using the same characteristic extraction and region exclusive OR method in the copyright authentication process, performing redundancy removal operation, and quantizing the image into a three-value matrix to realize copyright authentication of distinguishable similar images.

The low false alarm zero watermark algorithm based on the regional exclusive OR has the following beneficial effects:

(1) and constructing a zero watermark with sensitive characteristics based on the existing characteristic extraction algorithm, and providing a new solution for the problem of image zero watermark false alarm.

(2) The method has the advantages that the characteristic of the carrier is fused in the process of establishing the relation between the characteristics and the copyright, the fusion of multiple characteristics is realized, and the false alarm rate of the image zero watermark is reduced.

(3) And redundantly constructing image characteristics and copyright watermarks, in the copyright authentication process, removing redundancy of the extracted copyright watermarks, quantizing the extracted copyright watermarks into a three-value matrix, and when the unprotected characteristic similar images are subjected to copyright authentication, the copyright information cannot be extracted.

(4) Based on two existing feature extraction algorithms, a low false alarm zero watermark algorithm is provided, a new solution idea is provided for the image zero watermark false alarm problem, and the used feature extraction algorithm can be replaced at will.

The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical means of the present invention more clearly understood, the present invention may be implemented in accordance with the content of the description, and in order to make the above and other objects, features, and advantages of the present invention more clearly understood, the following detailed description is given in conjunction with the preferred embodiments, together with the accompanying drawings.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly described below.

FIG. 1 is a numerical distribution histogram;

fig. 2 is a diagram of a zero watermark construction process;

fig. 3 is a diagram of a copyright authentication process.

Detailed Description

Other aspects, features and advantages of the present invention will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, which form a part of this specification, and which illustrate, by way of example, the principles of the invention. In the referenced drawings, the same or similar components in different drawings are denoted by the same reference numerals.

As shown in fig. 1 to fig. 3, in the zero watermark generating process, an image feature matrix and a copyright watermark are redundantly constructed, a 32-bit robust binary sequence is extracted from an image and constructed into an area index matrix, the redundant watermark image is divided into areas according to the index matrix, and the areas are respectively subjected to exclusive or with the redundant feature and the chaotic matrix, and the zero watermark is obtained after the areas are combined. In the copyright authentication process, a characteristic matrix and a 32-bit robust binary sequence are extracted from an image, a watermark image with a redundant structure is extracted by using the same method as the zero watermark generation process, and after the redundant operation is removed, the redundant image is quantized into a matrix with only three values as an extraction result. Experimental results show that the algorithm can solve the false alarm problem of similar images on the premise of ensuring certain robustness, and is applicable to zero-watermark algorithms of various feature extraction methods.

The technical scheme adopted by the invention is as follows:

1. analyzing false alarm problem cause of zero watermark

In general, in a zero watermark algorithm based on image feature extraction, robust features extracted from an image uniquely represent an image, so that a false alarm problem is mainly caused by insufficient feature uniqueness, and can be divided into two categories according to whether carrier images are similar or not: 1) false alarm problem between similar images: when the similarity of the two images is high, the features extracted from the two images have high similarity, so that the zero watermark generated by the two images has high similarity, and a false alarm problem is caused. 2) False alarm problem between dissimilar images: in the zero-watermark algorithm, a binary matrix is usually extracted from a grayscale image with 8 bit depths and a three-channel color image as a feature for constructing a zero watermark, and the size of the feature matrix is smaller than that of an original image, so that the diversity of image features is certainly much smaller than that of the original image, and as the number of images increases, the feature similarity degree between all images is greater than that of the original image, so that images with obvious visual effect differences but high feature similarities may occur, and a false alarm problem may also be caused.

2. Amplification of vector characteristic differences

In the zero watermark algorithm, when the similarity between image features is high, a false alarm problem is caused, and by expanding the difference between different image features, the false alarm problem can be effectively avoided, such as redundant structural features and watermarks, ternary quantization, feature index regional XOR and other operations, and the difference between similar features is expanded, so that the zero watermark false alarm problem is solved.

2.1 redundant construction vehicle features

Extracting m from color image₁×m₂Size binary feature matrix f^FEAWhen a false alarm problem occurs, it usually appears that the feature similarity between the two images is high. Feature matrix f for the presence of another image^SIMAnd f is and^FEAerror rate of B^fAnd the number of error bits E^fCalculated as formulas (1) and (2).

Will f is^FEAAnd f^SIMRedundant configuration m₁×m₂F of size XR^FEAAnd F^SIMError rate B^FAnd the number of error bits E^FCalculated as follows.

Due to F^FEAAnd F^SIMIs formed by f^FEAAnd f^SIMA redundant configuration is obtained, thus

F^FEA(i,j,1)＝F^FEA(i,j,2)＝...＝F^FEA(i,j,R)＝f^FEA(i,j) (5)

F^SIM(i,j,1)＝F^SIM(i,j,2)＝...＝F^SIM(i,j,R)＝f^SIM(i,j) (6)

Where i ∈ {1,2,3₁}、j∈{1,2,3,...,m₂And obtaining the error rate relationship and the error bit number relationship before and after the characteristic redundancy structure, as shown in formulas (7) and (8).

Although the characteristic redundancy structure can not influence the error rate among the characteristics, the error bit number among different characteristics can be enlarged to meet the requirement E^F＝R×E^fAnd R is redundancy.

2.2 redundant construction watermarking

Will have a size of m₃×m₄Binary watermark M^wRedundant configuration m₃×m₄Redundant binary watermark M of x R^wwAt the redundant watermark M^wwThe binary data of random position is modified to simulate the process of extracting copyright information from zero watermark generated by a certain image by using irrelevant images, and the modification quantity is beta multiplied by m₃×m₄X R', where β ∈ [0,1 ]]Representing redundant watermarks M^wwDegree of modification after copyright authentication, matrix M^wwThe probability of each bit being modified can be calculated by equation (9).

Wherein

Representing an element M in a matrix^ww(i, j, k) probability of being modified, i ∈ {1,2,3₁}、j∈{1,2,3,...,m₂Is a matrix M, k belongs to {1,2,3^wwAnd the index of the middle subscript and R' are the redundancy of the redundant construction watermark. M^wwThe middle numerical value is modified to obtain M'^wwCopyright authentication redundant watermark is de-redundant by the method of formula (10),obtaining a matrix M'^wWherein the numerical value ranges from {0,1,2, 3., R' }.

In matrix M'^wThe amount of change in the value of each element

Is expected to

Calculated from equation (11).

Represents matrix M'^wIn formula (11), i ∈ {1,2,3₁}、j∈{1,2,3,...,m₂Is matrix M'^wwAnd M'^wIndex of subscript of (a). Because β ∈ [0,1 ]]M 'with increasing redundancy R'^wExpected change of middle element

Is also increased, i.e. removes the redundant matrix M'^wAnd M^wThe value of the phase difference between xr' becomes large.

2.3 triple value quantization of redundancy-free watermarks

In order to enlarge the difference degree between the characteristics of different images, a three-value quantization processing matrix M 'is used in the zero-watermark copyright authentication process'^wAs shown in formula (12).

Redundancy in which R' is a redundancy-constructing watermarkRedundancy, i ∈ {1,2,3₃}、j∈{1,2,3,...,m₄}，α∈[0,1]For the three-valued quantization parameter, (α × R ') and (1- α) × R' are quantization thresholds. Matrix M'^wIn the method, the change amount of each numerical value is

That is, the compounds represented by the formulas (12) and (13) can be obtained

When the matrix is M'^wAmount of change of medium numerical value

When the quantization threshold is not less than three values (alpha x R'), the matrix

The value of the corresponding position in (A) is marked as 0.5 when

Less than the three-valued quantization threshold (α x R'),

the value of the corresponding position in the watermark is marked as the original binary watermark M^wThe numerical value of (c). Due to modification of the matrix M^wwOf random position, thus matrix M'^wAmount of change of each numerical value in

Testing when the parameters beta is 0.25, constructing the binary watermark image redundancy, modifying the random position value and removing redundancy, and calculating the variation of all elements in the matrix

The distribution of the values is shown in FIG. 1.

Amount of change of elements in matrix

Substantially satisfies the normal distribution, and therefore the amount of change of the element according to equation (11) is expected

It can be known that the parameter β indicating the degree of difference between the copyright watermarks before and after authentication has a certain relationship with the ternary quantization parameter α, and that, when α ═ β ═ 0.25, the ternary matrix is used

About half the number has a value of 0.5, and the three-value matrix can be influenced by adjusting alpha and beta

0.5 in number. Three-value matrix

The values 0 and 1 in (1) mainly come from the watermark image M^wThe value 0.5 is expressed as a gray point, and when a small number of values in the matrix is 0.5, the copyright information can be still recognized therefrom due to the visual characteristics of human eyes, and when a large number of values in the matrix is 0.5, the three-valued image

There are a large number of gray pixels in the image, and copyright information cannot be recognized.

Since the copyright authentication result is a three-valued matrix, it is difficult to express the difference from the original-copyright image using the error rate and the NC value. After the ternary matrix is displayed as an image, the image only has three colors of black, white and gray, and compared with the binary copyright image, gray pixel points are added, wherein the larger the number of the gray pixel points is, the larger the difference with the binary copyright image is, and the difference between the ternary copyright image and the binary copyright image can be represented by calculating the percentage gamma of the element with the numerical value of 0.5 in the ternary matrix, as shown in formula (15).

Wherein the content of the first and second substances,

calculating the number of satisfying conditions for a ternary matrix obtained after copyright authentication by using a count (·) function, wherein i belongs to {1,2,. cndot., 32}, j belongs to {1,2,. cndot., 32} is an element index in the matrix, the value of an obtained percentage parameter gamma is between 0 and 1, and when gray pixel points in the ternary image are 12.5% and 25%, the information of the copyright image can be seen; when the gray pixel points in the ternary image are 75%, no effective information can be seen in the image; when the gray pixel points in the three-value image are 50%, copyright information can be seen in a hidden way, but the damage is serious.

With the parameter β determined, the adjustable parameter α will remove the redundancy matrix M'^wThe numerical value of different quantities in the process is quantized to 0.5; under the condition of determining the parameter alpha, different ternary images can be obtained according to the difference degree beta of the characteristics of the two images so as to distinguish the copyright authentication results of the attacked image and the false alarm image and avoid the problem of zero watermark false alarm caused by similarity of the image characteristics.

2.4 regional XOR of feature indices

In order to improve the safety of the zero-watermark algorithm, reduce the false alarm rate of the zero-watermark algorithm and visually represent copyright authentication information, a regional exclusive OR algorithm of feature indexes is provided, binary features are extracted from an image and are reconstructed into a matrix F 'with the size of M multiplied by N'^FEAWhile simultaneously watermarking the binary watermark M^wRedundant construction of a watermark M of size MxN^wwFrom matrix F'^FEAMedian selection watermark M^wwAnd performing exclusive or on the intermediate pixel and the two binary matrixes to obtain a zero watermark Z as shown in the formula.

In the copyright authentication process, a matrix F with the size of M multiplied by N is extracted from an image "^FEAAccording to the matrix F "^FEAAnd carrying out XOR on pixels in the median value selection zero watermark Z and two binary matrixes to obtain a copyright authentication image M'^wwAs shown in formula (17).

In equation (17), i ∈ {1,2, 3.. eta., M }, j ∈ {1,2, 3.. eta., N } is an index of the element position in the matrix, and M ∈ {1,2, 3.. eta., N }, which is an index of the element position in the matrix₁And M₂Are randomly generated binary matrixes of size M multiplied by N, and when copyright authentication is performed by using images similar to the characteristics causing the false alarm, the matrix F 'is extracted from the false alarm image'^FEAFeature matrix F with higher similarity "^FEAThe matrix F "^FEACan be regarded as being made of matrix F'^FEAModifying partial elements to obtain a difference matrix D of feature matrices^FAnd rate of difference D^FAnd a difference matrix D of the random matrix^MAs shown in the following formula.

D^F＝xor(F'^FEA,F”^FEA) (18)

D^M＝xor(M₁,M₂) (20)

D^FAnd D^MIs a binary matrix, respectively representing the difference of features and the difference of a random matrix when D^FAnd D^MIs 0, indicating that there is no difference at that point, when D is^FAnd D^MIs 1, indicating that there is a difference at that point, in the copyright authentication process of equation (16), when M is equal to₁(i,j)＝M₂(ii) at the time of (i, j),

from the formula (16) can be derivedWhen there is no difference in the characteristics at a certain point, i.e. D^FWhen (i, j) ═ 0, M 'obtained by authentication of formula (16) was used'^ww(i,j)＝M^ww(i, j); when there is a difference in the characteristics at a certain point, i.e. D^FWhen (i, j) ═ 1, M 'obtained by calculation from formula (16)'^ww(i, j) errors may occur, for random matrix differences D^MTo derive the correct M based on equation (16)^ww(i, j) calculation method.

From the formula (21), in feature F "^FEA(i, j) M 'obtained by copyright authentication in the case of error'^ww(i, j) not all errors, when D^M(i, j) ' 0, i.e. M ' when there is no difference at a certain point in the two random matrices '^ww(i,j)＝M^ww(i, j), namely M'^ww(i, j) error free; when D is present^MWhen (i, j) is 1, namely a difference exists between certain points in the two random matrixes,

namely M'^ww(i, j) error. It can be concluded that when D^F(i, j) and D^M(i, j) are simultaneously 1, and the copyright authentication result M'^ww(i, j) error. Due to M₁And M₂For the purpose of a randomly generated binary matrix,

representing the occurrence of an event D^M(i, j) is 1, so M₁And M₂The probability that the numerical values of certain elements are different is

Let feature matrix F'^FEAAnd F "^FEAThe error rate of (a) is mu,

representing the occurrence of an event D^F(i, j) ═ 1, feature index matrix F'^FEAAnd F "^FEAThe probability of a different value of an element is

It can be concluded that when the watermark image is subjected to the regional XOR of the two regions, the probability of numerical errors in the copyright authentication watermark is

I.e. the error rate of the copyright authentication watermark is about 0.5 x mu. In the application of the zero-watermark algorithm, the watermark can be divided into a plurality of areas corresponding to a certain number of random matrixes, and the generation and authentication processes of the zero-watermark are shown as formulas (22) and (23).

Wherein h is the number of divided regions of the watermark, and the error rate of the watermark obtained by copyright authentication is about

As h increases, the error rate of the copyright watermark approaches the error rate μ between features. Can be expanded into a plurality of random matrixes M_hTherefore, the chaotic binary matrix generated by a plurality of keys can be taken as M_hThe secret key is distributed to all people, so that multi-key sharing is realized; the redundantly constructed carrier characteristics can also be used as a matrix M_hTo expand the differences between similar features.

Extracting features from the image by two feature extraction algorithms, and respectively using the extracted two carrier features as index matrixes F'^FEASum matrix M₁Chaotic binary matrix generated by secret key as M₂. In the process of copyright authentication, the extracted matrix F "^FEAAnd M'₁Possibly with matrix F'^FEASum matrix M₁Exist at a certain levelWhen feature matrix F ', as previously demonstrated'^FEAAnd F "^FEAHas a bit error rate of mu and a matrix M₁＝M'₁Then, the error rate of the copyright authentication watermark is about 0.5 multiplied by mu; when the feature matrix M₁And M'₁Bit error rate of mu 'and matrix F'^FEA＝F”^FEAIndex matrix F'^FEAWhen the medium and two values are balanced, the error rate of the copyright authentication watermark is about 0.5 multiplied by mu'.

The false alarm problem of the zero watermark is mainly shown as high similarity of image features, the feature index region XOR algorithm provided by the invention can construct two features of an image into the zero watermark, when the uniqueness of one feature is insufficient to cause the false alarm problem, the code error rate value of the copyright authentication watermark can be improved by the uniqueness of the other feature, and the false alarm problem can be avoided by combining the three-value quantization algorithm in section 2.3. The false alarm problem occurs only when the similarity of all features of the two images is high, and can be avoided by increasing the kinds of features.

The invention relates to a low false alarm zero watermark algorithm based on regional XOR, which comprises a zero watermark construction process and a copyright authentication step:

the zero watermark construction includes the following:

in the zero watermark construction, firstly, frequency domain characteristics and space domain characteristics are extracted from an image, then the space domain characteristics, the frequency domain characteristics and copyright watermark redundancy are constructed into a matrix with the size of 128 multiplied by 128, the frequency domain characteristics are used as an index matrix for selecting pixel points of copyright watermarks, and the pixel points are subjected to exclusive OR respectively in the space domain characteristics and a key matrix to obtain the zero watermark.

1.1 extracting image characteristic sequence.

Dividing the brightness components of the color image with the size of 512 multiplied by 3 into blocks without overlapping, making each sub-block with the size of 16 multiplied by 16 for increasing the stability of the characteristics, constructing the characteristics according to the size relation of the average value of the sub-blocks and the integral average value of the brightness components, and obtaining the space domain characteristics f with the size of 32 multiplied by 32^SP. Extracting the frequency domain feature of the image, reducing the dimension of the color image feature, extracting the non-overlapping blocks of the brightness component, wherein the size of each sub-block is 2 multiplied by 2, calculating the average value of the sub-blocks to construct a size of 256 in256 matrix, extracting the frequency domain characteristic of the image by using Discrete Cosine Transform (DCT), selecting a coefficient matrix with the size of 4 multiplied by 8 in the low frequency part, and converting the coefficient matrix into a 32-bit binary sequence f according to the positive and negative of the coefficient^FRAs a frequency domain feature.

1.2 redundant construction vehicle features.

Feature f of size 32X 32^SPRedundant construction of features F of size 128X 128^SPAccording to a 32-bit signature sequence f^FRRedundant construction F^FRThe construction method is shown as the formula (24-27).

c_mn＝f^SP (25)

F^FR＝(c₁…c_k) (26)

c_k(i,j)＝f^FR(1,k) (27)

Wherein c is_kTo construct F^FRK ∈ {1,2, 3.., 32} each subblock size is 128 × 4, i ∈ {1,2, 3., 128}, j ∈ {1,2,3,4} is c_kIndex by subscript of internal element, 32 c_kTogether forming a feature matrix F of size 128 x 128^FR；c_mnTo construct F^SPM is equal to {1,2,3,4} and n is equal to {1,2,3,4}, and 16 c are contained in total_mnTogether forming a feature matrix F of size 128 x 128^SP。

1.3 redundantly constructing copyright watermark.

Dividing the copyright identification w of 32 multiplied by 32 into non-overlapping blocks, wherein the size of each sub-block is 32/p multiplied by 32/q, i belongs to {1,2, 3.. multidot.128 }, generating a random number matrix L with the size of (4 multiplied by p) × (4 multiplied by q), quantizing the numerical values into p multiplied by q, keeping the numerical values as a sub-block arrangement mode key K1, and arranging the sub-blocks in a redundant mode according to the method of the formula (29).

Wherein p and q are subblock arrangement position indexes, i belongs to {1,2, 3., 4 × p }, j belongs to {1,2, 3., 4 × q }, and a W matrix of a redundancy structure is formed by b_pqConstitute redundant copyright notice W of total size 128 × 128.

1.4 regional XOR of feature indices

Frequency domain feature F after redundant construction^FRDividing the redundant copyright mark W into two areas as an index matrix, and respectively connecting with a spatial domain characteristic F of a redundant structure^SPAnd a randomly generated binary matrix key K₂And XOR is carried out to obtain the zero watermark Z as shown in the formula (30).

Where i ∈ {1,2, 3., 128}, j ∈ {1,2, 3., 128} is the pixel position index.

The copyright authentication step includes the following contents:

spatial domain feature F 'of size 128 x 128 is extracted from the image needing copyright verification by using the feature extraction method in 1.1 and the redundant construction method in 1.2 and 1.3'^SPAnd frequency domain feature F'^FRSelecting pixel points in the zero watermark Z according to the frequency domain characteristics and respectively comparing the pixel points with the space domain characteristics F'^SPAnd a secret key K₂And XOR is carried out to obtain the redundant copyright watermark W', as shown in formula (31).

Where i ∈ {1,2, 3., 128}, j ∈ {1,2, 3., 128} is the pixel position index. The image is attacked by Gaussian noise (attack intensity is 0.2), and redundant frequency domain feature F 'is extracted'^FRAnd F 'after airspace feature'^SPAnd extracting the redundant copyright watermark W'.

Watermark redundancy removal and ternary quantization:

3.1 watermark redundancy removal

The redundant copyright watermark W 'does not overlap the sub-blocks, the sub-block size is consistent with that in the upper section and is 32/p multiplied by 32/q, and the sub-blocks are marked as b'_m,nWherein m is in the range of {1,2, 3., 4 × p }, n is in the range of {1,2, 3., 4 × q }, and the key K is arranged according to the arrangement mode₁Redundancy is removed as shown in equation (32).

Where m is equal to {1,2, 3.,. 4 × p }, n is equal to {1,2, 3.,. 4 × q } and is a subblock position index₁₁,a₁₂,...,a_pqIs a matrix with an initial value of 0 and a size of 32/p × 32/q, and all b'_m,nAfter the operation of the formula (33), the watermark w' with the size of 32 × 32 is obtained.

3.2 ternary quantization

Since the 128 × 128 size W 'is de-redundant to 32 × 32 size W', each pixel in W 'has a value ranging from 0 to 16, the watermark W' is quantized according to equation (34) to a matrix W of only three values^finalWhere the values are {0,0.5,1}, and w is^finalAnd as a final extraction result, completing copyright authentication.

Wherein i belongs to {1,2, 3.,. 32}, and j belongs to {1,2, 3.,. 32} is w^finalInner pixel index, s₁And s₂For two quantization thresholds s₁＝α×R'、s₂(1- α) × R' satisfying 0 ≦ s₁≤s₂And less than or equal to 16, alpha is a three-value quantization parameter, and R' is copyright watermark redundancy. At quantization threshold of s₁＝1、s₂When the watermark is 15, the obtained redundant copyright watermark W' is subjected to redundancy removal ternary quantization to obtain a final ternary matrix W^final。

While the foregoing is directed to the preferred embodiment of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.

Claims

1. The low false alarm zero watermark algorithm based on the regional XOR is characterized by comprising a zero watermark construction and copyright authentication:

the zero watermark construction comprises the following steps:

1.4 regional XOR of feature indices: frequency domain feature F after redundant construction^FRDividing the redundant copyright identification W into two areas as an index matrixSpatial domain feature F of discriminating redundant structure^SPAnd a randomly generated binary matrix key K₂Performing XOR to obtain a zero watermark;

the copyright authentication comprises the following steps:

2. The split-region exclusive or-based low false alarm zero watermark algorithm of claim 1, further comprising watermark de-redundancy and ternary quantization;

3. The split-region XOR-based low false alarm zero watermark algorithm of claim 2, wherein at quantization threshold s₁＝1、s₂When the watermark is 15, the obtained redundant copyright watermark W' is subjected to redundancy removal ternary quantization to obtain a final ternary matrix W^final。