CN110009547B - JPEG asymmetric digital image steganography method - Google Patents

Abstract

The invention relates to a JPEG asymmetric digital image steganography method, which obtains a carrier imageDividing the carrier image into two interleaved sub-images X ₁ X ₂ (ii) a Calculating an initialization distortion cost value and an initialization embedding modification map R of a carrier image ₁ 、R ₂ (ii) a Optimization of R ₁ Updating the sub-picture X in the case of intra-block embedded modification of each DCT block ₁ Obtaining the cost value of each element to obtain a new secret sub-graph Y' ₁ And a new embedded modified map R' ₁ (ii) a Constructing a four-neighbor domain model incorporating an embedding modified map R' ₁ And BBC policy update X ₂ Cost value of each element in the graph, and new secret sub graph Y 'is obtained' ₂ (ii) a Secret carrier sheet Y' ₁ And secret sub-figure Y' ₂ And merging to obtain a complete secret-carrying image Y ', and sending the complete secret-carrying image Y' into a steganography analyzer to detect the safety performance of the JPEG asymmetric steganography algorithm. According to the JPEG asymmetric digital image steganography method provided by the invention, a new cost value updating method is constructed, the safety performance of the algorithm is effectively improved, and the simultaneous updating and convergence speed block of +1/-1 distortion cost values in the horizontal direction and the vertical direction are realized through the establishment of a four-neighbor domain model.

Description

JPEG asymmetric digital image steganography method

Technical Field

The invention relates to the field of multimedia information security, in particular to a JPEG asymmetric digital image steganography method.

Background

In recent years, with the development of global networking and informatization, research on information security is driven to a hot trend, and multimedia information security is rapidly developed as an important branch of the information security field. Therefore, the digital image steganography technology is also concerned by the majority of research inventors as the important research content in the multimedia information security field, has important significance for the information communication security based on the network in the information era, and has great application in national defense.

The digital image steganography is that a digital image is used as a carrier, and secret information is embedded into an open carrier image, so that the effect of secret information transmission is achieved. Currently, there are three main indicators for measuring the performance of the digital image steganography method:

first, security, i.e., the detection of digital image steganalysis methods is to be maximally resisted. The digital image steganalysis is a technology for judging whether an input image contains steganographic information or not through characteristic analysis;

secondly, imperceptibility, namely, the carrier image and the secret image are similar to each other in statistical property as much as possible, so that the detectability is improved;

third, the embedded payload means the length of the embedded secret information. For the same carrier image, under the condition of ensuring the same security, if the more information is embedded, the more the steganography method has application value. In other words, the same secret information is embedded in the same carrier image, and the steganography method has higher application value if the security of the image is higher (the capability of resisting digital image steganography analysis is stronger).

In addition, according to different fields of secret information embedding, the digital image steganography technology which is currently mainstream can be mainly divided into two types, namely airspace digital image steganography and JPEG (Joint Photographic Experts Group) domain digital image steganography. The spatial domain digital image steganography is to embed secret information by modifying spatial domain pixel values, and the JPEG domain digital image steganography is to embed secret information by modifying DCT (Discrete Cosine Transform) coefficients. In recent years, since JPEG digital images have been widely used due to their high compression rate, research on steganography techniques using JPEG digital images as carriers has received attention from more research inventors in recent years.

At present, the mainstream JPEG image steganography method is basically realized based on a minimum embedded distortion framework, namely, the modification cost value of each DCT coefficient is calculated firstly, and then the secret information is embedded into a carrier and accompanied with minimum carrier total distortion by combining an encoding technology according to the size of the modification cost value. To our knowledge, J-UNIWARD ^[1] 、UERD ^[2] 、GUED ^[3] 、BET ^[4] Etc. are implemented based on the framework. With the proposition of breakthrough STCs (Syndrome Trellis Codes), the current mainstream steganographic algorithms focus on designing better distortion cost functions. The early distortion cost function assigns the same value to each element +1/-1 cost value, which we callA symmetric steganographic method. Recent studies have shown that more secure steganographic performance can be achieved when different values are assigned to the cost value of each element +1/-1, we call this type of method asymmetric steganographic methods.

Aiming at the asymmetric steganography technology, the current mainstream airspace steganography method comprises CMD ^[5] And Synch ^[6] The main ideas of the two methods are aggregation modification directions, but the adopted cost value strategies are different, the former method only optimizes the cost value of the coefficient close to the expected modification direction, and the latter method correspondingly adjusts the values of the same and opposite coefficients to the expected modification direction, so that the former method is superior to the latter method in terms of algorithm safety. For JPEG asymmetric digital image steganography, the method known in the prior art is disclosed as Dejoin _ J ^[7] The core of the method is inter-Block edge Continuity criterion (BBC criterion for short). The criterion is mainly based on the correlation between DCT blocks, giving the desired modification selection of 64 DCT modes within an 8 x 8DCT block-synchronous or asynchronous, i.e. either the desired co-orientation to the neighborhood modification direction (synchronous modification) or the desired inverse orientation to the neighborhood modification direction (asynchronous modification), for the horizontal and vertical directions, respectively, in combination with spatial Dejoin ^[8] And (4) strategy, thus realizing the asymmetrical image steganography of the JPEG domain.

However, the method Dejoin _ J only aims at maintaining the continuity of the edges (spatial domain) between DCT blocks, but does not discuss the influence of the embedded modification in the DCT blocks, and the convergence speed is slow. According to Dejoin _ J, the +1/-1 distortion cost value of the DCT coefficient is updated and adjusted by fixing the embedded change of a part of the DCT coefficients in the image, and then according to the fixed embedded changes and the proposed BBC criterion, the +1/-1 cost value of the DCT coefficient of the same mode between the adjacent blocks is updated according to the correlation between the blocks. However, the embedded change of these fixed DCT coefficients is not optimal and may affect the security performance of the algorithm to some extent.

Disclosure of Invention

The invention provides a JPEG asymmetric digital image steganography method, aiming at overcoming the technical defects that the prior art JPEG digital image steganography method is low in convergence speed, the embedded change of a fixed DCT coefficient is not optimal, and the safety performance of an algorithm is influenced.

In order to solve the technical problems, the technical scheme of the invention is as follows:

a JPEG asymmetrical digital image steganography method comprises the following steps:

s1: obtaining a carrier image X in JPEG format, dividing the carrier image into two interweaved subgraphs X ₁ 、X ₂ ；

S2: method for obtaining initialization distortion cost value C of carrier image based on JPEG image symmetric steganography algorithm _ori And initializing an embedded modification graph R ₁ 、R ₂ ；

S3: optimization of R ₁ Updating the sub-picture X in the case of intra-block embedded modification of each DCT block ₁ Obtaining a new secret sub-graph Y 'by embedding analog simulation of the cost value of each element' ₁ And a new embedded modified map R' ₁ ；

S4: constructing a four-neighbor domain model incorporating an embedding modification map R' ₁ And BBC policy update X ₂ The cost value of each element in the graph is obtained, and a new secret sub graph Y 'is obtained through analog simulation embedding' ₂ ；

S5: secret carrier sheet Y' ₁ And secret sub-figure Y' ₂ And merging to obtain a complete secret-carrying image Y ', and sending the complete secret-carrying image Y' into a steganography analyzer to detect the safety performance of the proposed JPEG asymmetric steganography algorithm.

Wherein, the step S1 comprises the following specific steps:

s11: JPEG image compression is carried out on the original airspace image to obtain a JPEG image which is compressed and coded under the condition of different quality factors and is used as a carrier image X;

s12: dividing a carrier image X into two interleaved subgraphs X ₁ 、X ₂ The division unit is 8 × 8DCT blocks.

Wherein the step S2 specifically comprises the following steps:

s21: calculating initialization distortion cost value C of carrier image X based on JPEG image symmetric steganography algorithm _ori ；

S22: will initialize a distortion cost value C _ori And carrier image X feedingAn analog simulation embedder is used for obtaining an initialized secret-carrying image Y;

s23: carrying out difference on the secret-carrying image Y and the carrier image X to obtain a residual image D, wherein the residual image D contains embedded and modified amplitude and direction information;

s24: ignoring the amplitude information of the embedding modification of the residual image D, and keeping the direction information of the embedding modification to obtain a carrier image X embedding modification image R, wherein R = sgn (D); obtaining two subgraphs simultaneously as R ₁ 、R ₂ Corresponding to an initialization distortion cost value of C _{ori_1} 、C _{ori_2} 。

Wherein the step S3 specifically comprises the following steps:

s31: statistical embedding modification graph R ₁ The modified number in the current DCT block is denoted by symbol N, and a constraint condition whether the DCT block is optimized is set, specifically:

s32: listing all possible embedding modification combinations, and randomly selecting half of the embedding modification combinations in order to reduce the computational complexity, namely, the total number of the embedding modification combinations is K =0.5 × 2 ^N ；

S33: respectively calculating spatial domain embedding distortion cost C corresponding to K modification combinations of the current DCT block _DCT ；

S34: selecting minimum distortion cost C from K combined spatial domain embedding distortion costs _{DCT_min} The corresponding embedding modification combination is used as the optimal embedding modification combination of the DCT, namely the expected modification direction of each DCT coefficient in the current DCT block is obtained;

s35: updating the +1/-1 cost value of each coefficient in the current DCT block according to the optimal embedded modification combination, which specifically comprises the following steps: setting an intra-block penalty factor w _inner (r _o ,r _a ) And updating a formula of the obtained cost value:

wherein r is _o Indicates a desired modification direction, r _a Modifying the direction for the reality; (u, v) represents the coordinates of the DCT block, which take on an integer ranging from 0. Ltoreq. U.ltoreq.63, 0. Ltoreq. V.ltoreq.63 for a 512 x 512 image; (m, n) represents the coordinates of the coefficients in the DCT block, and the values of m are integers, wherein m is more than or equal to 0 and less than or equal to 7, and n is more than or equal to 0 and less than or equal to 7;

represents an initial cost value of the (m, n) coefficient within the (u, v) th DCT block;

representing the new cost value of the updated (m, n) coefficient in the (u, v) th DCT block; from C' ₁ Representing a set of these new cost values;

s36: will be of cost value C' ₁ And subfigure X ₁ Sending the information into an analog simulation embedder to obtain a new secret sub-graph Y' ₁ Obtaining sub-graph X by subtracting the sub-graph X from the carrier image X and then taking symbol operation ₁ Novel Embedded modified map R' ₁ 。

Wherein the step S4 specifically includes:

s41: aiming at any DCT coefficient, constructing a four-adjacent domain model by using elements with the same mode in adjacent DCT blocks, wherein the adjacent DCT blocks are four, namely an upper DCT block, a lower DCT block, a left DCT block and a right DCT block which are directly connected with the current DCT block;

s42: embedding modified graph R' ₁ And sub-diagram X ₂ Is embedded in the modification map R ₂ Merging and traversing the merged embedded modified graph R ₂ Is traversed starting from the (0, 0) element, according to the embedded modification map R' ₁ Acquiring embedded modification combinations of the same pattern elements in 4 adjacent DCT blocks;

s43: combining with BBC criterion, updating the mode according to the expected horizontal and vertical updating modes of the mode ₂ Updating the cost value of the elements in the subgraph, wherein the specific calculation formula is as follows:

(p,q)∈{(u-1,v),(u+1,v),(u,v-1),(u,v+1)}；

wherein, the first and the second end of the pipe are connected with each other,

representing the updated cost value of the (m, n) th DCT coefficient in the (u, v) th DCT block; w is a _inter Representing an inter-block penalty coefficient;

representing an initial cost value of the (m, n) th DCT coefficient in the (u, v) th DCT block;

the cost value of the same-mode coefficient in four neighborhood DCT blocks of the current DCT block is represented as a subgraph X ₁ Cost value C 'after intra-block optimization updating' ₁ Element (p, q) represents the coordinates of the four neighborhood DCT blocks;

s44: go back to step S42 to update the next element until sub-graph X ₂ The cost values of all DCT coefficients are updated and are C' ₂ Representing a set of these new cost values;

s45: c 'new cost value' ₂ And subfigure X ₂ Sending the mixture into an analog simulation embedder again to obtain a secret carrier picture Y' ₂ 。

Wherein the step S33 specifically includes:

s331: construction of a wavelet direction filter bank K = { K ] using DB-8 decomposition filters, i.e., one-dimensional low-pass filter l and high-pass filter h ₁ ,K ₂ ,K ₃ In which K is ₁ ＝l·h ^T ，K ₂ ＝h·l ^T ，K ₃ ＝h·h ^T ；

S332: for the current embedded modification combination, sequentially acquiring the spatial variation of each modified element (i, j) through two-dimensional inverse DCT, namely:

wherein:

wherein, the first and the second end of the pipe are connected with each other,

representing the spatial variation of the pixel of the spatial (s, t) position corresponding to the element of the modified current DCT block (i, j);

s333: and accumulating the spatial domain changes corresponding to all element modifications in the current embedded modification combination, namely:

s334: calculating the relative change of the wavelet domain generated by the current embedding modification combination in three directions, specifically:

B ^(k) ＝W*K _k ,k＝{1,2,3}；

wherein: w denotes the spatial impact of the current embedding modification combination, K _k Represents the k-th directional filter;

s335: the wavelet domain influence in three directions is integrated to obtain the distortion cost value C of the wavelet domain of the current embedded combination _DCT The method specifically comprises the following steps:

wherein: (u, v) is BETA ^(k) And (u, v) coordinates of 0. Ltoreq. U and v. Ltoreq.22.

In step S11, the process of JPEG image compression specifically includes:

s111: 8 multiplied by 8 blocking the original spatial domain image;

s112: performing DCT (discrete cosine transformation) on the partitioned image and quantizing the partitioned image;

s113: and performing zig-zag scanning entropy coding on the quantized image to obtain a compressed JPEG image.

Wherein, the JPEG image symmetric steganography algorithm in the step S21 comprises one or more of J-UNIWARD, UERD and GUED.

In the scheme, the method optimizes the DCT block embedded modification diagram based on the minimum spatial domain embedding distortion principle, considers the spatial domain embedding influence in the block, finds a modification scheme which enables the carrier image block to have minimum embedded distortion by adjusting the preset embedding direction +1/-1, and then realizes the modification scheme by properly adjusting the +1/-1 distortion cost value and combining the existing coding technology. The optimized embedding change is used for updating reference of cost value between subsequent DCT blocks, so that the algorithm approaches to the optimal safety performance more quickly.

In the scheme, the concept of spatial domain asymmetric digital image steganography and BBC (base band coding) criteria are combined, the cost value iterative updating based on the correlation between DCT (discrete cosine transformation) blocks is realized by dividing two interlaced subgraphs, and the calculation complexity of the algorithm is reduced under the same safety performance; in addition, by combining with the DCT block internal embedding optimization, the safety performance of the algorithm can be further improved.

Compared with the prior art, the technical scheme of the invention has the beneficial effects that:

on the basis of the current mainstream JPEG steganography algorithm, a new cost value updating method is formed through DCT block embedding optimization based on space domain minimum embedding distortion and cost value updating based on DCT block correlation, so that the safety performance of the JPEG asymmetric digital image steganography method is higher than that of the current mainstream JPEG steganography algorithm; the method is based on the operation of the carrier image, does not limit the initialization algorithm, has low complexity of algorithm design and strong generalization, and can be applied to any initialization algorithm based on a minimum embedded distortion framework; through the establishment of the four-adjacent domain model, the simultaneous update of the +1/-1 distortion cost values in the horizontal direction and the vertical direction is realized, and the speed block is converged.

Drawings

FIG. 1 is a schematic flow chart of the steps of the method of the present invention;

fig. 2 is a schematic diagram of a cost value updating process based on IEO;

FIG. 3 is a diagram of a four-neighborhood model;

figure 4 is a schematic diagram of a cost value updating process based on SBC;

FIG. 5 is a BBC strategy illustration;

FIG. 6 is a schematic diagram of a JPEG image compression process.

Detailed Description

The drawings are for illustrative purposes only and are not to be construed as limiting the patent;

for the purpose of better illustrating the present embodiments, certain elements of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;

it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.

Example 1

As shown in fig. 1, a JPEG asymmetric digital image steganography method includes the following steps:

s1: taking a representative database BOSSbase ver1.01 of digital image steganography as a data source, respectively carrying out JPEG compression on the data source through quality factors 75 and 95 to obtain a carrier image X, and dividing the carrier image into two interweaved subgraphs X ₁ 、X ₂ ；

S2: method for obtaining initialization distortion cost value C of carrier image based on JPEG image symmetric steganography algorithm _ori And initializing an embedding modification map R ₁ 、R ₂ ；

S3: optimization of R ₁ Updating the sub-picture X in the case of intra-block embedded modification of each DCT block ₁ Obtaining a new secret sub-image Y 'by embedding analog simulation of the cost value of each element' ₁ And a new embedded modified map R' ₁ ；

S4: constructing a four-neighbor domain model incorporating an embedding modified map R' ₁ And BBC policy update X ₂ The cost value of each element in the tree is obtained by embedding the cost value in analog simulationTaking a New secret sub-graph Y' ₂ ；

S5: secret carrier sheet Y' ₁ And secret sub-figure Y' ₂ And merging to obtain a complete secret-carrying image Y ', sending the complete secret-carrying image Y' into a steganography analyzer, and detecting the safety performance of the proposed JPEG asymmetric steganography algorithm.

More specifically, the specific steps of step S1 are:

s11: JPEG image compression is carried out on the original airspace image to obtain a JPEG image which is compressed and coded under the condition of different quality factors and is used as a carrier image X;

s12: dividing the carrier image X into two interleaved sub-images X ₁ 、X ₂ The division unit is 8 × 8DCT blocks.

More specifically, the step S2 specifically includes:

s21: calculating initialization distortion cost value C of carrier image X based on JPEG image symmetric steganography algorithm _ori ；

S22: will initialize a distortion cost value C _ori Sending the carrier image X into an analog simulation embedder to obtain an initialized carrier secret image Y;

s23: carrying out difference on the secret-carrying image Y and the carrier image X to obtain a residual image D, wherein the residual image D contains embedded and modified amplitude and direction information;

s24: ignoring the magnitude information of the embedding modification of the residual image D, and keeping the direction information of the embedding modification to obtain a carrier image X embedding modification image R, wherein R = sgn (D); obtaining two subgraphs simultaneously as R ₁ 、R ₂ The corresponding initialization distortion cost value is C _{ori_1} 、C _{ori_2} 。

More specifically, as shown in fig. 2, the step S3 specifically includes:

s31: statistical embedding modification graph R ₁ The modified number in the current DCT block is denoted by symbol N, and a constraint condition whether the DCT block is optimized is set, specifically:

s32: listing all possible embedding modification combinations, and randomly selecting half of the embedding modification combinations in order to reduce the computational complexity, namely, the total number of the embedding modification combinations is K =0.5 × 2 ^N ；

S33: respectively calculating spatial domain embedding distortion cost C corresponding to K modification combinations of the current DCT block _DCT ；

S34: selecting minimum distortion cost C from K combined spatial domain embedding distortion costs _{DCT_min} The corresponding embedding modification combination is used as the optimal embedding modification combination of the DCT, namely the expected modification direction of each DCT coefficient in the current DCT block is obtained;

s35: updating the +1/-1 cost value of each coefficient in the current DCT block according to the optimal embedded modification combination, specifically: setting an intra-block penalty factor w _inner (r _o ,r _a ) And updating a formula of the obtained cost value:

wherein r is _o Indicates a desired modification direction, r _a Modifying the direction for the reality; (u, v) represents the coordinates of the DCT block, which take on an integer ranging from 0. Ltoreq. U.ltoreq.63, 0. Ltoreq. V.ltoreq.63 for a 512 x 512 image; (m, n) represents the coordinates of the coefficients in the DCT block, and the values are integers of 0-m 7, 0-n 7;

represents an initial cost value of the (m, n) coefficient within the (u, v) th DCT block;

representing the new cost value of the updated (m, n) coefficient in the (u, v) th DCT block; from C' ₁ Representing a set of these new cost values;

s36: will be of cost value C' ₁ And subfigure X ₁ Sending the secret information into an analog simulation embedder to obtain a new secret carrier graph Y' ₁ And obtaining a sub-image by performing a difference operation with the carrier image X and then performing a sign operationDrawing X ₁ Novel Embedded modified map R' ₁ 。

In the specific implementation process, the intra-block penalty coefficient w _inner (r _o ,r _a ) Is taken to be the direction r of desired modification within the DCT block _o And an initial modification direction r _a By definition, the value table is shown in table 1:

TABLE 1 penalty factor w in DCT blocks _inner (r _o ,r _a ) Watch (A)

Wherein the parameter alpha>And 1, taking alpha =5 optimal values through experimental verification. Taking the (2, 1) DCT block's element (0, 0) coefficient as an example, the desired modification direction is r, as shown in FIG. 4 _o = +1, then the corresponding intra-block penalty factor is:

the cost value of the element in the DCT block can be updated by substituting the above coefficients into the following equation:

wherein the content of the first and second substances,

is the updated cost value of the element (0, 0) coefficient of (2, 1) DCT block under the condition of different embedding modification directions,

is its initial cost value. Therefore, when the actual modification direction is the same as the desired modification direction, the distortion cost value decreases, and the possibility of embedding increases; conversely, when the actual modification direction is opposite to the desired modification direction, the distortion cost value is increased, thus ensuring the realityThe actual embedding modification is as close as possible to the optimal embedding modification.

More specifically, as shown in fig. 3 and 4, the step S4 specifically includes:

s41: aiming at any DCT coefficient, constructing a four-adjacent domain model by using elements with the same mode in adjacent DCT blocks, wherein the adjacent DCT blocks are four, namely an upper DCT block, a lower DCT block, a left DCT block and a right DCT block which are directly connected with the current DCT block; thus, as can be seen from the above-described division of the sub-picture in units of 8 × 8DCT blocks, for any one DCT block of the sub-picture there are four adjacent DCT blocks from the other sub-picture.

S42: embedding modified graph R' ₁ And sub-diagram X ₂ Initialization embedding modification graph R ₂ Merging and traversing the merged embedded modified graph R ₂ Is traversed starting from the (0, 0) element, according to the embedded modification map R' ₁ Acquiring embedded modification combinations of the same pattern elements in 4 adjacent DCT blocks;

s43: combining with BBC criterion, updating the mode according to the expected horizontal and vertical updating modes of the mode ₂ Updating the cost value of the elements in the subgraph, wherein the specific calculation formula is as follows:

(p,q)∈{(u-1,v),(u+1,v),(u,v-1),(u,v+1)}；

wherein the content of the first and second substances,

representing the updated cost value of the (m, n) th DCT coefficient in the (u, v) th DCT block; w is a _inter Representing an inter-block penalty coefficient;

representing an initial cost value of the (m, n) th DCT coefficient in the (u, v) th DCT block;

homomorphism in four neighborhood DCT blocks representing a current DCT blockCost value of coefficient, sub-graph X ₁ Cost value C 'after intra-block optimization updating' ₁ Element (p, q) represents the coordinates of the four neighborhood DCT blocks; wherein the inter-block penalty coefficient w _inter Like the intra-block penalty factor, the direction d is also updated as desired _e And the actual embedding modification direction d _a Determined, the values of which are shown in Table 2, wherein the parameter β>1, and experimentally determined β =2.

TABLE 2 DCT inter-Block penalty coefficients w _inter (d _e ,d _a ) Watch (CN)

Wherein the desired modification direction is determined according to the conclusion of the BBC policy and the embedded modification direction of the neighborhood elements. Taking (0, 1) element of (1, 2) DCT block as an example, as can be seen from FIG. 5, the horizontal direction in this mode expects asynchronous update mode, so when updating the cost value according to the correlation of left-neighboring domain elements, the expected modification direction should be opposite to the left-neighboring domain element embedding modification direction, and the figure shows that

It is therefore desirable to modify the direction to d _e = -1, the coefficient with smaller embedding distortion cost is more likely to be embedded, so the penalty factor based on left neighbor correlation should be

Then based on the correlation of the left-neighbor elements, the cost value update for (0, 1) elements in the (1, 2) DCT block is:

wherein the content of the first and second substances,

d _a e { -1,0, +1} represents the updated cost value,

the value of the initial cost is expressed,

representing the initial cost value of the same pattern element in the left neighborhood DCT block, the other three neighborhoods do the same.

S44: go back to step S42 to update the next element until sub-graph X ₂ The cost values of all DCT coefficients are updated and are C' ₂ Representing a set of these new cost values;

s45: c 'new cost value' ₂ And subfigure X ₂ Sending the mixture into an analog simulation embedder again to obtain a secret carrier picture Y' ₂ 。

More specifically, the step S33 specifically includes:

s331: construction of a wavelet direction filter bank K = { K ] using DB-8 decomposition filters, i.e., one-dimensional low-pass filter l and high-pass filter h ₁ ,K ₂ ,K ₃ In which K is ₁ ＝l·h ^T ，K ₂ ＝h·l ^T ，K ₃ ＝h·h ^T ；

S332: for the current embedded modification combination, sequentially obtaining the spatial variation of each modified element (i, j) through two-dimensional inverse DCT transformation, namely:

wherein:

wherein, the first and the second end of the pipe are connected with each other,

representation modificationThe current DCT block (i, j) element corresponds to the spatial domain variation of the spatial domain (s, t) position pixel;

s333: and accumulating the spatial domain changes corresponding to all element modifications in the current embedded modification combination, namely:

s334: calculating the relative change of the wavelet domain generated by the current embedding modification combination in three directions, specifically:

B ^(k) ＝W*K _k ,k＝{1,2,3}；

wherein: w denotes the spatial impact of the current embedding modification combination, K _k Represents the k-th directional filter;

s335: the wavelet domain influence in three directions is integrated to obtain the distortion cost value C of the wavelet domain of the current embedded combination _DCT The method specifically comprises the following steps:

wherein: (u, v) is BETA ^(k) And (u, v) coordinates of 0-22.

More specifically, as shown in fig. 6, the process of JPEG image compression in step S11 specifically includes:

s111: 8 multiplied by 8 blocking the original spatial domain image;

s112: performing DCT (discrete cosine transformation) on the partitioned image and quantizing the partitioned image;

s113: and performing zig-zag scanning entropy coding on the quantized image to obtain a compressed JPEG image.

More specifically, the JPEG image symmetric steganography algorithm in the step S21 comprises one or more of J-UNIWARD, UERD and GUED.

Example 2

On the basis of embodiment 1, in order to compare the security performance of the JPEG asymmetric digital image steganography method provided by the present invention, the present embodiment performs a performance test.

More specifically, all performance tests in the invention are based on a common image library Bossbase ver1.01 in digital image steganography, and 10000 gray-scale images with the size of 512 multiplied by 512 are obtained. First, to construct a carrier image, the above images are compressed into JPEG images with quality factors of 75 and 95, respectively. And then, selecting three JPEG digital image steganography algorithms which are the most safe at present, and respectively carrying out cost value initialization for J-UNIWARD, GUED and BET. Then, three steganalysis algorithms GFR with highest JPEG steganalysis performance are selected ^[9] ,CC-JRM ^[10] And SCA-GFR ^[11] . Classification error rate P _E As an index to measure the performance of the algorithm, the classification error rate is averaged over ten verifications. In addition, this embodiment also compares the security performance of the currently known JPEG asymmetric steganographic Dejoin _ J.

The performance results obtained when the carrier image quality factor is 75 are shown in tables 3-5, and it can be known from the tables that the IEO-SBC cost value updating strategy of the JPEG asymmetric digital image steganography method provided by the invention has improved performance under the initialization of the three algorithms and basically exceeds Dejoin _ J.

TABLE 3 classification error Rate (%) of GFR characteristic test under Q75

Wherein, the BET-HILL represents that the cost value of a hollow domain in the BET algorithm is calculated by adopting the HILL algorithm.

TABLE 4 Classification error Rate (%) for CC-JRM feature detection at Q75

TABLE 5 Classification error Rate (%) for SCA-GFR feature detection under Q75

When the image quality factor of the carrier is 95, the performance of the IEO-SBC and the performance of the SBC only strategy are respectively verified through experiments, and the result shows that the performance difference is not large, so that the intra-block optimization step is omitted under the condition of ensuring the algorithm security, the cost value is directly updated based on the DCT inter-block correlation, and the final performance is shown in tables 6-8. As can be seen from the table, the improvement of the security performance is still ensured in the case of using only the SBC. On this basis, we respectively test the average time of each image used for updating the cost value by the invention and the Dejoin _ J algorithm when Q95 is performed, as shown in Table 9, under the condition of ensuring the safety performance, the cost value updating time used by the invention is half of that of Dejoin _ J, the calculation complexity is lower, the time consumption is less, and the optimization speed is faster.

TABLE 6 Classification error Rate (%) for GFR characteristic test under Q95

TABLE 7 Classification error Rate (%) for CC-JRM feature detection under Q95

TABLE 8 Classification error Rate (%) for SCA-GFR feature detection under Q95

TABLE 9 average elapsed time for cost value update for each graph at Q95

Algorithm	Dejoin_J	SBC
			Time	4.743s	2.354s

It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

[1]Holub V,Fridrich J,Denemark T.Universal distortion function for steganography in an arbitrary domain[J].Eurasip Journal on Information Security,2014,no.1,pp.1-13,2014.

[2]L.Guo,J.Ni,and Y.Q.shi,“Using statistical image model for JPEG steganography:Uniform embedding revisited,”IEEE Transactions on Information Forensics and Security,vol.10,no.12,pp.2669-2680,2015.

[3]W.Su，J.Ni，X.Li，Y.Shi，“A New Distortion Function Design for JPEG Steganography Using the Generalized Uniform Embedding Strategy，”IEEE Transactions on Circuits and Systems for Video Technology，vol.28，no.12，pp.3545-3549，2018.

[4]X.Hu,J.Ni,and Y.Q.Shi,“Efficient jpeg steganography using domaintransformation of embedding entropy,”IEEE Signal Processing Letters,vol.25,no.6,pp.773–777,2018.

[5]B.Li,M.Wang,S.Tan,and J.Huang,“A strategy of clustering modification directions in spatial image steganography,”IEEE Transactions on Information Forensics and Security,vol.10,no.9,pp.1905-1917,2015.

[6]T.Denemark and J.Fridrich,“Improving steganographic security by synchronizing the selection channel,”in Proc.ACM,the 3rd ACM Workshop on Information Hiding and Multimedia Security,2015,pp.5-14.

[7]Weixiang Li,Weiming Zhang,Kejiang Chen,Wenbo Zhou and Nenghai Yu."Defining Joint Distortion for JPEG Steganography".Proceedings of the 6th ACM Workshop on Information Hiding and Multimedia Security,ACM,2018.

[8]W.Zhang,Z.Zhang,L.Zhang,H.Li,and N.Yu,“Decomposing joint distortion for adaptive steganography,”IEEE Transactions on Circuits and Systems for Vedio Technology,vol.27,no.10,pp.2274-2280,2017.

[9]X.Song,F.Liu,C.Yang,X.Luo,and Y.Zhang,“Steganalysis of adaptive JPEG steganography using 2D Gabor filters,”in Proc.ACM,the 3rd ACM Workshop on Information Hiding and Multimedia Security,2015,pp.15-23.

[10]J.Kodovsk’y and J.Fridrich,“Steganalysis of JPEG images using rich models,”in Proc.SPIE,Electronic imaging,Media Watermarking,Security,and Forensics XIV,vol.8303,2012,pp.0A 1-13.

[11]T.Denemark,M.Boroumand,and J.Fridrich,“Steganalysis features for content-adaptive JPEG steganography,”IEEE Transactions on Information Forensics and Security,vol.11,no.8,pp.1736-1746,2016.

Claims (7)

1. A JPEG asymmetric digital image steganography method is characterized by comprising the following steps:

s1: obtaining a JPEG gridCarrier image X of the formula, dividing the carrier image into two interleaved subpictures X ₁ 、X ₂ ；

S2: method for obtaining initialization distortion cost value C of carrier image based on JPEG image symmetric steganography algorithm _ori And initializing an embedding modification map R ₁ 、R ₂ ；

S3: optimization of R ₁ Updating the sub-picture X in the case of intra-block embedded modification of each DCT block ₁ Obtaining new carrier density graph Y by embedding analog simulation to the cost value of each element ₁ 'and New Embedded modified map R' ₁ ；

S4: constructing a four-neighbor domain model incorporating an embedding modification map R' ₁ And BBC policy update X ₂ The cost value of each element in the graph is embedded by analog simulation to obtain a new carrier density graph Y ₂ '；

S5: the carrier density is shown as Y ₁ ' and Carrier map Y ₂ Merging to obtain a complete secret-carrying image Y ', sending the complete secret-carrying image Y' into a steganography analyzer, and detecting the safety performance of the proposed JPEG asymmetric steganography algorithm;

the step S3 specifically comprises the following steps:

s31: statistical embedding modification graph R ₁ The modified number in the current DCT block is denoted by symbol N, and a constraint condition whether the DCT block is optimized is set, specifically:

s32: listing all possible embedding modification combinations, and randomly selecting half of the embedding modification combinations in order to reduce the computational complexity, namely, the total number of the embedding modification combinations is K =0.5 × 2 ^N ；

S33: respectively calculating spatial domain embedding distortion cost C corresponding to K modification combinations of the current DCT block _DCT ；

S34: selecting minimum distortion cost C from K combined spatial domain embedding distortion costs _{DCT_min} The corresponding embedding modification combination is used as the optimal embedding modification combination of the DCT, that is, each DCT coefficient in the current DCT block is obtainedA desired modification direction;

s35: updating the +1/-1 cost value of each coefficient in the current DCT block according to the optimal embedded modification combination, specifically: setting an intra-block penalty factor w _inner (r _o ,r _a ) And updating a formula of the obtained cost value:

wherein r is _o Indicates a desired modification direction, r _a Modifying the direction for the actual; (u, v) represents the coordinates of the DCT block, which take values of an integer in the range of 0. Ltoreq. U.ltoreq.63, 0. Ltoreq. V.ltoreq.63 for a 512 x 512 image; (m, n) represents the coordinates of the coefficients in the DCT block, and the values are integers of 0-m 7, 0-n 7;

represents an initial cost value of the (m, n) coefficient within the (u, v) th DCT block;

representing the new cost value of the updated (m, n) coefficient in the (u, v) th DCT block; from C' ₁ Representing a set of these new cost values;

s36: will be of cost value C' ₁ And subfigure X ₁ Sending the data into an analog simulation embedder to obtain a new secret carrier graph Y ₁ ', obtaining a sub-image X by subtracting the carrier image X and then taking a symbol for operation ₁ New embedded modified graph R' ₁ 。

2. The JPEG asymmetric digital image steganography method according to claim 1, wherein: the specific steps of the step S1 are as follows:

s11: JPEG image compression is carried out on the original airspace image to obtain a JPEG image which is compressed and coded under the condition of different quality factors and is used as a carrier image X;

s12: dividing the carrier image X into two interleaved sub-images X ₁ 、X ₂ Division unit of 8 × 8DCAnd T blocks.

3. The JPEG asymmetric digital image steganography method according to claim 2, wherein: the step S2 specifically comprises the following steps:

s21: calculating initialization distortion cost value C of carrier image X based on JPEG image symmetric steganography algorithm _ori ；

S22: will initialize the distortion cost value C _ori Sending the carrier image X into an analog simulation embedder to obtain an initialized carrier image Y;

s23: carrying out difference on the secret-carrying image Y and the carrier image X to obtain a residual image D, wherein the residual image D contains embedded and modified amplitude and direction information;

s24: ignoring the magnitude information of the embedding modification of the residual image D, and keeping the direction information of the embedding modification to obtain a carrier image X embedding modification image R, wherein R = sgn (D); obtaining two subgraphs simultaneously as R ₁ 、R ₂ Corresponding to an initialization distortion cost value of C _{ori_1} 、C _{ori_2} 。

4. The JPEG asymmetric digital image steganography method according to claim 3, wherein: the step S4 specifically comprises the following steps:

s41: for any DCT coefficient, constructing a four-adjacent domain model by using elements with the same mode in the DCT blocks adjacent to the DCT coefficient, wherein the adjacent DCT blocks are respectively four, namely an upper DCT block, a lower DCT block, a left DCT block and a right DCT block which are directly connected with the current DCT block;

s42: will embed the modification map R ₁ ' with sub-graph X ₂ Initialization embedding modification graph R ₂ Merging and traversing the merged embedded modified graph R ₂ Each element of each DCT block of (c), starting from the (0, 0) element, according to an embedding modification map R ₁ ' obtaining an embedded modified combination of the same pattern elements in the neighboring 4 DCT blocks;

s43: combining with BBC criterion, updating the mode according to the expected horizontal and vertical updating modes of the mode ₂ Updating the cost value of the elements in the subgraph, wherein the specific calculation formula is as follows:

(p,q)∈{(u-1,v),(u+1,v),(u,v-1),(u,v+1)}；

wherein the content of the first and second substances,

representing the updated cost value of the (m, n) th DCT coefficient in the (u, v) th DCT block; w is a _inter Representing an inter-block penalty coefficient;

representing an initial cost value of the (m, n) th DCT coefficient in the (u, v) th DCT block;

the cost value of the same-mode coefficient in four neighborhood DCT blocks of the current DCT block is represented as a subgraph X ₁ Cost value C 'after intra-block optimization updating' ₁ Element (b), (p, q) represents the coordinates of the four neighborhood DCT blocks;

s44: go back to step S42 to update the next element until sub-graph X ₂ The cost values of all DCT coefficients are updated and are C' ₂ Representing a set of these new cost values;

s45: the new cost value C' ₂ And subfigure X ₂ Re-sending the data into the analog simulation embedder to obtain a carrier density subgraph Y ₂ '。

5. The JPEG asymmetric digital image steganography method according to claim 1, wherein: the step S33 specifically includes:

s331: construction of a wavelet direction filter bank K = { K ] using DB-8 decomposition filters, i.e., one-dimensional low-pass filter l and high-pass filter h ₁ ,K ₂ ,K ₃ In which K is ₁ ＝l·h ^T ，K ₂ ＝h·l ^T ，K ₃ ＝h·h ^T ；

S332: for the current embedded modification combination, sequentially obtaining the spatial variation of each modified element (i, j) through two-dimensional inverse DCT transformation, namely:

wherein:

wherein the content of the first and second substances,

representing the spatial variation of the pixel of the spatial (s, t) position corresponding to the element of the modified current DCT block (i, j);

s333: and accumulating the spatial domain changes corresponding to all element modifications in the current embedded modification combination, namely:

s334: calculating the relative change of the wavelet domain generated by the current embedding modification combination in three directions, specifically:

B ^(k) ＝W*K _k ,k＝{1,2,3}；

wherein: w denotes the spatial impact of the current embedding modification combination, K _k Represents the k-th directional filter;

s335: the wavelet domain influence in three directions is integrated to obtain the distortion cost value C of the wavelet domain of the current embedded combination _DCT The method specifically comprises the following steps:

wherein: (u, v) is BETA ^(k) And (u, v) coordinates of 0. Ltoreq. U and v. Ltoreq.22.

6. The JPEG asymmetric digital image steganography method according to claim 2, wherein: in step S11, the JPEG image compression process specifically includes:

s111: 8 multiplied by 8 blocking the original spatial domain image;

s112: performing DCT (discrete cosine transform) on the partitioned image and quantizing the partitioned image;

s113: and performing zig-zag scanning entropy coding on the quantized image to obtain a compressed JPEG image.

7. The JPEG asymmetric digital image steganography method according to claim 2, wherein: the JPEG image symmetric steganography algorithm in the step S21 comprises one or more of J-UNIWARD, UERD and GUED.

Images