CN114820380A - Spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance - Google Patents
Spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance Download PDFInfo
- Publication number
- CN114820380A CN114820380A CN202210521584.1A CN202210521584A CN114820380A CN 114820380 A CN114820380 A CN 114820380A CN 202210521584 A CN202210521584 A CN 202210521584A CN 114820380 A CN114820380 A CN 114820380A
- Authority
- CN
- China
- Prior art keywords
- image
- adv
- carrier image
- carrier
- disturbance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000003044 adaptive effect Effects 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 230000009191 jumping Effects 0.000 claims description 4
- AYFVYJQAPQTCCC-GBXIJSLDSA-N L-threonine Chemical compound C[C@@H](O)[C@H](N)C(O)=O AYFVYJQAPQTCCC-GBXIJSLDSA-N 0.000 claims description 3
- 238000003064 k means clustering Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 8
- 230000002708 enhancing effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006555 catalytic reaction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 208000011580 syndromic disease Diseases 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for enhancing an airspace steganographic carrier image based on content self-adaption disturbance resistance, which comprises the following steps of: calculating the multi-granularity image texture characteristics of the carrier image; clustering the carrier images according to pixel values, and segmenting the images according to semantic category labels; calculating a weighted mask of the carrier image; calculating content adaptive countermeasure disturbance; and adding the content self-adaptive anti-disturbance into the carrier image to obtain an enhanced carrier image, and adding the real secret information to perform steganography to obtain a final enhanced secret-carrying image. The invention adds disturbance noise to the image texture rich area according to the characteristics of the carrier image, and improves the concealment of the disturbance noise. Therefore, the method and the device can successfully attack the deep learning-based steganalysis network under the condition of a small amount of disturbance noise, and can reduce the risk of detecting the anti-disturbance noise.
Description
Technical Field
The invention relates to the technical field of multimedia security, in particular to a spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance.
Background
With the rapid development of communication technology and network information technology, digital images have become a common information exchange method in people's daily life. The digital image has high redundancy, and if some pixel points of the image are modified, the ornamental value of people on the image cannot be influenced, so that the image has natural advantages in carrying secret information for communication. Image steganography is an important covert communication mode, and secret information is mainly embedded into a common digital image (called a carrier image), and the image embedded with the secret information (called the carrier image) is transmitted in a public channel without causing the doubt of a third party supervisor (such as a steganalyser). The image steganography and steganography analysis technologies are always in a game state and develop in opposition to each other. With the rise of deep learning in computer vision, the steganalysis technology based on deep learning is rapidly developed, which brings great challenges to the image steganalysis technology. At present, in order to further resist a deep learning-based steganalysis method, a steganographer provides two main flow modes: the image steganography method based on the generation of the countermeasure network and the image steganography method based on the countermeasure sample.
Among the patents currently published, there are the following patents which address similar problems to the present method: the publication number is CN108346125B, and the title is a patent of airspace image steganography method and system based on generation of countermeasure network. The method mainly comprises the steps of generating a modified probability map from a carrier image through a U-Net network, simulating the modified probability map by combining a hyperbolic sine function to generate a secret-carrying image, simultaneously, mutually confronting with a steganalysis network, obtaining a steganography probability generation model through training, and finally, combining STC (syndrome time coding) to obtain the secret-carrying image. The method effectively improves the security performance of image steganography, but still has the following defects: 1) because the network structure is huge, the model is still unstable in the training process and is difficult to converge; 2) the generative model is data driven and therefore relies on a training data set, with the security of steganography of the image still being low for different data sets.
Reference [1] (Yiwei Zhang, Weiming Zhang, Kejiang Chen, Jianying Liu, Yujia Liu, and Nenghai Yu.2018. adaptive example against new latent image present network based on disturbance noise [ C ]// Proceedings of the 6th ACM works hop on Information Hiding and Multimedia security.2018:67-72.) discloses a method of adding disturbance noise in a carrier image by using gradient Information of a steganalysis network for an input image and iteratively adding to the carrier image, generating a disturbance noise image, then adding secret Information, generating a secret image and inputting the generated secret image into the steganalysis network until the target steganalysis network cannot judge. The method effectively improves the security performance of the steganography, has higher steganography efficiency, and still has the following defects: 1) because disturbance noise is added in a plurality of iterations until the steganalysis is successfully fooled, the disturbance noise added in the method is large in quantity; 2) the disturbance noise adding mode does not consider image characteristic information and carries out global addition on the image; the defects of the two points show that the scheme is easy to leave relatively obvious disturbance traces in the final generated dense image (especially in the flat area of the image texture).
Reference [2] (Qin C, Zhang W, Dong X, et al, applied gradient based on sparse noise enhancement [ J ]. Journal of Visual Communication and Image reproduction, 2021,80:103325.) discloses a method of adding sparse disturbance noise based on a gradient map to a carrier Image, which visualizes gradient information of an input Image using a steganalysis network, selects a part of values with large Image gradients to iteratively add disturbance noise, generates a confrontation carrier Image, then performs steganography to obtain a carrier Image, and finally inputs the carrier Image into a steganalysis network for judgment until the steganalysis network is successfully fooled. The method can further reduce the redundant points of disturbance noise while maintaining steganography security, but the method still has the defects that: 1) in the process of visualizing the gradient information, the method can see that the points with larger gradient values still have parts in the image flat area, and obvious disturbance traces are still easy to leave after disturbance noise is added in the parts; 2) the method adopts a steganalysis network to calculate the gradient value of an input image, but the gradient value still has no clear physical meaning at present, and the size selection is difficult to define.
Disclosure of Invention
The invention aims to provide a spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance.
The technical scheme for realizing the purpose of the invention is as follows:
a spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance comprises
Step 1: calculating the multi-granularity image texture characteristics of the carrier image c;
1.1 calculating the fine grain texture description value L of each pixel point of the carrier image c fine (-) to obtain a fine-grained texture description feature matrix L fine (ii) a Wherein, the pixel point (i) o ,j o ) Fine grain texture description value L of fine (. phi.) is
i o And j o Respectively representing pixel points (i) o ,j o ) The lateral and longitudinal coordinates of; r denotes a pixel point (i) o ,j o ) Randomly selecting one pixel point as a starting point, and numbering according to a clockwise or anticlockwise direction; x is the number of o And x r Respectively representing pixel points (i) o ,j o ) And the pixel values of the neighborhood pixels; s (-) represents a symbolic function,
1.2 calculating the coarse grain texture description value L of each pixel point of the carrier image c coarse (. to) to obtain a coarse-grained texture description feature matrix L coarse (ii) a Wherein, the pixel point (i) o ,j o ) Coarse grain texture description value L of coarse (. phi.) is
The count is 0/1 or 1/0 jumping times of a bit string obtained after the value of 8 neighborhood pixels is taken through a sign function s (·), and thre is a threshold value;
1.3 calculating the Multi-granularity image texture feature matrix of the Carrier image c
L＝L fine +L coarse ；
Step 2: carrying out k-means clustering on the carrier image c according to the pixel values to obtain a semantic category label of each pixel value; segmenting according to the semantic category labels to obtain a classified semantic category label M matrix;
and step 3: calculating a weighted mask of the carrier image c;
3.1 calculating the texture of each image block; the image blocks are a set of pixel points (i, j) with the same semantic category label value M (i, j);
wherein the texture of the d-th image block
R d Set of pixel values, N, representing the d-th image block d Number of pixel values, L, representing the d-th image block i,j Expressing texture values of pixel points (i, j) of the carrier image c, namely values of ith row and jth column in a multi-granularity image texture feature matrix L of the carrier image c;
3.2 calculating the weighted mask matrix of the Carrier image c
Wherein,
and 4, step 4: calculating content adaptive countermeasure disturbance;
4.1 steganography is carried out on the carrier image c to obtain a secret carrier image s; inputting the carrier image c and the secret image s into a steganalysis network phi c,s (theta; t) training to obtain a steganalysis network model;
4.2 initializing the countermeasure disturbance δ adv Will oppose the disturbance delta adv Adding the obtained mixture into a carrier image c to obtain an enhanced carrier image c adv (ii) a Adding random secret information pair c adv Steganography is carried out to obtain the current enhanced secret-carrying image s adv ；
4.3 in the process of backward propagation of the steganalysis network model, solving the current enhanced secret-carrying image s adv Of the gradient value, updating the counterdisturbance, i.e.
Wherein, | | δ adv || 2 Is the 2-norm of the current countermeasure disturbance, k is the weighting coefficient, L D (f c,s (theta; t)) is a steganalysis network
Cross entropy loss function of (1);
4.4 Using weighted mask matrix
Weighting the updated counterdisturbance to obtain weighted counterdisturbance, wherein the weighted counterdisturbance is used as the current counterdisturbance:
4.5 adding the current counterdisturbance δ to the Carrier image c adv Obtaining an enhanced carrier image c adv Adding random secret information pair c adv Steganography is carried out to obtain the current enhanced secret-carrying image s adv (ii) a Using steganalysis networks
Determining a current enhanced secret-carrying image s adv : if steganalysis network
The current enhanced secret-carrying image s adv If the carrier image c is determined, the current counterdisturbance delta is made adv Adaptive countermeasure disturbance for the content, otherwise return to step 4.3;
and 5: adapting content against disturbance delta adv Adding the obtained mixture into a carrier image c to obtain an enhanced carrier image c adv Adding true secret information pair c adv Steganography is carried out to obtain a final enhanced secret-carrying image s adv 。
Preferably, the steganography adopts HUGO, WOW, UNIWARD or HILL algorithm; the steganalysis network
XuNet, YeNet, SRNet or ZhuNet.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention adds disturbance noise to the image texture rich area according to the characteristics of the carrier image, and improves the concealment of the disturbance noise. Therefore, the method and the device can successfully attack the deep learning-based steganalysis network under the condition of a small amount of disturbance noise, and can reduce the risk of detecting the anti-disturbance noise.
2. The invention constructs a weighted disturbance resisting mask mode. Firstly, an image texture description method based on multi-granularity is designed, and a pixel-by-pixel fine granularity and image feature statistics coarse granularity mode is combined for describing image texture features, so that compared with the mode of singly adopting the pixel-by-pixel fine granularity to describe image textures, the method is more robust to tiny noise. Secondly, segmenting the image according to the semantic features of the image, selecting an image texture information rich block as a candidate disturbance position, and constructing a disturbance mask to enable the anti-disturbance noise to have the self-adaption performance of the image content. Therefore, the method is more suitable for application scenes for accurately performing image steganography.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2a, 2b and 2c are schematic diagrams of enhanced carrier image visualization of the present invention.
Detailed Description
The invention is further illustrated by the following specific examples.
As shown in fig. 1, the method for enhancing an image of a spatial domain steganographic carrier based on content adaptive disturbance rejection includes:
step 1: calculating the multi-granularity image texture characteristics of the input carrier image c, wherein the method comprises the following steps:
step 1.1: and comparing pixel values of 8 neighborhood pixel points pixel by pixel aiming at the carrier image, wherein if the central pixel value is larger than the neighborhood pixels, the pixel value is 1, and otherwise, the pixel value is 0. In this embodiment, the upper left corner of the neighborhood is set as the starting point, 8 values 0/1 are recorded counterclockwise as a bit string with a length of 8, and the bit string is converted into a decimal number to represent the texture value of the current pixel, so that the fine-grained texture description value L of each pixel is obtained fine (. cndot.) is:
wherein i o And j o Respectively representing the central pixel coordinate value, x o And x r Respectively representing gray values of a central pixel point and a neighborhood pixel point, and s (-) represents a symbolic function:
step 1.2: based on step 1.1Setting the upper left corner of the neighborhood value as a starting point, recording 8 neighborhood values of each pixel as a bit string with the length of 8 counterclockwise, counting the jumping times count of the bit string 0/1 or 1/0, if the jumping times are smaller than a threshold value thre, the texture complexity is 1, otherwise, the texture complexity is accumulated, the texture complexity is a coarse-grained texture description value, and the coarse-grained texture description value L of each pixel is coarse (. cndot.) is:
step 1.3: the fine-grained texture description characteristics L of the images respectively obtained in the step 1.1 and the step 1.2 fine And image coarse grain texture description feature L coarse And adding pixel by pixel to obtain the final texture feature L of the image: l ═ L fine +L coarse Wherein L is fine ，L coarse And L each represents a matrix having the same size as the original image.
Step 2: carrying out k-means clustering on the input carrier image c according to the pixel values to obtain a semantic category label of each pixel value; and (4) segmenting according to the semantic category label to obtain a classified semantic category label M (the size of the matrix M is the same as that of the original image), wherein the set of the pixel points (i, j) with the same semantic label value M (i, j) belongs to the same image block.
And step 3: inputting the texture value matrix L of the carrier image c obtained in the step 1 and the semantic label M after the carrier image c obtained in the step 2 is classified, and calculating an image weighting mask, wherein the image weighting mask comprises the following steps:
step 3.1: assigning a value to the carrier image c by using the texture value matrix obtained in the step 1, and calculating a texture mean value of the same label of the image pixel (namely the texture mean value of each image block) according to the semantic label M obtained in the step 2, wherein the texture mean value is used for describing the texture of the current segmentation image block, and the texture L of the d-th image block d Comprises the following steps:
wherein R is d Set of pixel values, N, belonging to the d-th image block d Number of pixel values, L, representing the d-th image block i,j Representing a texture value with (i, j) image pixel coordinates.
Step 3.2: selecting an image block according to a threshold th, and selecting a sub-block with rich texture of the current image as a candidate weighting mask:
wherein,
representing the texture value with the resulting image pixel coordinates (i, j). Then:
(the matrix size is the same as the original carrier image size) represents the final resulting mask.
And 4, step 4: calculating disturbance noise, comprising:
step 4.1: steganography is carried out on an input carrier image c by adopting a traditional steganography method (such as a height undetectable algorithm HUGO, a wavelet weight algorithm WOW, a spatial domain unified relative wavelet algorithm UNIWARD, a high-pass low-pass filter algorithm HILL and the like) to obtain a corresponding secret-carrying image s, and the carrier and secret-carrying image pair are input into a steganography analysis network
(e.g., XuNet) [3] 、YeNet [4] 、SRNet [5] 、ZhuNet [6] Equal steganalysis network) to obtain a steganalysis network model;
reference [3 ]: xu, H. -Z.Wu, and Y. -Q.Shi, "Structural design of connected neural networks for hierarchical," IEEE Signal Process.Lett., May 2016, vol.23, No.5, pp.708-712.
Reference [4 ]: J.Ni, J.Ye, and Y.I.Yang, "Deep learning resonant representation for image segmentation," IEEE Trans. Inf. forces Security, Nov.2017, vol.12, No.11, pp.2545-2557.
Reference [5 ]: M.Boroum and, M.Chen, J.Fridrich, Deep residual network for stepana analysis of digital images, IEEE Trans.Inf.Forensecs Secur.14(5) (2019) 1181-.
Reference [6 ]: zhang R, Zhu F, Liu J, et al. depth-wise minor proportions and multi-level posing for an effective specific CNN-based catalysis [ J ]. IEEE Transactions on Information forms and Security,2019,15: 1138-.
Step 4.2: initializing the countering disturbance noise delta adv Will oppose the disturbance noise delta adv Adding to the carrier image c to obtain an enhanced carrier image c adv And adding a pair of random secret information c adv Performing steganography to obtain a current enhanced secret-carrying image s adv ；
Step 4.3: computing 2-norm [ Delta ] of the anti-disturbance noise adv || 2 As a constraint to minimize the countering disturbance noise. Meanwhile, calculating a cross entropy loss function L of the steganalysis network D (f c,s (theta; t)). For the current enhanced secret-carrying image s in the process of backward propagation of the steganalysis network model adv Obtaining a gradient value:
where k is a weighting coefficient, L D (. is a loss function of the steganalysis network;
step 4.4: for the disturbance value delta calculated in the step 4.3 adv Weighting the mask obtained in step 3
Obtaining a current disturbance value:
step 4.5: on the carrier imageAdding current anti-disturbance noise delta in c adv Obtaining the challenge vector c adv Then adding random secret information to steganographically obtain enhanced secret-carrying image s adv And will enhance the secret-carrying image s adv Input to an analyzer
If the steganalyser is determined
If the secret image s is judged as a carrier image, the anti-disturbance noise delta is output adv As content self-adaptive anti-disturbance noise, otherwise, continuously iterating the steps 4.3-4.5 to add the anti-disturbance noise;
and 5: adaptive disturbance noise resisting delta for the content obtained in the step 4 adv Adding the obtained mixture into a carrier image c to obtain an enhanced carrier image c adv And embedding real secret information, and performing steganography according to the traditional image steganography algorithm to obtain a final enhanced secret-carrying image s adv 。
Taking zhuet as an example, the invention and its safety performance ratio are as follows:
TABLE 1 false alarm rate p of different carrier enhancement methods for deep learning based steganalysis device ZhuNet MD (％)
In the table: the load (bpp, bit per pixel) refers to the secret information carrying capacity of the original steganography method, namely the number of bits of the average embedded secret information of each pixel in the carrier image; the modification rate is the ratio of the number of modifications of the confrontation sample to the carrier image.
False alarm rate p MD The definition is as follows:
wherein N is r Representation steganalysisThe network judges the secret-carrying image as the number of the carrier images, N t Indicating the number of all the secret images.
As shown in Table 1, the method of the invention can make the hidden write analyzer ZhuNet network based on deep learning have the highest false alarm rate and the best attack effect. For example, under the conditions that the load is 0.1bpp and the modification rate is 0.5, the method provided by the invention compares the SPS-ENH false alarm rate p which is the best method at present MD The improvement is 19.3 percent. In addition, with the improvement of the modification rate, the false alarm of all the methods is improved, which shows that when the anti-disturbance intensity is larger and larger, the successful attack can be basically realized on the steganalysis device ZhuNet network.
As shown in fig. 2a, 2b and 2c, fig. 2a is the original carrier image, the picture being selected from the BOSSBase data set. Fig. 2b shows the added noise immunity of the method of the present invention, the white point is the modified pixel area, the black point is the unmodified pixel area, and the modification amplitude is 1, so the graph shows the result after the pixel value is amplified by x255 times. Fig. 2c is the enhanced carrier image effect after adding the anti-disturbing noise to the carrier image of fig. 2 a. The white dashed box region represents an example of a region where the image texture is richer. As can be seen from FIG. 2, the disturbance information added by the present invention is mostly located in the image texture rich region.
Claims (2)
1. The spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance is characterized by comprising the following steps
Step 1: calculating the multi-granularity image texture characteristics of the carrier image c;
1.1 calculating the fine grain texture description value L of each pixel point of the carrier image c fine (-) to obtain a fine-grained texture description feature matrix L fine (ii) a Wherein, the pixel point (i) o ,j o ) Fine grain texture description value L of fine (. phi.) is
i o And j o Respectively representing pixel points (i) o ,j o ) The lateral and longitudinal coordinates of; r denotes a pixel point (i) o ,j o ) Randomly selecting one pixel point as a starting point, and numbering according to a clockwise or anticlockwise direction; x is the number of o And x r Respectively representing pixel points (i) o ,j o ) And the pixel values of the neighborhood pixels; s (-) represents a symbolic function,
1.2 calculating the coarse grain texture description value L of each pixel point of the carrier image c coarse (. to) to obtain a coarse-grained texture description feature matrix L coarse (ii) a Wherein, the pixel point (i) o ,j o ) Coarse grain texture description value L of coarse (. phi.) is
The count is 0/1 or 1/0 jumping times of a bit string obtained after the value of 8 neighborhood pixels is taken through a sign function s (·), and thre is a threshold value;
1.3 calculating the Multi-granularity image texture feature matrix of the Carrier image c
L＝L fine +L coarse ；
Step 2: carrying out k-means clustering on the carrier image c according to the pixel values to obtain a semantic category label of each pixel value; segmenting according to the semantic category labels to obtain a classified semantic category label M matrix;
and step 3: calculating a weighted mask of the carrier image c;
3.1 calculating the texture of each image block; the image blocks are a set of pixel points (i, j) with the same semantic category label value M (i, j);
wherein the texture of the d-th image block
R d Set of pixel values, N, representing the d-th image block d Number of pixel values, L, representing the d-th image block i,j Expressing texture values of pixel points (i, j) of the carrier image c, namely values of an ith row and a jth column in a multi-granularity image texture feature matrix L of the carrier image c;
3.2 computing the weighted mask matrix of the Carrier image c
Wherein,
and 4, step 4: calculating content adaptive countermeasure disturbance;
4.1 steganography is carried out on the carrier image c to obtain a secret carrier image s; inputting the carrier image c and the secret image s into a steganalysis network phi c,s (theta; t) training to obtain a steganalysis network model;
4.2 initializing the countermeasure disturbance δ adv Will oppose the disturbance delta adv Adding the obtained mixture into a carrier image c to obtain an enhanced carrier image c adv (ii) a Adding random secret information pair c adv Steganography is carried out to obtain the current enhanced secret-carrying image s adv ；
4.3 in hiddenIn the backward propagation process of the write analysis network model, the current enhanced secret-carrying image s is solved adv Of the gradient value, updating the counterdisturbance, i.e.
Wherein, | | δ adv || 2 Is the 2-norm of the current countermeasure disturbance, k is the weighting coefficient, L D (f c,s (θ; t)) is a steganalysis network
Cross entropy loss function of (1);
4.4 Using weighted mask matrix
Weighting the updated counterdisturbance to obtain weighted counterdisturbance, wherein the weighted counterdisturbance is used as the current counterdisturbance:
4.5 adding the current counterdisturbance δ to the Carrier image c adv Obtaining an enhanced carrier image c adv Adding random secret information pair c adv Steganography is carried out to obtain the current enhanced secret-carrying image s adv (ii) a Using steganalysis networks
Determining a current enhanced secret-carrying image s adv : if steganalysis network
The current enhanced secret-carrying image s adv If the image is determined as the carrier image c, the current counterdisturbance delta is made adv Adaptive countermeasure disturbance for the content, otherwise return to step 4.3;
and 5: content is transmitted toAdaptive countering disturbance delta adv Adding the obtained mixture into a carrier image c to obtain an enhanced carrier image c adv Adding true secret information pair c adv Steganography is carried out to obtain a final enhanced secret-carrying image s adv 。
2. The spatial domain steganography-based carrier image enhancement method based on content-adaptive disturbance rejection as claimed in claim 1, wherein the steganography employs HUGO, WOW, UNIWARD or HILL algorithm; the steganalysis network
XuNet, YeNet, SRNet or ZhuNet.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210521584.1A CN114820380B (en) | 2022-05-13 | 2022-05-13 | Spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210521584.1A CN114820380B (en) | 2022-05-13 | 2022-05-13 | Spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114820380A true CN114820380A (en) | 2022-07-29 |
| CN114820380B CN114820380B (en) | 2023-04-18 |
Family
ID=82515015
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210521584.1A Active CN114820380B (en) | 2022-05-13 | 2022-05-13 | Spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114820380B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115348360A (en) * | 2022-08-11 | 2022-11-15 | 国家电网有限公司大数据中心 | Self-adaptive embedded digital label information hiding method based on GAN |
| CN117131544A (en) * | 2023-10-27 | 2023-11-28 | 北京睿航至臻科技有限公司 | Data privacy protection method based on depth steganography |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537601A (en) * | 2014-12-23 | 2015-04-22 | 中山大学 | Gray level image aerial region steganography method based on nine grids |
| CN107133991A (en) * | 2017-03-17 | 2017-09-05 | 中山大学 | A kind of bianry image steganography method based on disturbance distortion and pixel selection |
| CN111222583A (en) * | 2020-01-15 | 2020-06-02 | 北京中科研究院 | Image steganalysis method based on confrontation training and key path extraction |
| US20200184592A1 (en) * | 2017-05-19 | 2020-06-11 | Google Llc | Hiding Information and Images via Deep Learning |
| CN111768325A (en) * | 2020-04-03 | 2020-10-13 | 南京信息工程大学 | Security improvement method based on generation of countermeasure sample in big data privacy protection |
| CN113538202A (en) * | 2021-08-05 | 2021-10-22 | 齐鲁工业大学 | Image steganography method and system based on generative steganography confrontation |
| CN113706636A (en) * | 2021-07-09 | 2021-11-26 | 重庆度小满优扬科技有限公司 | Method and device for identifying tampered image |
| CN114283080A (en) * | 2021-12-15 | 2022-04-05 | 复旦大学 | Multi-mode feature fusion text-guided image compression noise removal method |
-
2022
- 2022-05-13 CN CN202210521584.1A patent/CN114820380B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537601A (en) * | 2014-12-23 | 2015-04-22 | 中山大学 | Gray level image aerial region steganography method based on nine grids |
| CN107133991A (en) * | 2017-03-17 | 2017-09-05 | 中山大学 | A kind of bianry image steganography method based on disturbance distortion and pixel selection |
| US20200184592A1 (en) * | 2017-05-19 | 2020-06-11 | Google Llc | Hiding Information and Images via Deep Learning |
| CN111222583A (en) * | 2020-01-15 | 2020-06-02 | 北京中科研究院 | Image steganalysis method based on confrontation training and key path extraction |
| CN111768325A (en) * | 2020-04-03 | 2020-10-13 | 南京信息工程大学 | Security improvement method based on generation of countermeasure sample in big data privacy protection |
| CN113706636A (en) * | 2021-07-09 | 2021-11-26 | 重庆度小满优扬科技有限公司 | Method and device for identifying tampered image |
| CN113538202A (en) * | 2021-08-05 | 2021-10-22 | 齐鲁工业大学 | Image steganography method and system based on generative steganography confrontation |
| CN114283080A (en) * | 2021-12-15 | 2022-04-05 | 复旦大学 | Multi-mode feature fusion text-guided image compression noise removal method |
Non-Patent Citations (7)
| Title |
|---|
| JIE LUO 等: "Improving security for image steganography using content-adaptive adversarial perturbations" * |
| RUOHAN MENG 等: "High-Capacity Steganography Using Object Addition-Based Cover Enhancement for Secure Communication in Networks" * |
| VIPUL SHARMA 等: "Towards secured image steganography based on content-adaptive adversarial perturbation" * |
| 夏强 等: "基于普遍对抗噪声的高效载体图像增强算法" * |
| 李锦伟 等: "基于细粒度嵌入空间预留的密文域图像可逆信息隐藏方法" * |
| 陈君夫 等: "基于深度学习的图像隐写方法研究" * |
| 陈孟华 等: "基于注意力机制与生成对抗网络的彩色图像隐写算法" * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115348360A (en) * | 2022-08-11 | 2022-11-15 | 国家电网有限公司大数据中心 | Self-adaptive embedded digital label information hiding method based on GAN |
| CN115348360B (en) * | 2022-08-11 | 2023-11-07 | 国家电网有限公司大数据中心 | GAN-based self-adaptive embedded digital tag information hiding method |
| CN117131544A (en) * | 2023-10-27 | 2023-11-28 | 北京睿航至臻科技有限公司 | Data privacy protection method based on depth steganography |
| CN117131544B (en) * | 2023-10-27 | 2024-01-12 | 北京睿航至臻科技有限公司 | Data privacy protection method based on depth steganography |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114820380B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114820380B (en) | Spatial domain steganographic carrier image enhancement method based on content self-adaption disturbance resistance | |
| Li et al. | GAN-based spatial image steganography with cross feedback mechanism | |
| CN113222800A (en) | Robust image watermark embedding and extracting method and system based on deep learning | |
| CN112884758B (en) | Defect insulator sample generation method and system based on style migration method | |
| CN103034853A (en) | Universal steganalysis method for JPEG images | |
| Wang et al. | HidingGAN: High capacity information hiding with generative adversarial network | |
| CN115131188A (en) | Robust image watermarking method based on generation countermeasure network | |
| CN113723295A (en) | Face counterfeiting detection method based on image domain frequency domain double-flow network | |
| GB2614806A (en) | Method of crowd density estimation based on multi-scale feature fusion of residual network | |
| Zhou et al. | Adaptive digital watermarking scheme based on support vector machines and optimized genetic algorithm | |
| CN112861671A (en) | Method for identifying deeply forged face image and video | |
| CN115908095A (en) | Hierarchical attention feature fusion-based robust image watermarking method and system | |
| CN114359269A (en) | Virtual food box defect generation method and system based on neural network | |
| CN116091288A (en) | Diffusion model-based image steganography method | |
| CN116452401A (en) | Reversible robust watermark embedding and extraction model construction method for resisting image attack | |
| Liu et al. | Hiding functions within functions: Steganography by implicit neural representations | |
| CN115482139A (en) | Traceable deep learning model black box watermark scheme | |
| CN115131605A (en) | Structure perception graph comparison learning method based on self-adaptive sub-graph | |
| CN116757909B (en) | BIM data robust watermarking method, device and medium | |
| CN113628090A (en) | Anti-interference message steganography and extraction method and system, computer equipment and terminal | |
| CN113034332A (en) | Invisible watermark image and backdoor attack model construction and classification method and system | |
| Niu et al. | An image steganography method based on texture perception | |
| Meng et al. | A Novel Steganography Algorithm Based on Instance Segmentation. | |
| CN116112685A (en) | Image steganography method based on diffusion probability model | |
| Zhang et al. | LKAW: a robust watermarking method based on large kernel convolution and adaptive weight assignment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |