CN114782697B - Self-adaptive steganography detection method for anti-domain - Google Patents

Self-adaptive steganography detection method for anti-domain Download PDF

Info

Publication number
CN114782697B
CN114782697B CN202210467768.4A CN202210467768A CN114782697B CN 114782697 B CN114782697 B CN 114782697B CN 202210467768 A CN202210467768 A CN 202210467768A CN 114782697 B CN114782697 B CN 114782697B
Authority
CN
China
Prior art keywords
domain
adaptive
sub
field
steganography
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.)
Active
Application number
CN202210467768.4A
Other languages
Chinese (zh)
Other versions
CN114782697A (en
Inventor
王宏霞
章蕾
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan University
Original Assignee
Sichuan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Sichuan University filed Critical Sichuan University
Priority to CN202210467768.4A priority Critical patent/CN114782697B/en
Publication of CN114782697A publication Critical patent/CN114782697A/en
Application granted granted Critical
Publication of CN114782697B publication Critical patent/CN114782697B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

The invention discloses a self-adaptive steganography detection method in the field of countermeasures, which aims at the problem of carrier source mismatch in image steganography detection. The sub-domain adaptive classifier Y minimizes sub-domain adaptive loss while classifying to reduce the difference in the inter-domain related sub-domain feature distribution. The method is simple in calculation and easy to implement, can effectively improve the accuracy of the steganography detection model based on deep learning under the condition of carrier source mismatch, and has practical value.

Description

Self-adaptive steganography detection method for anti-domain
Technical Field
The invention relates to the technical field of information hiding, in particular to an adaptive steganography detection method in the field of countermeasures.
Background
Image steganography embeds secret information into an image, which can hide both the content of the secret information and the communication behavior. The purpose of steganography detection is to detect secret communications established by steganography. In recent years, many high-performance image steganography detection methods based on deep learning have emerged. Literature (Mehdi Boroumand, mo Chen, jessa Fridrich, IEEE Transactions on Information Forensics and Security,14 (5), 1181-1193) proposes an end-to-end depth residual network SRNet for steganography detection. When a steganographic detection model trained on one carrier source is used to detect images from a different carrier source, the detection error of the steganographic detection model is typically increased due to a mismatch between the two carrier sources, which is known as a carrier source mismatch problem. When the steganographic detection is performed, it is very difficult to obtain information such as an original carrier image, a steganographic algorithm, an embedding rate and the like adopted by a steganographic person for steganographic, that is, the difference between the training set and the test set is unavoidable, which is why the steganographic detection model is difficult to deploy in practical application. The carrier source mismatch problem and the domain adaptation problem in steganography detection are quite similar. The domain adaptive methods can be divided into two types, one is a metric-based domain adaptive method. The method comprises the steps of firstly mapping source domain and target domain features into a high-dimensional feature space, then measuring the distance between the mappings in the high-dimensional feature space based on a certain difference measure, and realizing the distribution alignment of the source domain and the target domain by minimizing the measure index of the inter-domain distribution difference. Literature (Ghifary M, kleijn W B, zhang M, pacific Rim International Conference on Artificial Intelligence, 2014:898-904) proposes a domain adaptive neural network (Domain adaptive neural network, daNN) that introduces a maximum mean difference (Maximum mean discrepancy, MMD) metric into a feed-forward neural network model. The MMD adaptation layer added after the feature layer is trained to reduce the distribution mismatch of the inter-domain features, and most of subsequent researches are developed based on the thought. Yet another approach learns a representation that can be both categorized and indistinguishable from which domain by selecting a resistance loss to minimize inter-domain distance. Literature (Ganin Y, ustinova E, ajakan H, et al the Journal ofMachine Learning Research,2016,17 (1): 2096-2030.) originally incorporated a challenge mechanism in domain adaptation to obtain universal features between domains by challenge with a domain arbiter. In this work, the learning objectives of the network are: the extracted features enable the classifier to correctly classify the sample as far as possible, and enable the discriminator to not distinguish whether the sample is from the source field or the target field. For the two domain adaptive methods described above, the use of the domain adaptive method alone against learning is insufficient to minimize the distance of the inter-domain feature distribution, and the use of the method alone of minimizing the metric lacks learning of domain independent features. Aiming at the carrier source mismatch problem in the image steganography detection, the invention utilizes the test set information, solves the carrier source mismatch problem by using a field self-adaption method, takes training set data as a source field and test set data as a target field, and improves the detection performance of a steganography detection model in the target field by minimizing the characteristic distribution difference between the source field and the target field.
Disclosure of Invention
In order to overcome the defects in the prior art and improve the accuracy of the steganography detection model under the mismatch condition, the invention provides an adaptive method for the domain of the antigen, which can enable the characteristics generated by the steganography detection model in the source domain and the target domain to be more similar, reduce the characteristic distribution difference between domains and improve the classification accuracy of the steganography detection model in the target domain.
The technical scheme for realizing the invention is as follows:
a self-adaptive steganography detection method for an anti-domain adopts an anti-learning strategy, and performs anti-learning with a steganography feature extractor F through a domain discriminator D, so that features generated by the steganography feature extractor F in a source domain and a target domain are more similar. The sub-domain adaptive classifier Y minimizes sub-domain adaptive loss while classifying to reduce the difference in the inter-domain related sub-domain feature distribution. Obtaining a steganography detection model MJ for classifying in the target field by simultaneously minimizing classification loss, counterloss and sub-field adaptive loss; the method comprises the following main steps:
step 1, taking a steganography detection model which is trained on the source field as a pre-training model M; firstly, removing a classification layer of the pre-training model M as a steganography feature extractor F; the tagged source domain data and the untagged target domain data are passed through the first layer of the steganographic feature extractor F to obtain features
Figure BDA0003625150410000021
And->
Figure BDA0003625150410000022
Step 2, the characteristics obtained in the step 1 are simultaneously sent into a classification branch and a discrimination branch; in the classification branch, the network structure of the sub-domain adaptive classifier Y is the same as the classification layer of the pre-training model M, and parameters are initialized. Sub-domain adaptive classifier Y detects feature f from source domain steganography sl And target field steganography detection feature f tl And respectively giving predicted values, and simultaneously calculating a classification loss L1 and a sub-field adaptive loss L2:
Figure BDA0003625150410000023
in n s The number of samples in the source domain; the classification loss L1 is the classifier predictive value
Figure BDA0003625150410000024
And Source Domain real Label->
Figure BDA0003625150410000025
Cross entropy loss between.
Figure BDA0003625150410000026
Wherein C is the label category, C is {0,1}, and C is the label category number; n is n s For the number of source field samples, n t The number of samples for the target field; w is the weight that the sample belongs to class c;
Figure BDA0003625150410000027
representing the inner product of the feature.
In the discrimination branch, the domain discriminator D is based on the input features
Figure BDA0003625150410000028
A label of the predicted domain is given, i.e. it is determined from which domain the feature comes. n is n t For the number of target field samples, the field prediction tag +.>
Figure BDA0003625150410000029
Sum field genuine label d i As the contrast loss L3:
Figure BDA00036251504100000210
step 3, calculating the total loss function
L＝L1+λL2+ωL3
λ and ω are trade-off parameters for sub-domain adaptive and countering losses, respectively; in the training process of the model, the parameters of the steganographic feature extractor F, the sub-domain self-adaptive classifier Y and the domain discriminator D are updated by minimizing the total loss and performing back propagation, so that the difference of the feature distribution of the source domain and the target domain is reduced.
And 4, stopping training when the set upper limit of the training times is reached, and combining the steganographic feature extractor F and the sub-field adaptive classifier Y to obtain a steganographic detection model MJ for classifying in the target field.
Further, the network structure of the domain identifier D in the step 2 is not unique, and can be adjusted according to actual situations.
Furthermore, the method can be applied to the space domain steganography detection and the JPEG domain steganography detection.
Compared with the prior art, the invention has the beneficial effects that,
1. the invention adopts an countermeasure learning strategy, and the countermeasure is carried out by introducing a domain discriminator and a steganography feature extractor, so that the source domain features and the target domain features generated by the steganography feature extractor are required to be similar as much as possible.
2. The distribution difference of the source domain features and the target domain features is reduced by calculating and minimizing sub-domain adaptive losses.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
FIG. 2 is a block diagram of an adaptive steganography detection method in the field of antibodies according to the present invention.
FIG. 3 is a graph comparing the experimental results of the present invention on UCIDv.2 data set with the prior art method.
Detailed Description
The invention will be further described with reference to the drawings and the specific examples.
The invention aims at the problem of carrier source mismatch in image steganography detection. Carrier source mismatch includes a variety of conditions such as dataset mismatch, steganographic algorithm mismatch, embedding rate mismatch, QF mismatch, image processing process mismatch, etc.
The specific steps of the invention as shown in fig. 1 are as follows:
S1：
Figure BDA0003625150410000031
representing a tagged source domain, comprising n s Sample with label->
Figure BDA0003625150410000032
Is sample->
Figure BDA0003625150410000033
Is a label vector of (a). Since steganography detection is a two-classification work, c=2. />
Figure BDA0003625150410000034
Representing a label-free target area. The steganographic detection model which has been trained on the source domain is taken as a pre-training model M.
As can be seen in connection with fig. 2, the classification layer of the pre-training model M is first removed as a steganographic feature extractor F; after the labeled source domain data and the unlabeled target domain data pass through the steganography feature extractor F, the source domain steganography detection feature F is obtained sl And target field steganography detection feature f tl The obtained characteristics are shared by the sub-domain adaptive classifier Y and the domain discriminator D, and are simultaneously sent into a classification branch and a discrimination branch.
S2: the field-adaptive approach using only countermeasure learning is not sufficient to minimize the distance of the inter-domain feature distribution, and the approach using only the minimization measure lacks learning of domain-independent features. Thus, in view of these limitations, the present invention proposes an anti-domain adaptive network for mismatch steganography detection. On the one hand, from the viewpoint of reducing inter-domain differences, the characteristic distribution of the related sub-domains is aligned, the inter-class distance is enlarged, and the intra-class distance is reduced. On the other hand, by constructing a domain discriminator to perform countermeasure learning with the steganography detection model, features generated by the model in two domains are more similar. As shown in FIG. 2, the anti-domain adaptive network provided by the invention mainly comprises a steganographic feature extractor F, a sub-domain adaptive classifier Y and a domain discriminator D, and is mainly aimed at improving pre-predictionAnd training the detection accuracy of the steganographic detection model under the mismatch condition. In the classification branch, the network structure of the sub-domain adaptive classifier Y is the same as the classification layer of the pre-training model M, and parameters are initialized. The sample passes through the first layer of the steganographic feature extractor F to obtain features
Figure BDA0003625150410000041
And->
Figure BDA0003625150410000042
The sub-domain adaptive classifier Y gives the predictive value for the source domain image +.>
Figure BDA0003625150410000043
By calculating->
Figure BDA0003625150410000044
Real label Y in source field s The cross entropy loss between them yields a classification loss L1: />
Figure BDA0003625150410000045
The classifier Y gives the predicted value of the target field image
Figure BDA0003625150410000046
And source domain feature f sl Target domain feature f tl Real label Y in source field s Together, used to calculate the sub-domain adaptive loss L2:
Figure BDA0003625150410000047
wherein c is the label category, c e {0,1}; n is n s For the number of source field samples, n t The number of samples for the target field; w is the weight that the sample belongs to class c;
Figure BDA0003625150410000048
representing the inner product of the feature;in the formula, sample x i The weights belonging to class c are calculated as:
Figure BDA0003625150410000049
y ic is the label vector y i C-th value of (c).
In the discrimination branch, the feature first goes through a gradient inversion layer (GRL) to simplify the challenge training process. The principle of the realization of the gradient inversion layer is as follows: the input is kept unchanged in the forward propagation process, and the gradient is multiplied by a negative constant value mu in the backward propagation process, so that the automatic negation of the gradient is realized to achieve the aim of countermeasure training. The role of the gradient inversion layer in forward and backward propagation can be expressed in terms of two pseudo functions:
forward propagation:
R μ (x)＝x
back propagation:
Figure BDA00036251504100000410
where I is an identity matrix. The domain discriminator D gives a label of the predicted domain from the inputted feature, i.e., judges from which domain the feature comes. n is n t For the number of target field samples, the field prediction labels are calculated
Figure BDA0003625150410000051
Sum field genuine label d i As the contrast loss L3:
Figure BDA0003625150410000052
s3: calculating the total loss function
L＝L1+λL2+ωL3
λ and ω are trade-off parameters for sub-domain adaptive and countering losses, respectively; in the training process of the model, the parameters of the steganographic feature extractor F, the sub-domain self-adaptive classifier Y and the domain discriminator D are updated by minimizing the total loss and carrying out back propagation, so that the difference of the feature distribution of the source domain and the target domain is reduced;
s4: and stopping training when the set upper limit of the training times is reached, and combining the steganographic feature extractor F and the sub-field adaptive classifier Y to obtain a steganographic detection model MJ for classifying in the target field.
The effectiveness of the method is described below in connection with a specific experiment
Given a pre-training model SRNet trained on a BOSSBase dataset with an S-uniwasd steganographic algorithm with an embedding rate of 0.4bpp (bits per pixel, bpp), it is possible that the detection accuracy of the pre-training model may be degraded to different extents due to differences between the training set and the test set when detecting S-uniwasd steganographic image pairs on other datasets, such as the ucidv.2 dataset, which belongs to dataset mismatch in carrier source mismatch problems. To verify the effectiveness of the present invention, data set mismatch experiments were performed on all three steganography algorithms, S-UNIWARD, WOW, HUGO. And randomly selecting 500 images on the BOSSBase data set as carrier images, and obtaining 500 secret-loaded images after steganography. The 500 pairs of images and labels are used as source fields. And randomly selecting 500 images on the UCIDv.2 data set as carrier images, and obtaining 500 secret-loaded images after steganography. The 500 pairs of images are taken as the target area.
Step 1, firstly, removing a classification layer of an SRNet pre-training model as a steganography feature extractor F; after the labeled source domain data and the unlabeled target domain data pass through the steganography feature extractor F, the source domain steganography detection feature F is obtained sl And target field steganography detection feature f sl The obtained features are simultaneously sent into a classification branch and a discrimination branch.
And 2, the network structure of the sub-field adaptive classifier Y is the same as that of the SRNet pre-training model, and is a full-connection layer, and the parameters of the full-connection layer are initialized. The sub-domain adaptive classifier Y gives a predictive value for the source domain image
Figure BDA0003625150410000053
By calculating->
Figure BDA0003625150410000054
Real label Y in source field s The cross entropy loss between the two is used for obtaining a classification loss L1, and a classifier Y gives a predicted value of the target field image +.>
Figure BDA0003625150410000055
And source domain feature f sl Target domain feature f tl Real label Y in source field s Together used to calculate the sub-domain adaptive loss L2. The domain identifier D consists of three linear layers to predict the domain label of the feature. The feature dimension of SRNet is 512 and the hidden layer dimension of domain arbiter D is set to 512. The domain discriminator D gives a label of the predicted domain based on the input characteristics, predicts the label by calculating the domain +.>
Figure BDA0003625150410000061
Sum field genuine label d i As the contrast loss L3.
And 3, calculating and minimizing a total loss function, carrying out back propagation, and updating parameters of the steganographic feature extractor F, the sub-domain self-adaptive classifier Y and the domain discriminator D.
And 4, after training 200 epochs, stopping training. Combining the steganographic feature extractor F and the sub-domain adaptive classifier Y results in a new SRNet model for classifying in the target domain.
As can be seen from the data set mismatch experimental results in FIG. 3, the detection accuracy of the SRNet model under the mismatch condition can be improved by 9.9% -10.9% by adopting the method for performing field adaptation on the SRNet model, and the method is effective when three different steganography algorithms are used, so that the method has practical value.

Claims (2)

1. The self-adaptive steganography detection method for the anti-domain adopts an anti-learning strategy, and performs anti-learning with the steganography feature extractor F through the domain discriminator D, so that the features generated by the steganography feature extractor F in the source domain and the target domain are more similar; the sub-domain adaptive classifier Y minimizes sub-domain adaptive loss while classifying to reduce the difference of inter-domain related sub-domain feature distribution; obtaining a steganography detection model MJ for classifying in the target field by simultaneously minimizing classification loss, counterloss and sub-field adaptive loss; the method comprises the following main steps:
step 1, taking a steganography detection model which is trained on the source field as a pre-training model M; firstly, removing a classification layer of the pre-training model M as a steganography feature extractor F; the tagged source domain data and the untagged target domain data are passed through the first layer of the steganographic feature extractor F to obtain features
Figure FDA0004178038770000011
And->
Figure FDA0004178038770000012
Step 2, the characteristics obtained in the step 1 are simultaneously sent into a classification branch and a discrimination branch; in the classification branch, the network structure of the sub-domain adaptive classifier Y is the same as the classification layer of the pre-training model M, the parameters are initialized, and the sub-domain adaptive classifier Y detects the characteristic f according to the source domain steganography sl And target field steganography detection feature f tl And respectively giving predicted values, and simultaneously calculating a classification loss L1 and a sub-field adaptive loss L2:
Figure FDA0004178038770000013
in n s The number of samples in the source domain; the classification loss L1 is the classifier predictive value
Figure FDA0004178038770000014
And Source Domain real Label->
Figure FDA0004178038770000015
Cross entropy loss between;
Figure FDA0004178038770000016
wherein C is the label category, C is {0,1}, and C is the label category number; n is n s For the number of source field samples, n t The number of samples for the target field; w is the weight that the sample belongs to class c;
Figure FDA0004178038770000017
representing the inner product of the feature;
in the discrimination branch, the domain discriminator D is based on the input features
Figure FDA0004178038770000018
Giving a label of the predicted domain, i.e. judging from which domain the feature comes; n is n t For the number of target field samples, the field prediction tag +.>
Figure FDA0004178038770000019
Sum field genuine label d i As the contrast loss L3:
Figure FDA00041780387700000110
step 3, calculating the total loss function
L＝L1+λL2+ωL3
λ and ω are trade-off parameters for sub-domain adaptive and countering losses, respectively; in the training process of the model, the parameters of the steganographic feature extractor F, the sub-domain self-adaptive classifier Y and the domain discriminator D are updated by minimizing the total loss and carrying out back propagation, so that the difference of the feature distribution of the source domain and the target domain is reduced;
and 4, stopping training when the set upper limit of the training times is reached, and combining the steganographic feature extractor F and the sub-field adaptive classifier Y to obtain a steganographic detection model MJ for classifying in the target field.
2. The adaptive steganography detection method for the antigen domain according to claim 1, wherein the network structure of the domain identifier D in the step 2 is not unique and can be adjusted according to practical situations.
CN202210467768.4A 2022-04-29 2022-04-29 Self-adaptive steganography detection method for anti-domain Active CN114782697B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210467768.4A CN114782697B (en) 2022-04-29 2022-04-29 Self-adaptive steganography detection method for anti-domain

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210467768.4A CN114782697B (en) 2022-04-29 2022-04-29 Self-adaptive steganography detection method for anti-domain

Publications (2)

Publication Number Publication Date
CN114782697A CN114782697A (en) 2022-07-22
CN114782697B true CN114782697B (en) 2023-05-23

Family

ID=82435691

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210467768.4A Active CN114782697B (en) 2022-04-29 2022-04-29 Self-adaptive steganography detection method for anti-domain

Country Status (1)

Country Link
CN (1) CN114782697B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112132738A (en) * 2020-10-12 2020-12-25 中国人民武装警察部队工程大学 Image robust steganography method with reference generation
CN112907431A (en) * 2021-02-26 2021-06-04 中国科学技术大学 Steganalysis method for resisting steganography robustness
CN114065861A (en) * 2021-11-17 2022-02-18 北京工业大学 Domain self-adaption method and device based on contrast counterwork learning

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8965038B2 (en) * 2012-02-01 2015-02-24 Sam Houston University Steganalysis with neighboring joint density
CN106920206B (en) * 2017-03-16 2020-04-14 广州大学 Steganalysis method based on antagonistic neural network
US10643602B2 (en) * 2018-03-16 2020-05-05 Microsoft Technology Licensing, Llc Adversarial teacher-student learning for unsupervised domain adaptation
CN111598762B (en) * 2020-04-21 2023-09-19 中山大学 Generating type robust image steganography method
CN111696021B (en) * 2020-06-10 2023-03-28 中国人民武装警察部队工程大学 Image self-adaptive steganalysis system and method based on significance detection
CN112634117B (en) * 2020-12-22 2023-05-05 中山大学 End-to-end JPEG domain image steganography method based on generation of countermeasure network
CN113538202A (en) * 2021-08-05 2021-10-22 齐鲁工业大学 Image steganography method and system based on generative steganography confrontation
CN114169462A (en) * 2021-12-14 2022-03-11 四川大学 Feature-guided depth sub-field self-adaptive steganography detection method
CN114257697B (en) * 2021-12-21 2022-09-23 四川大学 High-capacity universal image information hiding method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112132738A (en) * 2020-10-12 2020-12-25 中国人民武装警察部队工程大学 Image robust steganography method with reference generation
CN112907431A (en) * 2021-02-26 2021-06-04 中国科学技术大学 Steganalysis method for resisting steganography robustness
CN114065861A (en) * 2021-11-17 2022-02-18 北京工业大学 Domain self-adaption method and device based on contrast counterwork learning

Also Published As

Publication number Publication date
CN114782697A (en) 2022-07-22

Similar Documents

Publication Publication Date Title
CN113554089B (en) Image classification countermeasure sample defense method and system and data processing terminal
CN109583342B (en) Human face living body detection method based on transfer learning
Rajagopal Glioma brain tumor detection and segmentation using weighting random forest classifier with optimized ant colony features
Fındık et al. A color image watermarking scheme based on hybrid classification method: particle swarm optimization and k-nearest neighbor algorithm
CN111881722B (en) Cross-age face recognition method, system, device and storage medium
Myronenko et al. Accounting for dependencies in deep learning based multiple instance learning for whole slide imaging
CN113076994A (en) Open-set domain self-adaptive image classification method and system
WO2024051183A1 (en) Backdoor detection method based on decision shortcut search
Quiring et al. Adversarial machine learning against digital watermarking
CN114257697B (en) High-capacity universal image information hiding method
Liu et al. Ensemble of CNN and rich model for steganalysis
CN116912568A (en) Noise-containing label image recognition method based on self-adaptive class equalization
CN111726472A (en) Image anti-interference method based on encryption algorithm
Zhang et al. Advjnd: Generating adversarial examples with just noticeable difference
Hayes et al. Logan: Evaluating information leakage of generative models using generative adversarial networks
Rai et al. An optimized deep fusion convolutional neural network-based digital color image watermarking scheme for copyright protection
CN114782697B (en) Self-adaptive steganography detection method for anti-domain
CN113822377A (en) Fake face detection method based on contrast self-learning
Kaur et al. Feature selection using mutual information and adaptive particle swarm optimization for image steganalysis
CN112907431A (en) Steganalysis method for resisting steganography robustness
CN114169462A (en) Feature-guided depth sub-field self-adaptive steganography detection method
CN116680639A (en) Deep-learning-based anomaly detection method for sensor data of deep-sea submersible
Sultan et al. A new framework for analyzing color models with generative adversarial networks for improved steganography
Desai et al. Survey on universal image steganalysis
CN114638356A (en) Static weight guided deep neural network back door detection method and system

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