CN108961137B - Image steganalysis method and system based on convolutional neural network - Google Patents
Image steganalysis method and system based on convolutional neural network Download PDFInfo
- Publication number
- CN108961137B CN108961137B CN201810432764.6A CN201810432764A CN108961137B CN 108961137 B CN108961137 B CN 108961137B CN 201810432764 A CN201810432764 A CN 201810432764A CN 108961137 B CN108961137 B CN 108961137B
- Authority
- CN
- China
- Prior art keywords
- image
- function
- convolutional neural
- gabor
- neural network
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
-
- 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
- G06F18/29—Graphical models, e.g. Bayesian networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
The invention discloses an image steganalysis method and system based on a convolutional neural network, comprising three modules: the device comprises an image preprocessing part, a feature extraction part and a feature classification module. In the image preprocessing part, a group of Gabor filters with multidirectional and multiscale parameters are selected through experiments, and convolved with an input image to obtain an image residual error with high signal-to-noise ratio; the feature extraction portion uses a shortcut connection structure to directly connect the output of the shallow layer with the following layer to mitigate the overfitting phenomenon. The steganography method based on the convolutional neural network does not need a great deal of domain knowledge about steganography and steganalysis, and the characteristic extraction and characteristic classification processes are combined optimization processes, so that the design is simple and easy to implement; secondly, the scalability and directionality of the Gabor filter can help the network extract more efficient image residuals; finally, the steganography algorithm adaptive to the J-UNIWARD and the UED can obtain better detection effect.
Description
Technical Field
The invention belongs to the field of image steganalysis, and in particular relates to an image steganalysis method and system based on a convolutional neural network.
Background
With the continuous development of internet technology and image processing technology, digital images are one of the most widely used information transmission media in daily life, people can upload and download massive digital images on the internet, but this also gives criminals a opportunity to take advantage of, once criminals steal personal privacy, business confidentiality and national information through digital images on the network by using computer technology, the criminals will have a very bad influence. Therefore, the security reliability of digital images has become a hotspot problem in the field of information security.
The digital image steganography technology is an important component of an information hiding technology, is different from an information encryption technology in encrypting an original file into an unreadable file containing ciphertext but not hiding communication behavior, and is used for embedding information by modifying an original carrier image to finally obtain a carrier image containing the ciphertext which is visually indistinguishable from a common carrier, so that a transmission process of secret information on a public channel is hidden, and an attacker cannot confirm whether the carrier contains the hidden information. According to the difference of the working fields when the steganography algorithm embeds the secret information, the steganography algorithm can be divided into an airspace steganography algorithm for modifying image pixels and a JPEG domain steganography algorithm for modifying discrete cosine transform coefficients. Because of the popularity of the JPEG image transmission application in the network and the openness of the compression coding algorithm thereof, the hidden communication on the network by using the JPEG image as the carrier of the information hiding is more convenient and safer than other formats, and the research on the JPEG image hidden technology becomes the heavy weight of the research in the field of the information hiding.
In order to prevent the steganography technique from being used maliciously by criminals to transfer secret information, the purpose of the steganography analysis technique is to determine whether secret information is embedded in an image. Steganalysis methods based on information embedding changes to statistical features and machine learning theory are typically composed of two parts, feature extraction and training classifier. However, most of the content adaptive steganography algorithms with higher safety at present embed information into areas of complex textures of images by designing a distortion cost function, so that accurate modeling of the areas is difficult to carry out by steganography analysis algorithms based on artificial design features, and very effective features cannot be extracted. In recent years, with deep learning, especially, convolutional neural networks have made breakthrough progress in multiple fields of computer vision, and the field of steganalysis has also begun to use convolutional neural networks to replace traditional steganalysis methods of artificial design features, but the detection results of low embedding rate are not satisfactory by the existing JPEG steganalysis based on convolutional neural networks.
Disclosure of Invention
In order to overcome the defect of the conventional JPEG steganalysis method based on the convolutional neural network, the invention selects a group of Gabor filters with multidirectional and multi-scale parameters through experiments to convolve with an input image to obtain an image residual with high signal to noise ratio, which is also a first method for extracting the residual in a preprocessing part of the JPEG steganalysis network by using the Gabor filters.
In order to solve the technical problems, the technical scheme of the invention is as follows:
an image steganalysis system based on a convolutional neural network comprises an image preprocessing part, a feature extraction part and a feature classification module.
An image steganalysis method based on a convolutional neural network comprises the following steps:
s1: designing an image database required by an experiment, and cutting, compressing and embedding secret information on images in the image database to obtain a secret-loaded image; dividing an original image and a secret-loaded image into a training set, a verification set and a test set which are mutually disjoint;
s2: designing a network model based on a convolutional neural network, wherein the network model comprises a preprocessing layer, a feature extraction layer using a convolutional layer and a pooling layer, a full-connection layer and a feature classification layer of a Softmax function;
s3: determining an initialized parameter of the filter based on image residual data of the preprocessing layer designed according to the experimental result;
s4: the image of the training set is subjected to data enhancement and then is input into a convolutional neural network-based network model of S2 for training;
s5: and (3) selecting the optimal N network models based on the convolutional neural network obtained after training in the step (S4) to analyze the images of the test set, wherein N is a positive integer.
In a preferred embodiment, n=5.
In a preferred embodiment, the image database of S1 is a BOSSbase v1.01 database.
In a preferred embodiment, the secret information embedding of S1 is implemented by adaptive steganography algorithms J-UNIWRD and UED.
In a preferred scheme, the preprocessing layer decompresses an input JPEG image to a space domain to amplify the change of an embedding operation on the image, then uses a multidirectional multi-scale Gabor filter to convolve with the image, suppresses image content irrelevant to classification, increases the signal to noise ratio of the image, and finally cuts off the obtained image residual, thereby further reducing the data range input to the next network. The network structure uses a shortcut connection structure to directly add shallow output and deep output, and the method increases the diversity of the characteristics.
In a preferred embodiment, the truncated linear function is as follows:
in a preferred embodiment, S3 includes the following:
the preprocessing layer adopts a Gabor filter to extract image residual data of the image decompressed to the airspace, and the kernel function of the Gabor filter is obtained by multiplying a Gaussian function and a cosine function and is expressed by the following formula:
wherein, x '=xcos theta+ysin theta and y' = -xsin theta+ycos theta represent pixels of the image, lambda represents cosine function wavelength in the Gabor kernel function, theta represents direction of parallel stripes of the Gabor function,
representing the cosine function phase offset in the Gabor kernel function,/->
Sigma represents the standard deviation of the Gaussian function in the Gabor kernel function, +.>
Representing the cosine function wavelength in the Gabor kernel function, γ being the spatial aspect ratio, representing the ellipticity of the Gabor filter, γ=1; the θ and σ are determined experimentally.
In the preferred scheme, the Gabor filter has good time-frequency characteristics, features can be extracted from multiple scales and directions, and the Gabor filter can extract effective image residual data from an image in a traditional feature-based machine learning method.
In a preferred embodiment, the direction parameter θ is
In a preferred embodiment, σ= {0.75,1}.
In a preferred embodiment, S4 includes the following:
the original image and the secret image are randomly selected to rotate clockwise or anticlockwise by 90 degrees and then are input into a network model based on a convolutional neural network.
The preferred scheme can help the network learn more diversified steganographic features. The Gabor filter parameters of the network model pre-processing layer seek more optimized quantization factors along with the network training, thereby better capturing the trace of steganographic modification.
In a preferred embodiment, S5 includes the following:
and 5 network models with the best performance on the verification set in the S4 are selected to be used for the test set, and the average value of the prediction probabilities of the 5 network models for the two categories output by the test set is used as an analysis result.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
1. the steganography method based on the convolutional neural network does not need a great deal of domain knowledge about steganography and steganalysis, and the characteristic extraction and characteristic classification processes are combined optimization processes, so that the design is simple and easy to implement;
2. the first proposed JPEG steganalysis network that uses a learnable Gabor filter for preprocessing convolution operations, the scalability and directionality of the Gabor filter can be used to help the network extract more efficient image residuals.
3. When the content adaptive steganography algorithm is detected, the detection capability is obviously better than that of the traditional machine learning steganography analysis algorithm based on the characteristics, and the method is not only suitable for detecting one steganography algorithm, but also can be used for detecting different steganography algorithms.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is an overall framework diagram of a steganalysis network in accordance with the present invention.
Fig. 3 is a diagram of a network sub-module in accordance with the present invention.
Fig. 4 is a diagram of a preprocessing layer Gabor filter according to the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the present patent;
for the purpose of better illustrating the embodiments, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the actual product dimensions;
it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
As shown in fig. 1, a JPEG image steganalysis method based on a convolutional neural network includes the following steps:
s1: an image database required for the experiment is generated. The invention is a research in the field of steganography analysis, so that the image library selects a BOSSBase v1.01 database commonly used in the steganography field and the steganography analysis field to measure steganography analysis performance, firstly, an image cutting center 256×256-sized region of a PGM format in the image library is subjected to JPEG compression, finally, the most secret JPEG steganography algorithm J-UNIWRD and the self-adaptive steganography algorithm UED are used for carrying out steganography information embedding on the image of the image library, and an original image and a corresponding carrier image are divided into a training set, a verification set and a test set which are mutually disjoint.
S2, a network framework of a JPEG image steganalysis method based on a convolutional neural network comprises the following steps: a preprocessing layer, a feature extraction layer using a convolution layer and a pooling layer, a fully connected layer, and a feature classification layer of Softmax functions. Decompressing an input JPEG image to a space domain under the condition of no quantization in an image preprocessing layer, extracting an image residual error by 20 Gabor filtering with the size of 8 multiplied by 8 from the decompressed image, and truncating the image residual error by a truncated linear function with the truncation threshold value of 8, wherein the used truncated linear function is as follows:
an overall framework diagram of the steganalysis network is shown in fig. 2, with a network submodule diagram shown in fig. 3. In the network sub-module, each convolution group is followed by a band batch normalization layer (Batch Normalization) and a modified linear unit (ReLU) structure. The convolution layer adopts a shortcut structure, and the characteristic layer jump transmission input before convolution is multiplied by the convolution result to avoid gradient dispersion phenomenon generated by the network along with depth deepening. In the submodule, all other parameters except the number of convolution kernels of each convolution layer are completely consistent, and the number of the convolution kernels of the convolution layer 1 is half of that of each other convolution layer. The number of convolution kernels of the convolution layer 1 in the five sub-modules is 12, 24, 48, 96 and 192 respectively. The final output features of the feature extraction part are subjected to global average pooling, then are input into a full connection layer to fuse the learned depth features, and finally are used as a network objective function to guide the learning process through a Softmax loss function, so that the prediction probability that the input image belongs to two categories of an original carrier image and a carrier image is obtained.
S3: parameters for initializing a preprocessing layer image residual extraction filter are designed, and in view of the excellent performance of a Gabor filter in a traditional steganography analysis method and the fact that the Gabor filter can extract characteristics from multiple dimensions and multiple directions, the preprocessing layer of the network mechanism adopts the Gabor filter to extract residual of an image decompressed to a airspace. The kernel function of the Gabor filter is obtained by multiplying a gaussian function and a cosine function, and the formula is as follows:
wherein x '=xcos θ+ysin θ, y' = -xsin θ+ycos θ represents a pixel of the image, λ represents a cosine function wavelength in the Gabor kernel function, θ represents a direction of parallel stripes of the Gabor function,
representing cosine function phase in Gabor kernel functionOffset, σ represents the standard deviation of the Gaussian function in the Gabor kernel function, ++>
The cosine function wavelength and gamma in the Gabor kernel function are the spatial aspect ratio, and the ellipticity of the Gabor filter is shown.
The Gabor filter has a phase parameter of
Gamma=1, and the parameters are determined to be the direction parameter θ and the scale parameter σ through experiments, and because of symmetry between different directions of the Gabor function, the range of the selected direction parameter is set to be
First we tested +.>
With sampling interval +.>
The performance of the experiment when selecting the direction parameters found that when the direction parameters of the Gabor filter were +.>
The test performance is best. Experiments were then performed on different combinations of σ= {0.5,0.75,1,1.25} and finally the scale parameter was determined as σ= {0.75,1}. An initialization diagram of the 20 filters of the preprocessing layer of the present invention is shown in fig. 4.
S4: the data of the training set is input to train the network model, the iteration times are 90000, and the training model is stored every 500 iterations. The data of the validation set is input into the trained model to select the optimal training model of 5. Inputting the data of the test set into an optimal training model, and taking the average value of the two types of prediction probabilities output by the 5 network models to the test set as a final detection result. The detection performance of the invention on the UED algorithm and the J-UNWARD algorithm is shown in table 1, and corresponding documents of the steganography algorithm are L.Guo, J.Ni, and Y.Q.Shi.an efficient JPEG steganographic scheme using uniform embedding [ C ]. IEEE International Workshop on Information Forensics and Security (WIFS). Spain:IEEE 2012:169-174, and Fridrich J.digital image steganography using universal distortion [ C ]. ACM Workshop on Information Hiding and Multimedia security.ACM,2013:59-68, respectively. SCA-GFR is The most effective JPEG steganographic analysis technique in The traditional method, and The corresponding literature is "Xia C, guan Q, zhao X, et al, improving GFR Steganalysis Features by Using Gabor Symmetry and Weighted Histograms [ C ]. The ACM workshop. ACM,2017:55-66 ]. Under the same conditions, the performance results of the SCA-GFR detection J-UNIWARD steganography algorithm are also shown in Table 1, and compared with the SCA-GFR method, the detection performance of the model provided by the embodiment is best under different embedding capacities, and the performance is greatly improved under the condition of high embedding capacity. The invention can also improve the detection performance more effectively for the case of low embedding capacity, which is a small modification of the original carrier image.
Table 1 example vs. conventional algorithm
The terms describing the positional relationship in the drawings are merely illustrative, and are not to be construed as limiting the present patent; it is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.
Claims (9)
1. The image steganalysis method based on the convolutional neural network is characterized by comprising the following steps of:
s1: designing an image database required by an experiment, and cutting, compressing and embedding secret information on images in the image database to obtain a secret-loaded image; dividing an original image and a secret-loaded image into a training set, a verification set and a test set which are mutually disjoint;
s2: designing a network model based on a convolutional neural network, wherein the network model comprises a preprocessing layer, a feature extraction layer using a convolutional layer and a pooling layer, a full-connection layer and a feature classification layer of a Softmax function;
the preprocessing layer decompresses an input image to an airspace, then convolves the decompressed image with a multidirectional and multiscale Gabor filter, and finally truncates the obtained image residual data through a truncating function; the truncation function is as follows:
the f (x) represents a truncation function, x represents image residual data, and T represents a truncation threshold;
s3: designing initialization parameters of a preprocessing layer filter according to experimental results;
s4: the image of the training set is subjected to data enhancement and then is input into a convolutional neural network-based network model of S2 for training;
s5: and (3) selecting the optimal N network models based on the convolutional neural network obtained after training in the step (S4) to analyze the images of the test set, wherein N is a positive integer.
2. The method of claim 1, wherein S3 comprises the following steps:
the preprocessing layer adopts a Gabor filter to extract image residual data of the image decompressed to the airspace, and the kernel function of the Gabor filter is obtained by multiplying a Gaussian function and a cosine function and is expressed by the following formula:
wherein said x ′ =xcos θ+ysinθ and y ′ = -xsin theta + ycos theta represents the pixels of the image, lambda represents the cosine function wavelength in the Gabor kernel, theta represents the direction of parallel stripes of the Gabor function, phi represents the cosine function phase offset in the Gabor kernel,
sigma represents the standard deviation of the Gaussian function in the Gabor kernel function, +.>
Representing the cosine function wavelength in the Gabor kernel function, γ being the spatial aspect ratio, representing the ellipticity of the Gabor filter, γ=1; the θ and σ are determined experimentally.
3. The method of claim 2, wherein the direction parameter θ is
4. A method of steganalysis according to claim 2 or 3, wherein σ= {0.75,1}.
5. The method of claim 4, wherein S4 comprises the following steps:
the original image and the secret image are randomly selected to rotate clockwise or anticlockwise by 90 degrees and then are input into a network model based on a convolutional neural network.
6. The method for steganalysis according to claim 5, wherein S5 comprises the following steps:
and 5 network models with the best performance on the verification set in the S4 are selected to be used for the test set, and the average value of the prediction probabilities of the 5 network models for the two categories output by the test set is used as an analysis result.
7. The method of claim 6, wherein the image database of S1 is a BOSSbase v1.01 database.
8. The method for steganographic analysis of the image of claim 7, wherein the secret information embedding of S1 is implemented by steganographic algorithm J-UNIWRD and adaptive steganographic algorithm UED.
9. The convolutional neural network-based image steganalysis system is characterized by comprising an image preprocessing part, a feature extraction part and a feature classification module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810432764.6A CN108961137B (en) | 2018-07-12 | 2018-07-12 | Image steganalysis method and system based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810432764.6A CN108961137B (en) | 2018-07-12 | 2018-07-12 | Image steganalysis method and system based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108961137A CN108961137A (en) | 2018-12-07 |
| CN108961137B true CN108961137B (en) | 2023-05-23 |
Family
ID=64499101
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810432764.6A Active CN108961137B (en) | 2018-07-12 | 2018-07-12 | Image steganalysis method and system based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108961137B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109859091B (en) * | 2018-12-24 | 2023-05-16 | 中国人民解放军国防科技大学 | Image steganography detection method based on Gabor filtering and convolutional neural network |
| CN109934761B (en) * | 2019-01-31 | 2022-11-29 | 中山大学 | JPEG image steganalysis method based on convolutional neural network |
| CN109995520A (en) * | 2019-03-06 | 2019-07-09 | 西北大学 | Cipher key transmission methods, image processing platform based on depth convolutional neural networks |
| CN109949200B (en) * | 2019-03-11 | 2022-10-11 | 河南工业大学 | Filter subset selection and CNN-based steganalysis framework construction method |
| CN110033083B (en) * | 2019-03-29 | 2023-08-29 | 腾讯科技（深圳）有限公司 | Convolutional neural network model compression method and device, storage medium and electronic device |
| CN110334805B (en) * | 2019-05-05 | 2022-10-25 | 中山大学 | JPEG domain image steganography method and system based on generation countermeasure network |
| CN112149834B (en) * | 2019-06-28 | 2023-11-07 | 北京百度网讯科技有限公司 | Model training method, device, equipment and medium |
| CN110503157B (en) * | 2019-08-27 | 2021-11-19 | 安徽大学 | Image steganalysis method of multitask convolution neural network based on fine-grained image |
| CN112561766B (en) * | 2019-09-10 | 2024-03-05 | 阿里巴巴集团控股有限公司 | Image steganography and extraction method and device and electronic equipment |
| CN112487365B (en) * | 2019-09-11 | 2024-01-30 | 阿里巴巴集团控股有限公司 | Information steganography method and information detection method and device |
| CN111199507B (en) * | 2019-12-25 | 2023-05-05 | 深圳大学 | Image steganalysis method, intelligent terminal and storage medium |
| CN111050174A (en) * | 2019-12-27 | 2020-04-21 | 清华大学 | Image compression method, device and system |
| CN112529758B (en) * | 2020-12-18 | 2022-08-09 | 海南大学 | Color image steganography method based on convolutional neural network |
| CN112734622B (en) * | 2021-03-30 | 2021-07-20 | 深圳大学 | Image steganalysis method and terminal based on Tucker decomposition |
| CN113076549B (en) * | 2021-04-08 | 2023-05-12 | 上海电力大学 | Novel U-Net structure generator-based countermeasures network image steganography method |
| CN112926607B (en) * | 2021-04-28 | 2023-02-17 | 河南大学 | Two-branch network image steganography framework and method based on convolutional neural network |
| CN113709323A (en) * | 2021-08-06 | 2021-11-26 | 合肥高维数据技术有限公司 | JPEG image processing method, robust steganography method, system and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778742B (en) * | 2016-12-09 | 2020-03-31 | 东南大学 | Car logo detection method based on Gabor filter background texture suppression |
| CN107330845A (en) * | 2017-06-26 | 2017-11-07 | 中山大学 | A kind of digital picture steganalysis method based on convolutional neural networks |
-
2018
- 2018-07-12 CN CN201810432764.6A patent/CN108961137B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN108961137A (en) | 2018-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961137B (en) | Image steganalysis method and system based on convolutional neural network | |
| Ahmaderaghi et al. | Blind image watermark detection algorithm based on discrete shearlet transform using statistical decision theory | |
| Kang et al. | Identifying tampered regions using singular value decomposition in digital image forensics | |
| Makbol et al. | Block‐based discrete wavelet transform‐singular value decomposition image watermarking scheme using human visual system characteristics | |
| Gul et al. | A new methodology in steganalysis: breaking highly undetectable steganograpy (HUGO) | |
| Monga et al. | Perceptual image hashing via feature points: performance evaluation and tradeoffs | |
| Meikap et al. | Directional PVO for reversible data hiding scheme with image interpolation | |
| CN112529758B (en) | Color image steganography method based on convolutional neural network | |
| Fridrich et al. | Steganalysis of LSB replacement using parity-aware features | |
| Wu et al. | Steganalysis via deep residual network | |
| CN110968845B (en) | Detection method for LSB steganography based on convolutional neural network generation | |
| Qureshi et al. | Bibliography of digital image anti‐forensics and anti‐anti‐forensics techniques | |
| Muralidharan et al. | The infinite race between steganography and steganalysis in images | |
| Kang et al. | Robust median filtering forensics based on the autoregressive model of median filtered residual | |
| Subhedar et al. | Image steganography using contourlet transform and matrix decomposition techniques | |
| CN109523452B (en) | Color image steganography detection method based on channel differential residual error | |
| CN115809953A (en) | Attention mechanism-based multi-size image robust watermarking method and system | |
| CN115345768A (en) | Robust watermark attack method and system based on neural network | |
| CN115908095A (en) | Hierarchical attention feature fusion-based robust image watermarking method and system | |
| Jana et al. | A new DCT based robust image watermarking scheme using cellular automata | |
| Chhikara et al. | A review on digital image steganalysis techniques categorised by features extracted | |
| Tang et al. | Detection of various image operations based on CNN | |
| Bashir et al. | Towards deep learning-based image steganalysis: Practices and open research issues | |
| CN116308986A (en) | Hidden watermark attack algorithm based on wavelet transformation and attention mechanism | |
| Wei et al. | Auto-generating neural networks with reinforcement learning for multi-purpose image forensics |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Li Xuan Inventor after: Kang Xiangui Inventor before: Li Xuan |
|
| CB03 | Change of inventor or designer information |