CN108805161A - A kind of Stego-detection method of multi-embedding rate stego-image - Google Patents
A kind of Stego-detection method of multi-embedding rate stego-image Download PDFInfo
- Publication number
- CN108805161A CN108805161A CN201810375675.2A CN201810375675A CN108805161A CN 108805161 A CN108805161 A CN 108805161A CN 201810375675 A CN201810375675 A CN 201810375675A CN 108805161 A CN108805161 A CN 108805161A
- Authority
- CN
- China
- Prior art keywords
- image
- efld
- sample
- training
- fld
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Editing Of Facsimile Originals (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of Stego-detection methods of multi-embedding rate stego-image.This method is to train to obtain two integrated FLD graders by different embedded rate images.First integrated FLD grader is trained to obtain by low embedded rate image and carrier image, it can select the more preferable training sample for representing carrier image and hidden image feature, and the sample selected and Qi Gao insertion rates version are trained together and obtain second integrated FLD grader.It is had main steps that unknown classification image classification:First integrated FLD grader classifies to the image of votes in a certain range, then is classified to the residual image of votes in a certain range by second integrated FLD grader, and last unfiled image is classified by first integrated FLD grader.The method of the present invention is utilized respectively the advantage of the two integrated FLD graders, it is thus possible to promote the recognition accuracy to JPEG hidden images.
Description
Technical field
The present invention relates to a kind of Stego-detection methods of multi-embedding rate stego-image, and multiple insertions can be obtained when for training
The characteristics of hidden image of rate, promotes the recognition accuracy of low embedded rate image by high embedded rate image.
Background technology
Steganalysis is the important component in multi-media information security, and main task is detected in digital carrier
Whether secret information is had.The most-often used digital picture in numerous digital carriers, because of the number of 512 × 512 sizes
Word image can carry enough secret informations.Digital picture can be divided into many classes, JPEG due to the difference of its format
(Joint Picture Expert Group Joint Photographic Experts Groups) image because its compression and image reconstruction effect and
It is largely used in life.Many steganography methods are designed based on jpeg image, will contain the hidden of secret information
It writes image and is mixed in internet and largely transmitted into row information in normal jpeg image, thus there is very high concealment.It is excellent at present
The characteristics of elegant JPEG image steganography method is:The constant and sufficient information embedded quantity of statistical property.If when embedding information
Embedded rate in 0.05bpac, (per non-zero discrete cosine, hand over bits per nonzero AC DCT coefficient bits by transformation
Flow coefficient), when the low embedded rate such as 0.1bpac, the recognition accuracy of the steganalysis JPEG image steganography method outstanding to these
It is not high.One research contents of steganalysis is by catching difference and steganography calculation between hidden image and normal digital image
The principle of method extracts the steganalysis feature that can more preferably distinguish hidden image and normal digital image, another research contents
It is the grader that design meets application scenarios.
Invention content
It is insufficient existing for prior art the purpose of the present invention is being directed to, a kind of steganography inspection of multi-embedding rate stego-image is provided
Survey method can obtain the hidden image feature set of multiple and different embedded rates, design for steganalysis in training grader
New integrated FLD (taking house linear discriminant) grader, the identification of low embedded rate image is promoted using the embedded rate hidden image of height
Accuracy rate.
In order to achieve the above objectives, the present invention adopts the following technical scheme that:
1. a kind of Stego-detection method of multi-embedding rate stego-image, it is characterised in that operating procedure is as follows:
(1) low embedded rate image and carrier image train to obtain EFLD1(integrated to take house linear discrimination classification device 1):By carrier figure
Picture and low embedded rate image train to obtain an integrated FLD graders EFLD1, and pass through EFLD1Select the high training of number of votes obtained
Image;
(2) low embedded rate image, the embedded rate image of height and carrier image train to obtain EFLD2It is (integrated to take house linear discrimination classification
Device 2):The training image and its high embedded rate version selected by previous step train to obtain second integrated FLD grader
EFLD2;
(3) cross validation determines parameter value:The ginseng used when finally being classified to unfiled image by cross validation determination
Number size, that is, determine number of votes obtained range;
(4) classify to unfiled image:EFLD is used first1Classify to the test image within the scope of number of votes obtained, remains
Under test image use EFLD2Classify to the image within the scope of number of votes obtained, also non-classified test image is by EFLD1Point
Class
The step(1)Low embedded rate image and carrier image train to obtain EFLD1:
It is sample number to have the steganalysis data set of 3 n × p, n first, and p is the dimension of sample, and N is integrated classifier subclassification
Device quantity.First data set is carrier image steganalysis feature set, and second data set is that low embedded rate stego-image is special
Collection, third data set are high embedded rate stego-image feature sets.It generates one group of length and is equal to 0.5n, maximum value is equal to n's
Random Positive Integer Set does not repeat mutually between digital, is that serial number extracts trained sample to these three data sets with these numbers
This, it is remaining to give over to test sample;The shared 0.5n width carrier images after handling in this way, 0.5n low embedded rate image, and
0.5n panel height insertion rate images participate in training;It also needs to leave part training figure before first integrated FLD grader of training
As being used as cross validation collection, remaining training set that can train to obtain first integrated FLD grader;
In first integrated FLD grader of training, stochastic subspace first is extracted to sample characteristics, i.e., from the original p of sample
Certain dimension is randomly selected in dimensional feature;After the stochastic subspace for extracting carrier image and low embedded rate image pattern, it can calculate
Obtain the best projection direction of the FLD sub-classifiers;It can be optimized according to formula (1) and best projection direction is calculated,
Following alternatives can be used:[1] Scatter Matrix in class is calculated:The equal of each characteristic value is calculated separately to this two classes sample
Value, and obtain the equal value difference of this two category feature;Each category feature subtract unique characteristics value mean value and and the result of calculation turn
Matrix multiple is set, then two category feature result of calculations are added to obtain Scatter Matrix in class.[2] the equal value difference matrix of two classes is calculated:It should
Matrix is actually the equal value difference matrix of two classes being calculated in [1].[3] increase L2 and optimize item:Scatter Matrix in class is added
Upper one identical with its matrix size and be multiplied by 10-10Unit matrix.[4] best projection direction is calculated:It is excellent to increase L2
Change Scatter Matrix divided by the equal value difference matrix of two classes in the class of item and just obtains approximate best projection direction。
(1)
(2)
(3)
WhereinIt is Scatter Matrix in class,It is class scatter matrix,Indicate theThe mean value of class,Indicate theClass
InA image feature vector,Indicate the mean value of all training datas.Obtain best projection directionAfterwards, decision gate is calculated
Limit b;Circular is to be multiplied by training sampleProjection result is obtained, it is the minimum value of projection result to take b, then not
The size of disconnected increase b, when classification accuracy maximum, b at this time is exactly decision threshold.
An integrated FLD graders EFLD for sharing N number of sub-classifier is obtained after repeating n times1;Use EFLD1To training sample
This is classified, and the voting results of all training images can be obtained;It is taken out the training image of number of votes obtained in a certain range
High embedded rate image corresponding with these training images;These images taken out illustrate to have more since obtained poll is higher
The judgement of sub-classifier be identical, then these images can preferably embody the spy of hidden image or carrier image
Point.
The step(2)Low embedded rate image, the embedded rate image of height and carrier image train to obtain EFLD2:
After the version for having the carrier image and low embedded hidden image and its high embedded rate that number of votes obtained is high, one new collection of training
At FLD graders EFLD2, but being slightly different in training process and the first step;Because selecting the quantity of image in general
It is fewer than the training image quantity in the first step, uses bootstrap (to repeat first when so training sub-classifier every time
Sampling) method carries out resampling to these a small amount of samples and generates new training set;Usually have one after each bootstrap
The sample of fixed number amount is not drawn into, these samples are used for OOB (bag bags of outer error rates of out of) tests to determine ginseng
Several values;In training, a certain number of high embedded rate images slowly can be continuously added or remove, and select in this process
Select out a best sub-classifier of OOB effects;Integrated FLD points that one shares N number of sub-classifier are obtained after same cycle n times
Class device EFLD2。
The step(3)Cross validation determines parameter value:
Training obtains two integrated FLD graders EFLD1And EFLD2Afterwards, it is also necessary to the ginseng used when one test of extra computation
Number, the discrimination threshold of integrated classifier when the effect of this parameter is to determine test;Specific method is to utilize not join in step (1)
Cross validation is carried out with the training sample of training grader, takes the parameter value for so that cross validation results are best.
The step(4)Classify to unfiled image:
EFLD is used first1Test sample is judged, EFLD1The only test sample of classification votes in a certain range;
Then using the EFLD for thering is high embedded rate sample to participate in training2Classify to remaining test sample, equally only classification ballot
The test sample of number in a certain range, directly skips the test sample being classified;By having one after this two step
A little test samples are also unfiled, these test samples are by EFLD1Classify;This is because low embedded rate image and carrier image
Optimal judgement plane and decision threshold be in EFLD1Sub-classifier in obtain, thus remaining unfiled test sample
In EFLD1Under have highest recognition accuracy;
Final sample predictions result is made of three parts, EFLD1Differentiation to high poll sample, EFLD2To high poll sample
Differentiation and EFLD1Differentiation to all remaining samples, if in Part III there are voting results be 0 sample,
EFLD1It will be at random to sample classification, it usually needs the quantity of the sample judged at random is considerably less;After sample is classified, nothing
By to mistake, grader later will not again classify to this sample, thus this three parts is not in the same sample
Repeat classification situation;
The new integrated FLD graders for having the characteristics of multiple embedded rates for hidden image in this way and designing, can promote steganography point
Analyse accuracy rate.
The present invention compared with prior art, have following obvious prominent substantive distinguishing features and significant technology into
Step:Existing Steganalysis has to be hoisted for the recognition accuracy of low embedded rate image and there is no hidden in view of carrying out
High embedded rate image can be got when writing analysis, this method takes into account the embedded rate image of height, utilizes the embedded rate image of height
Promote the recognition accuracy to low embedded rate image.
Description of the drawings
Fig. 1 is the flowsheet of the method for the present invention.
Fig. 2 is the flow chart of the integrated FLD graders of training first of the method for the present invention.
Fig. 3 be the method for the present invention second integrated FLD grader in a sub- classifier training flow chart.
Fig. 4 is the classification process figure of the method for the present invention.
Specific implementation mode
Below in conjunction with the accompanying drawings, it elaborates to the specific embodiment of the invention.
Embodiment one:
Referring to Fig. 1, the Stego-detection method of this multi-embedding rate stego-image, it is characterised in that operating procedure is as follows:
(1) low embedded rate image and carrier image train to obtain EFLD1:It trains to obtain one by carrier image and low embedded rate image
A integrated FLD graders EFLD1, and pass through EFLD1Select the high training image of number of votes obtained;
(2) low embedded rate image, the embedded rate image of height and carrier image train to obtain EFLD2:The instruction selected by previous step
Practice image and its high embedded rate version trains to obtain second integrated FLD graders EFLD2;
(3) cross validation determines parameter value:The ginseng used when finally being classified to unfiled image by cross validation determination
Number size, that is, determine number of votes obtained range;
(4) classify to unfiled image:EFLD is used first1Classify to the test image within the scope of number of votes obtained, remains
Under test image use EFLD2Classify to the image within the scope of number of votes obtained, also non-classified test image is by EFLD1Point
Class.
Embodiment two:The present embodiment and embodiment one are essentially identical, and special feature is:
Referring to Fig. 2, the step(1)Low embedded rate image and carrier image train to obtain EFLD1:
An integrated FLD grader for sharing 50 sub-classifiers is trained using 90% training sample first, sub-classifier is to carry
The FLD graders of L2 optimizations.
Training EFLD1Specific steps it is as shown in Figure 2.There are 3 10000 × 8000 steganalysis data sets, 10000 are
Sample number, 8000 be the dimension of sample, and 50 be integrated classifier sub-classifier quantity.First data set is carrier image steganography
Feature set is analyzed, second data set is low embedded rate stego-image feature set, and third data set is that high embedded rate contains close figure
As feature set.Generate one group of length be equal to 5000, maximum value be equal to 10000 random Positive Integer Set, number between mutually not
It repeats, is that serial number extracts training sample to these three data sets with these numbers, it is remaining to give over to test sample, by this way
5000 width carrier images are shared after processing, 5000 low embedded rate images and 5000 panel height insertion rate images participate in training;It is instructing
Also need to leave 10% training set before practicing first integrated FLD grader as cross validation collection, remaining 90% training set can instruct
It gets to first integrated FLD grader;
In first integrated FLD grader of training, random 2000 n-dimensional subspace n first is extracted to sample characteristics, i.e., from sample
2000 dimensional features are randomly selected in original 8000 dimensional feature;Extract the stochastic subspace of carrier image and low embedded rate image pattern
Afterwards, the best projection direction of the FLD sub-classifiers can be calculated;One is obtained after repeating 50 times shares 50 sub-classifiers
Integrated FLD graders EFLD1;Use EFLD1Classify to training sample, the voting results of all training images can be obtained;From
Middle taking-up number of votes obtained is more than 10 or the high embedded rate image corresponding with these training images of the training image less than -10.
Embodiment three:The present embodiment and embodiment one are essentially identical, and special feature is:
Referring to Fig. 3, the step(2)Low embedded rate image, the embedded rate image of height and carrier image train to obtain EFLD2:
After obtaining the high carrier image of number of votes obtained and low embedded hidden image and its version of high embedded rate, one new collection of training
At FLD graders, but training process and previous it is slightly different;The sample for first using bootstrap methods a small amount of to these
It carries out resampling and generates new training set, in each sub-classifier of training, can be added on the basis of height wins the vote training image
Its height is embedded in rate version to train to obtain a better sub-classifier of classifying quality;Specific steps are as shown in figure 3, to picking out
Training sample and its high embedded rate version choose stochastic subspace, pick out training sample using 90% and add extra random
100 panel height insertion rate images carry out the training of FLD sub-classifiers, record this sub-classifier and are imitated to the OOB for being left 10% sample
Fruit;It then adds random 100 panel height insertion rate image and equally carries out FLD sub-classifier training, most with current OOB effects by it
Good FLD sub-classifiers are compared, and continue to add high embedded rate figure if OOB effects are better than current optimal sub-classifier
Otherwise picture removes the 100 panel height insertion rate images currently added and chooses the next instruction of 100 panel height insertion rate images progress again
Practice;Deconditioning after reaching preset total frequency of training takes out the best sub-classifier of OOB effects;Same cycle 50
An integrated FLD graders EFLD for sharing 50 sub-classifiers is obtained after secondary2。
Example IV:The present embodiment and embodiment one are essentially identical, and special feature is:
The step(3)Cross validation determines parameter value:
Training obtains two integrated FLD graders EFLD1And EFLD2Afterwards, it is also necessary to the ginseng used when one test of extra computation
Number, the discrimination threshold of integrated classifier when the effect of this parameter is to determine test;Specific method is to utilize step(1)In do not join
Cross validation is carried out with 10% training sample of training grader, takes the number of votes obtained range for so that cross validation results are best, range
Between [0.5,1].
Embodiment five:The present embodiment and embodiment one are essentially identical, and special feature is:
Referring to Fig. 4, the step(4)Classify to unfiled image:
It is as shown in Figure 4 to the classifying step of unknown images.EFLD is used first1Test sample is judged, EFLD1Only classify
The test sample of votes in a certain range;Then using the EFLD for thering is high embedded rate sample to participate in training2To remaining survey
Sample is originally classified, the test sample of votes in a certain range of equally only classifying, to the test sample being classified
Directly skip;By being had after this two step, some test samples are also unfiled, these test samples are by EFLD1Classify;
Final sample predictions result is made of three parts, EFLD1Differentiation to high poll sample, EFLD2To high poll sample
Differentiation and EFLD1Differentiation to all remaining samples, if in Part III there are voting results be 0 sample,
EFLD1It will be at random to sample classification, it usually needs the quantity of the sample judged at random is considerably less;After sample is classified, nothing
By to mistake, grader later will not again classify to this sample, thus this three parts is not in the same sample
Repeat classification situation;
The new integrated FLD graders for having the characteristics of multiple embedded rates for hidden image in this way and designing, can promote steganography point
Analyse accuracy rate.
Claims (5)
1. a kind of Stego-detection method of multi-embedding rate stego-image, it is characterised in that operating procedure is as follows:
(1) low embedded rate image and carrier image train to obtain EFLD1:It trains to obtain one by carrier image and low embedded rate image
A integrated FLD graders EFLD1, and pass through EFLD1Select the high training image of number of votes obtained;
(2) low embedded rate image, the embedded rate image of height and carrier image train to obtain EFLD2:The training selected by previous step
Image and its high embedded rate version train to obtain second integrated FLD graders EFLD2;
(3) cross validation determines parameter value:The ginseng used when finally being classified to unfiled image by cross validation determination
Number size, that is, determine number of votes obtained range;
(4) classify to unfiled image:EFLD is used first1Classify to the test image within the scope of number of votes obtained, is left
Test image use EFLD2Classify to the image within the scope of number of votes obtained, also non-classified test image is by EFLD1Classification.
2. a kind of Stego-detection method of multi-embedding rate stego-image according to claim 1, it is characterised in that:The step
Suddenly(1)Low embedded rate image and carrier image train to obtain EFLD1:
It is sample number to have the steganalysis data set of 3 n × p, n, and p is the dimension of sample, and N is integrated classifier sub-classifier number
Amount, first data set is carrier image steganalysis feature set, and second data set is low embedded rate stego-image feature set,
Third data set is high embedded rate stego-image feature set, generates one group of length and is equal to 0.5n, maximum value equal to n it is random just
Integer set does not repeat mutually between digital, is that serial number extracts training sample to these three data sets with these numbers, is left
Give over to test sample;0.5n width carrier images, 0.5n low embedded rate image and 0.5n panel heights are shared after handling in this way
Embedded rate image participates in training;Also need to leave part training image before first integrated FLD grader of training as friendship
Fork verification collection, remaining training set can train to obtain first integrated FLD grader;
In first integrated FLD grader of training, stochastic subspace first is extracted to sample characteristics, i.e., from the original p of sample
Certain dimension is randomly selected in dimensional feature;After the stochastic subspace for extracting carrier image and low embedded rate image pattern, it can calculate
Obtain the best projection direction of the FLD sub-classifiers;The integrated FLD for sharing a N number of sub-classifier classification is obtained after repeating n times
Device EFLD1;Use EFLD1Classify to training sample, the voting results of all training images can be obtained;It is taken out gained vote
The training image of number in a certain range high embedded rate image corresponding with these training images.
3. a kind of Stego-detection method of multi-embedding rate stego-image according to claim 1, it is characterised in that:The step
Suddenly(2)Low embedded rate image, the embedded rate image of height and carrier image train to obtain EFLD2:
After obtaining the high carrier image of number of votes obtained and low embedded hidden image and its version of high embedded rate, one new collection of training
At FLD graders, but training process and previous it is slightly different;The sample for first using bootstrap methods a small amount of to these
It carries out resampling and generates new training set, in each sub-classifier of training, can be added on the basis of height wins the vote training image
Its height is embedded in rate version to train to obtain a better sub-classifier of classifying quality;In training, can slowly be continuously added
Or a certain number of high embedded rate images of removal, and a best sub-classifier of OOB effects is selected in this process;Together
An integrated FLD graders EFLD for sharing N number of sub-classifier is obtained after sample cycle n times2。
4. a kind of Stego-detection method of multi-embedding rate stego-image according to claim 1, it is characterised in that:The step
Suddenly(3)Cross validation determines parameter value:
Training obtains two integrated FLD graders EFLD1And EFLD2Afterwards, it is also necessary to the ginseng used when one test of extra computation
Number, the discrimination threshold of integrated classifier when the effect of this parameter is to determine test;Specific method is to utilize not join in step (1)
Cross validation test is carried out with the training sample of training grader, takes the parameter value for so that cross validation results are best.
5. a kind of Stego-detection method of multi-embedding rate stego-image according to claim 1, it is characterised in that:The step
Suddenly(4)Classify to unfiled image:
EFLD is used first1Test sample is judged, EFLD1The only test sample of classification votes in a certain range;So
Afterwards using the EFLD for thering is high embedded rate sample to participate in training2Classify to remaining test sample, votes of equally only classifying
Test sample in a certain range directly skips the test sample being classified;By being had after this two step
Test sample is also unfiled, these test samples are by EFLD1Classify;
Final sample predictions result is made of three parts, EFLD1Differentiation to high poll sample, EFLD2To high poll sample
Differentiation and EFLD1Differentiation to all remaining samples, if there are the sample that voting results are 0, EFLD in Part III1
It will be at random to sample classification, it usually needs the quantity of the sample judged at random is considerably less;It is no matter right after sample is classified
Mistake, grader later will not again classify to this sample, thus this three parts is not in be repeated to the same sample
Classification situation;
The new integrated FLD graders for having the characteristics of multiple embedded rates for hidden image in this way and designing, can promote steganography point
Analyse accuracy rate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810375675.2A CN108805161B (en) | 2018-04-25 | 2018-04-25 | Steganography detection method for multi-embedding-rate encrypted image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810375675.2A CN108805161B (en) | 2018-04-25 | 2018-04-25 | Steganography detection method for multi-embedding-rate encrypted image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108805161A true CN108805161A (en) | 2018-11-13 |
CN108805161B CN108805161B (en) | 2022-08-05 |
Family
ID=64093838
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810375675.2A Active CN108805161B (en) | 2018-04-25 | 2018-04-25 | Steganography detection method for multi-embedding-rate encrypted image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108805161B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110059754A (en) * | 2019-04-22 | 2019-07-26 | 厦门大学 | A kind of batch data steganography method, terminal device and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103258123A (en) * | 2013-04-25 | 2013-08-21 | 中国科学院信息工程研究所 | Steganalysis method based on blindness of steganalysis systems |
CN106203492A (en) * | 2016-06-30 | 2016-12-07 | 中国科学院计算技术研究所 | The system and method that a kind of image latent writing is analyzed |
CN106504177A (en) * | 2016-09-30 | 2017-03-15 | 广东技术师范学院 | A kind of low embedded rate compressed sensing general steganalysis method of coloured image |
CN107273916A (en) * | 2017-05-22 | 2017-10-20 | 上海大学 | The unknown Information Hiding & Detecting method of steganographic algorithm |
-
2018
- 2018-04-25 CN CN201810375675.2A patent/CN108805161B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103258123A (en) * | 2013-04-25 | 2013-08-21 | 中国科学院信息工程研究所 | Steganalysis method based on blindness of steganalysis systems |
CN106203492A (en) * | 2016-06-30 | 2016-12-07 | 中国科学院计算技术研究所 | The system and method that a kind of image latent writing is analyzed |
CN106504177A (en) * | 2016-09-30 | 2017-03-15 | 广东技术师范学院 | A kind of low embedded rate compressed sensing general steganalysis method of coloured image |
CN107273916A (en) * | 2017-05-22 | 2017-10-20 | 上海大学 | The unknown Information Hiding & Detecting method of steganographic algorithm |
Non-Patent Citations (8)
Title |
---|
BIN CHEN ET AL.: ""Mixing high-dimensional features for JPEG steganalysis with ensemble classifer"", 《SIVIP》 * |
FENGYONG LI ET AL.: ""JPEG Steganalysis With High-Dimensional Features and Bayesian Ensemble Classifier"", 《IEEE SIGNAL PROCESSING LETTERS》 * |
XIKAI XU ET AL.: ""An Adaptive Ensemble Classifier for Steganalysis Based on Dynamic Weighted Fusion"", 《PROCEEDING OF THE 2015 INTERNATIONAL CONFERENCE ON COMMUNICATIONS,SIGNAL PROCESSING,AND SYSTEMS》 * |
万宝吉 等: ""基于多分类器融合的位置嵌入率图像隐写分析方法"", 《数据采集与处理》 * |
孙物一 等: ""基于高维特征正则化的JPEG图像隐写分析"", 《应用科学学报》 * |
曹镇 等: ""基于层次修剪和加权投票的图像隐写分析"", 《计算机工程与设计》 * |
朱婷婷 等: ""基于不确定性推理的JPEG图像通用隐藏信息检测技术"", 《电子学报》 * |
钱银龙: ""深度学习的图像隐写分析方法研究"", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110059754A (en) * | 2019-04-22 | 2019-07-26 | 厦门大学 | A kind of batch data steganography method, terminal device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108805161B (en) | 2022-08-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107077734A (en) | Determining method and program | |
CN103942562A (en) | Hyperspectral image classifying method based on multi-classifier combining | |
CN105654066A (en) | Vehicle identification method and device | |
CN106203492A (en) | The system and method that a kind of image latent writing is analyzed | |
CN108280480B (en) | Latent image carrier security evaluation method based on residual error co-occurrence probability | |
CN110874604A (en) | Model training method and terminal equipment | |
CN105989001B (en) | Image search method and device, image search system | |
CN112241494A (en) | Key information pushing method and device based on user behavior data | |
CN109543674A (en) | A kind of image copy detection method based on generation confrontation network | |
CN113593714A (en) | Method, system, equipment and medium for detecting multi-classification new coronary pneumonia cases | |
CN107273916A (en) | The unknown Information Hiding & Detecting method of steganographic algorithm | |
CN103310235A (en) | Steganalysis method based on parameter identification and estimation | |
CN109523452A (en) | A kind of color image Stego-detection method based on channel difference residual | |
CN108268889A (en) | To carved gravure true-false detection method, detection platform and detecting system | |
CN111488939A (en) | Model training method, classification method, device and equipment | |
CN108229505A (en) | Image classification method based on FISHER multistage dictionary learnings | |
TWI483127B (en) | Adaptable categorization method and computer readable recording medium using the adaptable categorization method | |
CN108805161A (en) | A kind of Stego-detection method of multi-embedding rate stego-image | |
CN108510483B (en) | Method for generating color image tampering detection by adopting VLAD coding and SVM calculation | |
CN111461135B (en) | Digital image local filtering evidence obtaining method integrated by convolutional neural network | |
CN107729918A (en) | Cellular automata based on Cost Sensitive Support Vector Machines emerges in large numbers the sorting technique of phenomenon | |
CN102592151B (en) | Blind detection method for median filter in digital image | |
CN106530199B (en) | Multimedia integration steganalysis method based on window type hypothesis testing | |
CN105138984A (en) | Sharpened image identification method based on multi-resolution overshoot effect measurement | |
Zhao et al. | Pornographic image recognition in compressed domain based on multi-cost sensitive decision tree |
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 |