CN1710594A - Picture concealed blind-detection system based high-order statistic - Google Patents
Picture concealed blind-detection system based high-order statistic Download PDFInfo
- Publication number
- CN1710594A CN1710594A CN 200510026803 CN200510026803A CN1710594A CN 1710594 A CN1710594 A CN 1710594A CN 200510026803 CN200510026803 CN 200510026803 CN 200510026803 A CN200510026803 A CN 200510026803A CN 1710594 A CN1710594 A CN 1710594A
- Authority
- CN
- China
- Prior art keywords
- picture
- analyzer
- training
- controller
- pretreater
- 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.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The system includes preprocessor, trainer, controller and analyzer. Controller controls and calls preprocessor, trainer, and analyzer. If user's input information is to analyze whether a photograph to be detected hides information, then controller loads analyzer needed from disk and gives the photograph to be detected to the preprocessor to carry out pretreatment, and pretreated result is distributed to analyzer, which returns a synthetical analyzed result back to user. If user's input information is to train new analyzer, then controller calls preprocessor and training photograph library, and generated new analyzer is stored in disk for further use. Advantages are: high precision for detecting photographs hided by F5 and Outguess, flexibility, and capable to generate new analyzer.
Description
Technical field
What the present invention relates to is a kind of system of areas of information technology, specifically, is a kind of picture concealed blind-detection system based on high-order statistic.
Background technology
Information hiding in existing information, is compared the Information hiding that exchanges with cryptography, allow the enemy to detect, and intercepts and captures, and does not violate the safe limit of system.The purpose of Information hiding be with Information hiding in other harmless information, and do not allow the enemy to detect the existence of out of Memory.The carrier of hiding is general numerical information, and at present modal is the JPEG picture.The information that is hidden is any information that can embed in the data stream, and embedding information and information carrier have formed pseudo-carrier together.
The instrument that is used for Information hiding at present mainly contains F5, Outguess, and Jpeg etc., but also there is not special picture concealed blind-detection system.For the image concealing message context, certain methods was once proposed in the world, blind checking method as the farid proposition, at first utilize multistage discrete wavelet to decompose the higher order statistical characteristic of extracting the training pictures, obtain the proper vector of picture, expense She Er linear separator (FLD) is classified to the proper vector of training pictures then, obtain the classification thresholding according to certain algorithm, characteristic vector space is divided, obtain some intervals, the high-order statistic for the treatment of the mapping sheet again calculates, and sees which interval it drops on, and just can judge whether picture contains to hide Info.Though FLD can handle more parameter, but its recognition function is linear, is difficult to handle the classification problem of multidimensional data.
Summary of the invention
The objective of the invention is to overcome deficiency of the prior art, a kind of picture concealed blind-detection system based on high-order statistic is provided.Make it by analyzing the high-order statistic of picture to be measured, detect picture and whether contain useful F5, the information that the Outguess instrument is hidden.
The present invention is achieved by the following technical solutions, the present invention includes: pretreater, training aids, control phase and analyzer.Controller is the core of each member of control system and the I of total system, pretreater, analyzer and training aids are controlled and called to controller, when whether the information of user input is to analyze certain picture to be measured to contain when hiding Info, controller is written into the analyzer that needs from disk, give pretreater with the picture to be measured of input and carry out pre-service, be distributed to relevant analyzer then, and return to comprehensive analysis result of user; When the input information that receives the user when controller is the new analyzer of training, directly call pretreater and training aids training plan valut, again the analyzer that generates is stored in the disk, detect picture afterwards for controller and whether contain when hiding Info and call.
Pretreater is that the input picture is carried out pre-service, comprises gray processing, standard picture size and extraction picture feature value, and gray processing is picture to be carried out color mode transform (as the JPEG picture), to obtain the gray processing picture; Then the picture behind the gray processing is carried out the standard size, as the picture of all size (360*460,520*391 etc.) is unified the picture that standard is 640*480; At last the picture after the standard size is extracted proper vector, carrying out 2 grades of discrete wavelets decomposes, and calculate preceding four high-order statistics of wavelet coefficients at different levels: average, variance, gradient and kurtosis (mean, variance, skewness, kurtosis), every like this width of cloth picture just can obtain 24 proper vectors.
Training aids makes up all kinds of analyzers by the training fuzzy neural network, it sends into pretreater to the picture of training plan valut earlier, obtain the proper vector of pictures,, can generate the analyzer of corresponding format again with these proper vector training fuzzy neural networks.The function of training aids is to generate the various analyzers that comprise fuzzy rule base by the training fuzzy neural network.In controller input training information, (promptly require to generate the analyzer of certain newtype) the user, a training plan valut also must be provided.Training aids is earlier sent into pretreater to pictures and is carried out pre-service, and the eigenwert that pretreater is returned is as the input vector of fuzzy neural network then.Every width of cloth picture has 24 eigenwerts in native system, if the training plan valut has 1800 width of cloth pictures, the input of fuzzy neural network is exactly the vector of 1800 24 dimensions so.With fuzzy neural network of these input vector training, just can generate the analyzer of a particular type.Make up the fuzzy neural network of other type according to this kind method, can obtain the analyzer of other types after the repetition training.
Controller is the core component of total system, also is the I of system, and it can receive the demand information of user's input, selects for use corresponding member to carry out analyzing and processing, returns corresponding result at last and gives the user.When the user imports the information that needs certain width of cloth picture of detection, whether promptly need to analyze a width of cloth or a series of pictures contains and hides Info, and when having used which kind of hide tools, controller at first loads required analyzer, receive the picture to be measured of extraneous input, then picture to be measured is sent into pretreater, extract proper vector, proper vector with picture to be measured is distributed to corresponding analyzer again, such as, the proper vector of JPEG picture is distributed to JPEG analyzer and the hiding JPEG analyzer of Outguess that F5 hides, and the result that controller returns according to each analyzer finally returns a comprehensive result.The information of importing as the user is when needing the analyzer of certain newtype of generation, controller calls pretreater, and the training plan valut that the user is provided carries out pre-service, obtains the eigenwert of pictures, these eigenwerts are sent into fuzzy neural network train, just can generate new analyzer.
Native system has been trained the analyzer of generation: comprise hiding hiding JPEG picture analyzing device and the hiding GIF picture analyzing device of Outguess of GIF picture analyzing device, Outguess of JPEG picture analyzing device, F5 that F5 hides, but every alanysis device all controlled device calls, they are generated by the training fuzzy neural network by training aids, every alanysis device can controlled device call, picture to corresponding format carries out analyzing and testing, sees whether it contains the information that corresponding hide tools is hidden.As the JPEG picture analyzing device that F5 hides, it comes down to a fuzzy neural network, the picture feature vector of jpeg format is sent into this analyzer analyze, and can obtain a value between 0-1.If this value, illustrates that picture contains the information that F5 hides greater than 0.5, right and wrong nature picture; If this value illustrates that picture does not contain to hide Info less than 0.5, is the nature picture.
The present invention has overcome the linear caused low precision identification problem to hiding Info by the FLD recognition function, has made full use of the learning ability of the non-linear and neural network of fuzzy clustering, has improved accuracy of identification.Show the present invention to F5 through after the performance test, the information that Outguess hides has very high discrimination, can reach more than 85%.And the present invention has very big dirigibility, can generate the analyzer that be applicable to other picture format or other hide tools by the training aids training.
Description of drawings
Fig. 1 system architecture figure
Fig. 2 is used for the fuzzy neural network in generation rule storehouse
Embodiment
As shown in Figure 1, native system mainly comprises 4 parts: pretreater, training aids, controller and analyzer.Pretreater can be called by training aids and controller, and the input picture is carried out pre-service, comprises gray processing, standard picture size and extraction picture feature value.Picture after native system requires pretreater to all gray processings is unified the picture that standard is the 640*480 size, then every width of cloth picture being carried out 2 grades of discrete wavelets decomposes, the runic content of obtaining each grade and detail content (coefficient of level, vertical, diagonal line subband), calculate preceding four high-order statistics of wavelet coefficients at different levels again: average, variance, gradient and kurtosis (mean, variance, skewness, kurtosis), every like this width of cloth picture just can obtain 24 proper vectors.These proper vectors turn back to training aids or the controller that calls pretreater at last, so that they carry out other operations.
Training aids can directly generate all kinds of analyzers, and this mainly realizes by the training fuzzy neural network.In order to construct certain type analyzer, the JPEG picture analyzing device hiding as F5 needs to set up a fuzzy neural network, as shown in Figure 2.The input of this fuzzy neural network is the proper vector of picture, and every width of cloth picture has 24 proper vectors, if the input picture has 1800 width of cloth, the input of fuzzy neural network is exactly the proper vector of 1800 24 dimensions so.Input feature value is carried out fuzzy clustering, be divided into the R group, generate R corresponding neural network and a degree of membership neural network then, this degree of membership neural network is to be used for providing and the relevance grade μ of R the corresponding neural network of grouping to input picture feature value
j,, just can generate relevance grade and R bar fuzzy rule f to this R+1 neural network repetition training
j(x), (j=1,2 ..., R) calculate formula
Just can obtain the output of whole fuzzy neural network.So an analyzer is input as picture feature vector X exactly, is output as the fuzzy neural network of y.Training aids can carry out repetition training to given training plan valut under the calling of controller, generate various types of analyzers that comprise fuzzy rule base.
Controller is the core of total system, also is the I of system, and it can directly call pretreater, training aids and analyzer.If whether controller receives user's input information is to detect picture to contain when hiding Info, picture to be measured is sent into pretreater carry out pre-service, obtain the proper vector of picture, calling corresponding analyzer according to picture format then analyzes, value of analyzer output, scope is between 0 to 1, and the result that controller returns according to each analyzer finally returns a comprehensive result.Such as picture format to be measured is JPEG, controller just calls the GIF picture analyzing device that F5 hides, the JPEG picture analyzing device that Outguess hides, the proper vector of picture is sent into this two analyzers, these two analyzers can return each contented r1 as a result, r2, at this moment, system makes decision in the following manner:
| The result that controller returns | The JPEG picture analyzing device that Outguess hides | ||
| ??＜0.5 | ??≥0.5 | ||
| The JPEG picture analyzing device that F5 hides | ??＜0.5 | ??＜0.5 | ??≥0.5 |
| ??≥0.5 | ??≥0.5 | ??≥0.5 | |
Wherein,＜0.5 explanation picture does not contain and hides Info, and 〉=0.5 explanation picture contains and hides Info.R=max{r as a result as seen from the above table
1, r
2..., r
n.Such as, the probable value that the JPEG picture analyzing device that F5 hides returns is 0.82, but the probable value of the JPEG picture analyzing device that Outguess hides is 0.7, system just tells the user so: this picture comprises hiding Info of F5 embedding, and possibility is 82%.If the user profile that controller receives is to train when generating new analyzer, directly calls training aids training fuzzy neural network and generate required analyzer.
Native system has trained the analyzer of generation to have 4: the JPEG picture analyzing device that F5 hides, the GIF picture analyzing device that F5 hides, the GIF picture analyzing device that JPEG picture analyzing device that Outguess hides and Outguess hide.Each analyzer all is to be generated by training aids, and can controlled device call.The input of analyzer is the proper vector value of picture, and output is one 0～1 positive integer.Send into the JPEG picture analyzing device that Outguess hides such as the picture feature vector of certain jpeg format, output 0.8 illustrates that this picture contains to hide Info, and the probability that contains the information that useful Outguess hides is 0.8.Each analyzer all is a fuzzy neural network independently, can analyze a picture separately, exports a probable value.
Claims (10)
1, a kind of picture concealed blind-detection system based on high-order statistic, it is characterized in that, comprise: pretreater, training aids, control phase and analyzer, pretreater, analyzer and training aids are controlled and called to controller, when whether the information of user input is to analyze certain picture to be measured to contain when hiding Info, controller is written into the analyzer that needs from disk, gives pretreater with the picture to be measured of input and carries out pre-service, be distributed to analyzer then, and return to comprehensive analysis result of user; When the input information that receives the user when controller is the new analyzer of training, directly call pretreater and training aids training plan valut, again the analyzer that generates is stored in the disk, detect picture afterwards for controller and whether contain when hiding Info and call.
2, the picture concealed blind-detection system based on high-order statistic according to claim 1 is characterized in that, described controller is the core of each member of control system and the I of total system.
3, picture concealed blind-detection system based on high-order statistic according to claim 1, it is characterized in that, described pretreater, be that the input picture is carried out pre-service, comprise gray processing, standard picture size and extraction picture feature value, gray processing is picture to be carried out color mode transform, and to obtain the gray processing picture, then the picture behind the gray processing is carried out the standard size, at last the picture after the standard size is extracted proper vector, carry out 2 grades of discrete wavelets and decompose, and calculate preceding four high-order statistics of wavelet coefficients at different levels: average, variance, gradient and kurtosis, every like this width of cloth picture just can obtain 24 proper vectors.
4, the picture concealed blind-detection system based on high-order statistic according to claim 3 is characterized in that, described standard size is meant the picture of all size is unified the picture that standard is 640*480.
5, the picture concealed blind-detection system based on high-order statistic according to claim 1, it is characterized in that, described analyzer, the system that is meant has trained the analyzer of generation, comprise: the GIF picture analyzing device that JPEG picture analyzing device that the GIF picture analyzing device that the JPEG picture analyzing device that F5 hides, F5 hide, Outguess hide and Outguess hide, picture to corresponding format carries out analyzing and testing, see that whether it contains the information that corresponding hide tools is hidden, and obtains a value between 0-1.
6, the picture concealed blind-detection system based on high-order statistic according to claim 5 is characterized in that, described analyzer, but they all controlled device call, they by training aids by the training fuzzy neural network generate.
7, the picture concealed blind-detection system based on high-order statistic according to claim 5, it is characterized in that, the JPEG picture analyzing device that described F5 hides, come down to a fuzzy neural network, the picture feature vector of jpeg format is sent into this analyzer analyze, obtain a value, if this value is greater than 0.5 between 0-1, illustrate that picture contains the information that F5 hides, right and wrong nature picture; If this value illustrates that picture does not contain to hide Info less than 0.5, is the nature picture.
8, picture concealed blind-detection system based on high-order statistic according to claim 1, it is characterized in that, described training aids, make up all kinds of analyzers by the training fuzzy neural network, earlier the picture of training plan valut is sent into pretreater, obtain the proper vector of pictures, again with these proper vector training fuzzy neural networks, can generate the analyzer of corresponding format, the user in controller input training information, a training plan valut also must be provided, and training aids is earlier sent into pretreater to pictures and is carried out pre-service, and the eigenwert that pretreater is returned is as the input vector of fuzzy neural network then.
9, the picture concealed blind-detection system based on high-order statistic according to claim 1, it is characterized in that described controller is the demand information that receives user's input, select for use corresponding member to carry out analyzing and processing, return corresponding result at last and give the user.
10, the picture concealed blind-detection system based on high-order statistic according to claim 9, it is characterized in that, described controller, when the user imports the information that needs certain width of cloth picture of detection, whether promptly need to analyze a width of cloth or a series of pictures contains and hides Info, and when having used which kind of hide tools, controller at first loads required analyzer, receive the picture to be measured of extraneous input, then picture to be measured is sent into pretreater, extract proper vector, the proper vector with picture to be measured is distributed to corresponding analyzer again; The information of importing as the user is when needing the analyzer of certain newtype of generation, controller calls pretreater, and the training plan valut that the user is provided carries out pre-service, obtains the eigenwert of pictures, these eigenwerts are sent into fuzzy neural network train, just can generate new analyzer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200510026803 CN1710594A (en) | 2005-06-16 | 2005-06-16 | Picture concealed blind-detection system based high-order statistic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200510026803 CN1710594A (en) | 2005-06-16 | 2005-06-16 | Picture concealed blind-detection system based high-order statistic |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN1710594A true CN1710594A (en) | 2005-12-21 |
Family
ID=35706835
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200510026803 Pending CN1710594A (en) | 2005-06-16 | 2005-06-16 | Picture concealed blind-detection system based high-order statistic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1710594A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101615286B (en) * | 2008-06-25 | 2012-01-04 | 中国科学院自动化研究所 | Blind hidden information detection method based on analysis of image gray run-length histogram |
| CN101414378B (en) * | 2008-11-24 | 2012-09-05 | 罗向阳 | Hidden blind detection method for image information with selective characteristic dimensionality |
| CN107103630A (en) * | 2017-03-03 | 2017-08-29 | 中国科学院信息工程研究所 | A kind of carrier-free concealed communication method based on GIF attribute interval division mapping codes |
-
2005
- 2005-06-16 CN CN 200510026803 patent/CN1710594A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101615286B (en) * | 2008-06-25 | 2012-01-04 | 中国科学院自动化研究所 | Blind hidden information detection method based on analysis of image gray run-length histogram |
| CN101414378B (en) * | 2008-11-24 | 2012-09-05 | 罗向阳 | Hidden blind detection method for image information with selective characteristic dimensionality |
| CN107103630A (en) * | 2017-03-03 | 2017-08-29 | 中国科学院信息工程研究所 | A kind of carrier-free concealed communication method based on GIF attribute interval division mapping codes |
| CN107103630B (en) * | 2017-03-03 | 2019-11-26 | 中国科学院信息工程研究所 | A kind of carrier-free concealed communication method based on GIF attribute interval division mapping code |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Fu et al. | Detection of image splicing based on Hilbert-Huang transform and moments of characteristic functions with wavelet decomposition | |
| CN108052980B (en) | Image-based air quality grade detection method | |
| CN109948692B (en) | Computer-generated picture detection method based on multi-color space convolutional neural network and random forest | |
| CN102063907B (en) | Steganalysis method for audio spread-spectrum steganography | |
| CN105740707A (en) | Malicious file identification method and device | |
| CN106203492A (en) | The system and method that a kind of image latent writing is analyzed | |
| CN105701470A (en) | Analog circuit fault characteristic extraction method based on optimal wavelet packet decomposition | |
| Kaur et al. | Review of various steganalysis techniques | |
| CN115759640B (en) | Public service information processing system and method for smart city | |
| CN110287888A (en) | A kind of TV station symbol recognition method and system | |
| CN103310235A (en) | Steganalysis method based on parameter identification and estimation | |
| CN116363440A (en) | Deep learning-based identification and detection method and system for colored microplastic in soil | |
| CN115277180A (en) | Block chain log anomaly detection and tracing system | |
| CN1710594A (en) | Picture concealed blind-detection system based high-order statistic | |
| CN110020637A (en) | A kind of analog circuit intermittent fault diagnostic method based on more granularities cascade forest | |
| CN117373136A (en) | Face counterfeiting detection method based on frequency mask and attention consistency | |
| CN110071884A (en) | A kind of Modulation Recognition of Communication Signal method based on improvement entropy cloud feature | |
| CN106530199B (en) | Multimedia integration steganalysis method based on window type hypothesis testing | |
| CN110309737A (en) | A kind of information processing method applied to cigarette sales counter, apparatus and system | |
| CN112487991B (en) | High-precision load identification method and system based on characteristic self-learning | |
| Yu et al. | A multi-scale feature selection method for steganalytic feature GFR | |
| CN114677333A (en) | Image contrast enhancement detection method based on histogram | |
| CN103778210A (en) | Method and device for judging specific file type of file to be analyzed | |
| KR20100082683A (en) | Methods for designing self-educated digital image analyzing engine and database | |
| CN117648607B (en) | Cloud computing-based data comprehensive research and judgment analysis system and method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |