CN110390941A - MP3 audio hidden information analysis method and device based on coefficient correlation model - Google Patents
MP3 audio hidden information analysis method and device based on coefficient correlation model Download PDFInfo
- Publication number
- CN110390941A CN110390941A CN201910586062.8A CN201910586062A CN110390941A CN 110390941 A CN110390941 A CN 110390941A CN 201910586062 A CN201910586062 A CN 201910586062A CN 110390941 A CN110390941 A CN 110390941A
- Authority
- CN
- China
- Prior art keywords
- coefficient
- audio
- qmdct
- model
- analyzed
- 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
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/018—Audio watermarking, i.e. embedding inaudible data in the audio signal
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
Abstract
The MP3 audio hidden information analysis method and device based on coefficient correlation model that the invention discloses a kind of, wherein method is the following steps are included: obtain MP3 audio QMDCT coefficient matrix to be analyzed, using QMDCT coefficient as mode input；A variety of QMDCT associative mode coefficient vectors are extracted from MP3 audio QMDCT coefficient matrix by Recognition with Recurrent Neural Network；The audio associative mode coefficient vector analyzed by tagsort network handles is classified, and the probability whether MP3 audio to be analyzed carries out steganography is obtained.This method can realize higher Stego-detection rate to a variety of steganographic algorithms based on MP3 quantization encoding process, there is certain versatility, and dependence for manual feature is eliminated on model, and the model designed is fairly simple, it can be realized lower detection time-consuming.
Description
Technical field
The present invention relates to digital audio Stego-detection technical field, in particular to a kind of MP3 based on coefficient correlation model
Audio hidden information analysis method and device.
Background technique
Existing to be simply divided into two classes for MP3 compression domain Stego-detection method, one kind is conventional method, a kind of
It is by the way of deep learning, traditional approach mainly passes through the extraction feature of manpower work, is then divided using classifier
Class.
Detection method based on deep learning is then largely carried out feature extraction and is classified using neural network.It is based on
The input of the method for MP3 compression domain Stego-detection is mainly with QMDCT (Quantified Modified Discrete Cosine
Transform, audio quantization correct cosine transform coefficient matrix) based on.For a MP3 audio, usually for
The audio of given frame number, each available n*576 QMDCT coefficient of frame, it is number of channels and granule number that wherein n, which depends on,
Amount.The existing conventional representative steganographic algorithm analyzed based on QMDCT coefficient mainly has ADOTP steganalysis and MDI2
Algorithm, wherein ADOTP algorithm uses first-order difference to that filtering of QMDCT coefficient first, then using markov matrix to being
Number carries out modeling and extracts feature, is then input in classifier and is trained.MDI2 algorithm has calculated separately each ranks
Then difference calculates markov one-step transition probability, feature input classifier is calculated.Algorithm based on deep learning is such as
Multiple dimensioned correlation also substantially carries out feature construction by the way of first manual designs high-pass filtering filtering, then using mind
Mode through network carries out further feature extraction.
All in all, the either conventional method or neural network method of method before this depends primarily on manual extraction
Feature (is related to designing filter), has certain subjectivity, conventional method has ignored the transfer characteristic of high-order, and reason is to mention
Take complexity excessively high, the method based on deep learning lays particular emphasis on level characteristics based on then mainly modeling with convolutional neural networks, and
The depth network number of plies generally constructed is excessive, and detection is time-consuming too long.In addition, detection algorithm before this is in versatility and complicated item
Verification and measurement ratio under part is not generally high.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.
For this purpose, an object of the present invention is to provide a kind of audio hidden information analysis sides MP3 based on coefficient correlation model
Method, this method can be realized higher steganalysis verification and measurement ratio, and still performance is good under the conditions of low insertion rate, and coefficient
Correlation model complexity is relatively low, can guarantee preferable Real time detection performance.
It is another object of the present invention to propose a kind of MP3 audio hidden information analysis device based on coefficient correlation model.
In order to achieve the above objectives, one aspect of the present invention embodiment proposes a kind of MP3 audio based on coefficient correlation model
Steganalysis method, comprising the following steps: MP3 audio QMDCT coefficient matrix to be analyzed is obtained, by the QMDCT coefficient
As mode input；A variety of QMDCT relevant modes are extracted from the MP3 audio QMDCT coefficient matrix by Recognition with Recurrent Neural Network
Formula coefficient vector；Classified by tagsort network to the audio associative mode coefficient vector to be analyzed, obtain to
Whether the MP3 audio of analysis carries out the probability of steganography.
The MP3 audio hidden information analysis method based on coefficient correlation model of the embodiment of the present invention, again with the viewpoint of sequence
MP3 tonic train is modeled, and using Recognition with Recurrent Neural Network building coefficient correlation model between captured frame and frame, it is each in frame
Correlated characteristic between coefficient, and directly eliminate the process filtered by hand using QMDCT coefficient, and can modeling comparison it is remote
The coefficient of distance relies on and the directional information of coefficient, so as to realize higher steganalysis verification and measurement ratio, and in low insertion rate
Under the conditions of still performance it is good, and coefficient correlation model complexity is relatively low, can guarantee preferable Real time detection performance.
In addition, the MP3 audio hidden information analysis method according to the above embodiment of the present invention based on coefficient correlation model may be used also
With following additional technical characteristic:
Further, in one embodiment of the invention, the QMDCT coefficient sequence that each MP3 audio obtains indicates
Are as follows:
X=[x1,x2,…,xK],
Wherein, xKFor the coefficient vector after each frame amount.
Further, in one embodiment of the invention, further includes: right described in coefficient of utilization associated extraction model
QMDCT coefficient sequence carries out correlation analysis, obtains coefficient correlated characteristic vector.
Further, in one embodiment of the invention, the tagsort network are as follows:
st=tanh (Wt·yt+bt),
Zk=Vk*sk+bk,
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding pre-set categories.
Further, in one embodiment of the invention, Stego-detection formula are as follows:
Wherein, z is the probability value of output, and threshold is preset judgment threshold, and T is the class label of model.
In order to achieve the above objectives, another aspect of the present invention embodiment proposes a kind of MP3 sound based on coefficient correlation model
Frequency hidden information analysis device, comprising: module is obtained, it, will be described for obtaining MP3 audio QMDCT coefficient matrix to be analyzed
QMDCT coefficient is as mode input；Extraction module, for passing through Recognition with Recurrent Neural Network from the MP3 audio QMDCT coefficient matrix
It is middle to extract a variety of QMDCT associative mode coefficient vectors；Categorization module, for passing through tagsort network to the sound to be analyzed
Frequency associative mode coefficient vector is classified, and the probability whether MP3 audio to be analyzed carries out steganography is obtained.
The MP3 audio hidden information analysis device based on coefficient correlation model of the embodiment of the present invention, again with the viewpoint of sequence
MP3 tonic train is modeled, and using Recognition with Recurrent Neural Network building coefficient correlation model between captured frame and frame, it is each in frame
Correlated characteristic between coefficient, and directly eliminate the process filtered by hand using QMDCT coefficient, and can modeling comparison it is remote
The coefficient of distance relies on and the directional information of coefficient, so as to realize higher steganalysis verification and measurement ratio, and in low insertion rate
Under the conditions of still performance it is good, and coefficient correlation model complexity is relatively low, can guarantee preferable Real time detection performance.
In addition, the MP3 audio hidden information analysis device according to the above embodiment of the present invention based on coefficient correlation model may be used also
With following additional technical characteristic:
Further, in one embodiment of the invention, the QMDCT coefficient sequence that each MP3 audio obtains indicates
Are as follows:
X=[x1,x2,…,xK],
Wherein, xKFor the coefficient vector after each frame amount.
Further, in one embodiment of the invention, further includes: analysis module is used for coefficient of utilization associated extraction
Correlation analysis is carried out to QMDCT coefficient sequence described in model, obtains coefficient correlated characteristic vector.
Further, in one embodiment of the invention, the tagsort network are as follows:
st=tanh (Wt·yt+bt),
Zk=Vk*sk+bk,
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding pre-set categories.
Further, in one embodiment of the invention, Stego-detection formula are as follows:
Wherein, z is the probability value of output, and threshold is preset judgment threshold, and T is the class label of model.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is the MP3 audio encoding process schematic diagram according to the embodiment of the present invention；
Fig. 2 is the MP3 audio decoding process schematic diagram according to the embodiment of the present invention；
Fig. 3 is schematic diagram of being classified according to the compressed domain audio steganography main algorithm of the embodiment of the present invention；
Fig. 4 is the flow chart according to the MP3 audio hidden information analysis method based on coefficient correlation model of the embodiment of the present invention；
Fig. 5 is to implement main process schematic diagram according to the algorithm of the embodiment of the present invention；
Fig. 6 is the coefficient correlation model structural schematic diagram according to the embodiment of the present invention；
Fig. 7 is the Recognition with Recurrent Neural Network basic framework schematic diagram according to the embodiment of the present invention；
Fig. 8 is the structural schematic diagram according to the bidirectional circulating neural network of the embodiment of the present invention；
Fig. 9 is the structural schematic diagram according to the shot and long term memory network of the embodiment of the present invention；
Figure 10 is the frame training algorithm flow chart according to the supervised learning of the embodiment of the present invention；
Figure 11 is to be shown according to the structure of the MP3 audio hidden information analysis device based on coefficient correlation model of the embodiment of the present invention
It is intended to.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
Before introducing the MP3 audio hidden information analysis method and device based on coefficient correlation model, first simply introduce
Steganography and MP3 steganographic algorithm.
Steganography is the key technology in system of hiding.The main target of Steganography is based on disclosed various media informations
As carrier, it is embedded in information as much as possible, and guarantees that the various statistical natures of steganography front and back carrier do not change as far as possible.Steganography
The purpose of analysis is to detect the information of various insertions as far as possible in all suspicious carriers.In general more after steganography
Media file has good not sentience, during transmission the general more difficult secret information for being detected its carrying,
The specific crowd that only hidden writer specifies can extract secret information from the multimedia file after steganography.Since information is hidden
This characteristic write, it is easy to by criminal using carrying out information transmitting, it is more likely that caused to national security very big
Danger.Therefore Steganalysis has obtained more and more concerns.
In general the Digital Media of steganography includes image, audio and text etc..Audio file is due in recent years in net
Widespread in network, to become the hot spot of many hidden writers and steganalysis detection researcher's concern.Particularly,
MP3 audio is instantly the most popular one of audio file formats, and compared to wav format, MP3 format has very high compression
Than, and it is able to maintain extraordinary sound quality.
The coding principle of MP3 as shown in Figure 1, the decoding principle of MP3 as shown in Fig. 2, for MP3 audio steganography algorithm, root
It is believed that the difference of breath embedded location, as shown in figure 3, can be mainly divided into following several classes: (1) information based on time-frequency domain signal is embedding
Enter；(2) it is embedded in based on quantization and the information of programming process；(3) the information insertion based on code stream domain.
Particularly, the method for the embodiment of the present invention is the insertion steganalysis for quantization and cataloged procedure, based on quantization
With the insertion steganalysis of cataloged procedure be mainly say telescopiny in compressed encoding quantization compression combine, may be implemented compared with
High concealment.
MP3 steganographic algorithm than more typical compression domain mainly includes MP3Stego, based on the hidden of Huffman code word mapping
Write algorithm (HCM) and adaptive M P3 steganographic algorithm (EECS) scheduling algorithm based on the replacement of isometric entropy code word.Wherein MP3stego
It is that insertion is completed in the interior circulation of MP3 coding, algorithm is according to the parity of part_2_3_length come embedding information.HCM is hidden
It writes algorithm to modify to the Huffman code word in MP3 in the case where not destroying MP3 code flow structure, mainly to code word
Classify and is then embedded according to different code-word types.EECS algorithm is based primarily upon isometric entropy code word replacement and realizes that password disappears
The insertion of breath.In the design for being adaptively mainly reflected in cost function in telescopiny, syndrome trellis coding and cost
Construction of function greatly enhances the safety of steganography.
The MP3 audio steganography based on coefficient correlation model proposed according to embodiments of the present invention point is described with reference to the accompanying drawings
Method and device is analysed, describes the MP3 sound based on coefficient correlation model proposed according to embodiments of the present invention with reference to the accompanying drawings first
Frequency steganalysis method.
Fig. 4 is the flow chart of the MP3 audio hidden information analysis method based on coefficient correlation model of one embodiment of the invention.
As shown in figure 4, should MP3 audio hidden information analysis method based on coefficient correlation model the following steps are included:
In step S401, MP3 audio QMDCT coefficient matrix to be analyzed is obtained, QMDCT coefficient is defeated as model
Enter.
Specifically, as shown in figure 5, the audio file such as wav formatted file of 2N different style of selection, sound intermediate frequency
File number N > 10000 carry out audio compression to wherein N number of audio using mp3 condensing encoder and obtain MP3 audio file.It is right
Remaining N number of audio uses the MP3 audio steganography tool based on quantization or coding to believe the secret of different content different length
Breath is embedded in obtain the MP3 audio file of steganography.Steganography audio and non-steganography audio constitute steganalysis sample database.
Decompression is decoded to MP3 audio using MP3 audio decoder, in decompression process, available solution
The coefficient of compression.For giving sample, each frame decompression of audio is commonly available n*576 QMDCT coefficient, and n depends on
The sound channel and numbers of particles of audio.If single audio frequency is segmented into K frame, it is obtained to each audio QMDCT system
Number Sequence can indicate are as follows:
X=[x1,x2,…,xK]。
In step S402, a variety of QMDCT phases are extracted from MP3 audio QMDCT coefficient matrix by Recognition with Recurrent Neural Network
Close mode coefficient vector.
Further, in one embodiment of the invention, the method for the embodiment of the present invention further include: coefficient of utilization is related
It extracts model and correlation analysis is carried out to QMDCT coefficient sequence, obtain coefficient correlated characteristic vector.
Specifically, as shown in figure 5, using coefficient associated extraction model as shown in FIG. 6 to QMDCT coefficient sequence H into
Row correlation analysis, available coefficient correlated characteristic vector W=[g1；g2,…；gn].The wherein specific implementation of coefficient correlation model
Based on Recognition with Recurrent Neural Network, basic framework with formula as shown in fig. 7, expressed are as follows:
Wherein stT moment recirculating network node state is represented, x is t moment input vector, and b is linear bias, and W is input
Weight, V are output weight.
In order to capture the direction relations between coefficient, the embodiment of the present invention uses bidirectional circulating neural network come the side of extraction
To information, basic structure using formula as shown in figure 8, can be expressed as:
Wherein, h is node state vector, and W is weighing vector, and b is bias vector, and y is output vector.Wherein, H is length
Phase memory network (LSTM), structure is as shown in Figure 9.It can be stated with formula are as follows:
Wherein, W is weight vector, and i is input gate vector, and f is to forget door vector, and o is out gate vector, x be input to
Amount, b are linear bias, and h is node state, and σ is usually sigmoid function, and ⊙ is point-by-point operation.
In step S403, the audio associative mode coefficient vector analyzed by tagsort network handles is classified,
Obtain the probability whether MP3 audio to be analyzed carries out steganography.
Specifically, as shown in figure 5, coefficient correlated characteristic vector obtained in step 402 is used tagsort model
Classify, wherein tagsort network can be described as:
st=tanh (Wt·yt+bt),
Zk=Vk*sk+bk,
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding a certain classification, Z is to finally obtain
Each classification normalization probability.Particularly for Stego-detection, have:
Wherein, z is the probability value exported before this, and threshold is given judgment threshold, and T is the class label of model.
Further, for above-mentioned two models, that is, coefficient associated extraction model and tagsort model, the present invention is real
It applies example to be trained using the frame of supervised learning, algorithm flow is as shown in Figure 10, passes through available two models of training
Optimized parameter.For any one MP3 audio, according to above-mentioned step, obtain in an identical manner the QMDCT feature of audio to
Amount, then using train come two model parameters detected, to judge whether MP3 audio passes through steganography.
To sum up, the audio quantization obtained for MP3 audio compression process corrects cosine transform coefficient matrix QMDCT, proposes
A kind of coefficient correlation model carries out steganalysis.The model uses Recognition with Recurrent Neural Network using QMDCT coefficient as input
Itd is proposed coefficient correlated characteristic is extracted, and uses tagsort network as disaggregated model to carry out steganography classification, it is involved
Model obtains model coefficient by there is the learning framework of supervision.The coefficient correlation model binding characteristic classification net finally trained
Network can carry out MP3 audio steganography detection.Present invention method can apply the MP3 sound a variety of based on quantization and coding
Frequency steganographic algorithm such as MP3Stego in EECS scheduling algorithm, has certain versatility and can obtain very high steganography
Detection accuracy.That is, the method for the embodiment of the present invention can be to a variety of steganographic algorithms based on MP3 quantization encoding process
It can realize higher Stego-detection rate, there is certain versatility, and eliminate the dependence for manual feature on model, and set
The model of meter is fairly simple, and it is time-consuming to can be realized lower detection.
The MP3 audio hidden information analysis method based on coefficient correlation model proposed according to embodiments of the present invention, with sequence
Viewpoint models MP3 tonic train again, and using Recognition with Recurrent Neural Network building coefficient correlation model between captured frame and frame,
Correlated characteristic in frame between each coefficient, and the process filtered by hand directly is eliminated using QMDCT coefficient, and can build
The more remote coefficient of mould relies on and the directional information of coefficient, so as to realize higher steganalysis verification and measurement ratio, and
Still performance is good under the conditions of low insertion rate, and coefficient correlation model complexity is relatively low, can guarantee preferable real-time
Detection performance.
The MP3 audio steganography based on coefficient correlation model point proposed according to embodiments of the present invention referring next to attached drawing description
Analysis apparatus.
Figure 11 is that the structure of the MP3 audio hidden information analysis device based on coefficient correlation model of one embodiment of the invention is shown
It is intended to.
As shown in figure 11, should MP3 audio hidden information analysis device 10 based on coefficient correlation model include: obtain module 100,
Extraction module 200 and categorization module 300.
Wherein, module 100 is obtained for obtaining MP3 audio QMDCT coefficient matrix to be analyzed, and QMDCT coefficient is made
For mode input.Extraction module 200 is used to extract from MP3 audio QMDCT coefficient matrix by Recognition with Recurrent Neural Network a variety of
QMDCT associative mode coefficient vector.Categorization module 300 is used for the audio associative mode system analyzed by tagsort network handles
Number vector is classified, and the probability whether MP3 audio to be analyzed carries out steganography is obtained.The device 10 of the embodiment of the present invention can
Higher Stego-detection rate can be realized to a variety of steganographic algorithms based on MP3 quantization encoding process, there is certain versatility, and
The dependence for manual feature is eliminated on model, and the model designed is fairly simple, it is time-consuming to can be realized lower detection.
Further, in one embodiment of the invention, the QMDCT coefficient sequence that each MP3 audio obtains indicates
Are as follows:
X=[x1,x2,…,xK],
Wherein, xKFor the coefficient vector after each frame amount.
Further, in one embodiment of the invention, the device 10 of the embodiment of the present invention further include: analysis module.
Wherein, analysis module carries out correlation analysis to QMDCT coefficient sequence for coefficient of utilization associated extraction model, obtains coefficient correlation
Feature vector.
Further, in one embodiment of the invention, tagsort network are as follows:
st=tanh (Wt·yt+bt),
Zk=Vk*sk+bk,
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding pre-set categories.
Further, in one embodiment of the invention, Stego-detection formula are as follows:
Wherein, z is the probability value of output, and threshold is preset judgment threshold, and T is the class label of model.
It should be noted that the aforementioned explanation to the MP3 audio hidden information analysis embodiment of the method based on coefficient correlation model
Illustrate the MP3 audio hidden information analysis device based on coefficient correlation model for being also applied for the embodiment, details are not described herein again.
The MP3 audio hidden information analysis device based on coefficient correlation model proposed according to embodiments of the present invention, with sequence
Viewpoint models MP3 tonic train again, and using Recognition with Recurrent Neural Network building coefficient correlation model between captured frame and frame,
Correlated characteristic in frame between each coefficient, and the process filtered by hand directly is eliminated using QMDCT coefficient, and can build
The more remote coefficient of mould relies on and the directional information of coefficient, so as to realize higher steganalysis verification and measurement ratio, and
Still performance is good under the conditions of low insertion rate, and coefficient correlation model complexity is relatively low, can guarantee preferable real-time
Detection performance.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. a kind of MP3 audio hidden information analysis method based on coefficient correlation model, which comprises the following steps:
MP3 audio QMDCT coefficient matrix to be analyzed is obtained, using the QMDCT coefficient as mode input；
Extracted from the MP3 audio QMDCT coefficient matrix by Recognition with Recurrent Neural Network a variety of QMDCT associative mode coefficients to
Amount；
Classified by tagsort network to the audio associative mode coefficient vector to be analyzed, is obtained to be analyzed
Whether MP3 audio carries out the probability of steganography.
2. the method according to claim 1, wherein the QMDCT coefficient sequence that each MP3 audio obtains indicates
Are as follows:
X=[x1,x2,…,xK],
Wherein, xKFor the coefficient vector after each frame amount.
3. according to the method described in claim 2, it is characterized by further comprising:
Correlation analysis is carried out to QMDCT coefficient sequence described in coefficient of utilization associated extraction model, obtains coefficient correlated characteristic vector.
4. the method according to claim 1, wherein the tagsort network are as follows:
st=tanh (Wt·yt+bt),
Zk=Vk*sk+bk,
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding pre-set categories.
5. according to the method described in claim 4, it is characterized in that, Stego-detection formula are as follows:
Wherein, z is the probability value of output, and threshold is preset judgment threshold, and T is the class label of model.
6. a kind of MP3 audio hidden information analysis device based on coefficient correlation model characterized by comprising
Module is obtained, it is for obtaining MP3 audio QMDCT coefficient matrix to be analyzed, the QMDCT coefficient is defeated as model
Enter；
Extraction module, for extracting a variety of QMDCT phases from the MP3 audio QMDCT coefficient matrix by Recognition with Recurrent Neural Network
Close mode coefficient vector；
Categorization module, for being classified by tagsort network to the audio associative mode coefficient vector to be analyzed,
Obtain the probability whether MP3 audio to be analyzed carries out steganography.
7. device according to claim 6, which is characterized in that the QMDCT coefficient sequence that each MP3 audio obtains indicates
Are as follows:
X=[x1,x2,…,xK],
Wherein, xKFor the coefficient vector after each frame amount.
8. device according to claim 7, which is characterized in that further include:
Analysis module obtains coefficient for carrying out correlation analysis to QMDCT coefficient sequence described in coefficient of utilization associated extraction model
Correlated characteristic vector.
9. device according to claim 6, which is characterized in that the tagsort network are as follows:
st=tanh (Wt·yt+bt),
Zk=Vk*sk+bk,
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding pre-set categories.
10. device according to claim 9, which is characterized in that Stego-detection formula are as follows:
Wherein, z is the probability value of output, and threshold is preset judgment threshold, and T is the class label of model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910586062.8A CN110390941A (en) | 2019-07-01 | 2019-07-01 | MP3 audio hidden information analysis method and device based on coefficient correlation model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910586062.8A CN110390941A (en) | 2019-07-01 | 2019-07-01 | MP3 audio hidden information analysis method and device based on coefficient correlation model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110390941A true CN110390941A (en) | 2019-10-29 |
Family
ID=68286135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910586062.8A Pending CN110390941A (en) | 2019-07-01 | 2019-07-01 | MP3 audio hidden information analysis method and device based on coefficient correlation model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110390941A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110968845A (en) * | 2019-11-19 | 2020-04-07 | 天津大学 | Detection method for LSB steganography based on convolutional neural network generation |
| CN111768792A (en) * | 2020-05-15 | 2020-10-13 | 宁波大学 | Audio steganalysis method based on convolutional neural network and domain confrontation learning |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102143001A (en) * | 2011-04-02 | 2011-08-03 | 西南科技大学 | Semantic comprehension-based audio resource management method |
| CN102917227A (en) * | 2012-10-29 | 2013-02-06 | 山东省计算中心 | Compressive sensing-based adaptive video information hiding method |
| US20140142958A1 (en) * | 2012-10-15 | 2014-05-22 | Digimarc Corporation | Multi-mode audio recognition and auxiliary data encoding and decoding |
| CN103886864A (en) * | 2014-03-03 | 2014-06-25 | 南京邮电大学 | Method for detecting quantitatively-embedded secret information for DWT domain fidelity compensation |
| CN104123935A (en) * | 2014-07-16 | 2014-10-29 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN104282310A (en) * | 2014-09-26 | 2015-01-14 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
| CN104637484A (en) * | 2015-02-03 | 2015-05-20 | 宁波大学 | MP3 audio steganography detection method based on co-occurrence matrix analysis |
| CN105118512A (en) * | 2015-07-23 | 2015-12-02 | 武汉大学 | General steganalysis method facing AAC digital audio |
| CN106251860A (en) * | 2016-08-09 | 2016-12-21 | 张爱英 | Unsupervised novelty audio event detection method and system towards safety-security area |
| CN107197297A (en) * | 2017-06-14 | 2017-09-22 | 中国科学院信息工程研究所 | A kind of video steganalysis method of the detection based on DCT coefficient steganography |
| CN108875000A (en) * | 2018-06-14 | 2018-11-23 | 广东工业大学 | A kind of semantic relation classification method merging more syntactic structures |
| CN109344244A (en) * | 2018-10-29 | 2019-02-15 | 山东大学 | A kind of the neural network relationship classification method and its realization system of fusion discrimination information |
| CN109817233A (en) * | 2019-01-25 | 2019-05-28 | 清华大学 | Voice flow steganalysis method and system based on level attention network model |
-
2019
- 2019-07-01 CN CN201910586062.8A patent/CN110390941A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102143001A (en) * | 2011-04-02 | 2011-08-03 | 西南科技大学 | Semantic comprehension-based audio resource management method |
| US20140142958A1 (en) * | 2012-10-15 | 2014-05-22 | Digimarc Corporation | Multi-mode audio recognition and auxiliary data encoding and decoding |
| CN102917227A (en) * | 2012-10-29 | 2013-02-06 | 山东省计算中心 | Compressive sensing-based adaptive video information hiding method |
| CN103886864A (en) * | 2014-03-03 | 2014-06-25 | 南京邮电大学 | Method for detecting quantitatively-embedded secret information for DWT domain fidelity compensation |
| CN104123935A (en) * | 2014-07-16 | 2014-10-29 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN104282310A (en) * | 2014-09-26 | 2015-01-14 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
| CN104637484A (en) * | 2015-02-03 | 2015-05-20 | 宁波大学 | MP3 audio steganography detection method based on co-occurrence matrix analysis |
| CN105118512A (en) * | 2015-07-23 | 2015-12-02 | 武汉大学 | General steganalysis method facing AAC digital audio |
| CN106251860A (en) * | 2016-08-09 | 2016-12-21 | 张爱英 | Unsupervised novelty audio event detection method and system towards safety-security area |
| CN107197297A (en) * | 2017-06-14 | 2017-09-22 | 中国科学院信息工程研究所 | A kind of video steganalysis method of the detection based on DCT coefficient steganography |
| CN108875000A (en) * | 2018-06-14 | 2018-11-23 | 广东工业大学 | A kind of semantic relation classification method merging more syntactic structures |
| CN109344244A (en) * | 2018-10-29 | 2019-02-15 | 山东大学 | A kind of the neural network relationship classification method and its realization system of fusion discrimination information |
| CN109817233A (en) * | 2019-01-25 | 2019-05-28 | 清华大学 | Voice flow steganalysis method and system based on level attention network model |
Non-Patent Citations (4)
| Title |
|---|
| HUANG YONGFENG ET AL.: "Steganography in low bit-rate speech streams based on quantization index modulation controlled by keys", 《SCIENCE CHINA》 * |
| LIGUANG ZHANG ET AL.: "Data hiding in MP3 audio by modifing QMDCT coefficients", 《2009 ISECS INTERNATIONAL COLLOQUIUM ON COMPUTING, COMMUNICATION, CONTROL, AND MANAGEMENT》 * |
| ZHONG-LIANG YANG: "RNN-Stega: Linguistic Steganography Based on Recurrent Neural Networks", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 * |
| 郭洪刚 等: "基于差分统计量的MP3Stego隐写分析算法", 《计算机工程与应用》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110968845A (en) * | 2019-11-19 | 2020-04-07 | 天津大学 | Detection method for LSB steganography based on convolutional neural network generation |
| CN110968845B (en) * | 2019-11-19 | 2023-10-27 | 天津大学 | Detection method for LSB steganography based on convolutional neural network generation |
| CN111768792A (en) * | 2020-05-15 | 2020-10-13 | 宁波大学 | Audio steganalysis method based on convolutional neural network and domain confrontation learning |
| CN111768792B (en) * | 2020-05-15 | 2024-02-09 | 天翼安全科技有限公司 | Audio steganalysis method based on convolutional neural network and domain countermeasure learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113936339B (en) | Fighting identification method and device based on double-channel cross attention mechanism | |
| CN108648746A (en) | A kind of open field video natural language description generation method based on multi-modal Fusion Features | |
| Petridis et al. | Deep complementary bottleneck features for visual speech recognition | |
| Sargin et al. | Audiovisual synchronization and fusion using canonical correlation analysis | |
| CN111640444B (en) | CNN-based adaptive audio steganography method and secret information extraction method | |
| CN108960063B (en) | Multi-event natural language description method in video facing event relation coding | |
| CN111652357B (en) | Method and system for solving video question-answer problem by using specific target network based on graph | |
| CN110349136A (en) | A kind of tampered image detection method based on deep learning | |
| CN109817233A (en) | Voice flow steganalysis method and system based on level attention network model | |
| CN109145745B (en) | Face recognition method under shielding condition | |
| Zhang et al. | Generative steganography by sampling | |
| JPH1055444A (en) | Recognition of face using feature vector with dct as base | |
| CN111582225A (en) | Remote sensing image scene classification method and device | |
| CN110968845B (en) | Detection method for LSB steganography based on convolutional neural network generation | |
| CN110827832A (en) | Video identity recognition equipment and method | |
| CN106650617A (en) | Pedestrian abnormity identification method based on probabilistic latent semantic analysis | |
| CN110390941A (en) | MP3 audio hidden information analysis method and device based on coefficient correlation model | |
| CN110232564A (en) | A kind of traffic accident law automatic decision method based on multi-modal data | |
| CN114694255B (en) | Sentence-level lip language recognition method based on channel attention and time convolution network | |
| Nefian et al. | A Bayesian approach to audio-visual speaker identification | |
| CN114999525A (en) | Light-weight environment voice recognition method based on neural network | |
| CN113887573A (en) | Human face forgery detection method based on visual converter | |
| CN109192217B (en) | Multi-class low-rate compressed voice steganography-oriented general information hiding detection method | |
| CN108734144A (en) | A kind of speaker's identity identifying method based on recognition of face | |
| CN111104868B (en) | Cross-quality face recognition method based on convolutional neural network characteristics |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191029 |