CN101847411A - MIDI (Musical Instrument Digital Interface) audio hidden information analysis method and device - Google Patents
MIDI (Musical Instrument Digital Interface) audio hidden information analysis method and device Download PDFInfo
- Publication number
- CN101847411A CN101847411A CN201010188419A CN201010188419A CN101847411A CN 101847411 A CN101847411 A CN 101847411A CN 201010188419 A CN201010188419 A CN 201010188419A CN 201010188419 A CN201010188419 A CN 201010188419A CN 101847411 A CN101847411 A CN 101847411A
- Authority
- CN
- China
- Prior art keywords
- order
- midi
- parameter
- dynamics
- audio frequency
- 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
Images
Landscapes
- Auxiliary Devices For Music (AREA)
- Electrophonic Musical Instruments (AREA)
Abstract
The embodiment of the invention provides an MIDI audio hidden information analysis method, which comprises the following steps: reading the dynamic parameters and the channel parameters of hidden cipher-carrying information in MIDI audio channel instructions; reordering the dynamic parameters according to the ascending order of the channel parameters; abstracting the high-order statistical moment characteristics of first-order and second-order differences of the spatial domains of the ordered dynamic parameters and the high-order statistical characteristics of histogram characteristic functions to constitute 20-dimension characteristic vectors; inputting the 20-dimension characteristic vectors to a support vector machine to carry out training to form a sorter; and inputting MIDI audio to the sorter to detect whether the MIDI audio carries cipher or not. The embodiment of the invention also provides an MIDI audio hidden information analysis device, which comprises a reordering module, a characteristic abstracting module and a training detection module. The blind detection on various kinds of LSB (Linux Standard Base) hidden information can be realized according to the method and the device, and the hidden information analysis algorithm of the invention has very good applicability and has very high sorting accuracy.
Description
Technical field
The present invention relates to the audio hidden information analysis field of authentication, particularly, the present invention relates to a kind of MIDI audio hidden information analysis method and device.
Background technology
Along with the development of infotech, relevant latent of information security write the very big concern that causes people with steganalysis.MIDI (Musical Instrument Digital Interface, musical instrument digital interface) is to propose for the communication issue that solves between the electrophonic musical instrument early 1980s.What MIDI transmitted is not voice signal, but instructions such as note, controlled variable, what its indication MIDI equipment will do, and how to do, as play which note, much volumes etc.They are by the unified MIDI message (MIDI Message) that is expressed as.Adopt asynchronous serial communication during transmission, the standard traffic baud rate is 31.25 * (1 ± 0.01) KBaud.
MIDI is the communication protocol of exchange message between musical instruments and the computing machine, the MIDI audio frequency have file size little, be convenient to advantages such as modification, be widely used in fields such as internet, mobile phones.At present, be that the Information hiding of carrier occurs with the MIDI audio frequency, for ensuring information safety, prevent to utilize the MIDI file to carry out the confidential corespondence activity, the MIDI file is authenticated with steganalysis have very big application potential.
MIDI voice data storage be the instruction sequence of playing back music how, the music note numeral, musical instruments or computing machine carry out reading of MIDI file according to the value that the MIDI instruction sequence is stored, the MIDI file has the different characteristic of other audio files.Latent write that the LSB spatial domain is latent to be write at the MIDI file is modal at present, its principle is that the lowest order that secret information is embedded into MIDI audio frequency dynamics component is embedded secret data, and based on auditory masking effect, people's ear can not effectively be distinguished.
Latent the writing of the LSB of MIDI audio frequency do not change file size, can slightly change MIDI tonequality, and experiment shows, embeds secret information minimum 3 of dynamics component, and people's ear can not effectively be distinguished.At present at the latent detection method of writing of MIDI audio frequency also seldom, existing detection method is based on the latent back file architectural feature of writing, and finds out the latent some features in back of writing, and judges whether the MIDI audio frequency is write by latent.Because MIDI audio frequency individual difference is huge, judges that with some features not only verification and measurement ratio is lower, and can bring very big false alarm rate.
Summary of the invention
Purpose of the present invention is intended to solve at least one of above-mentioned technological deficiency, to realizing the latent blind Detecting of writing of multiple LSB, improves classification accuracy rate especially, has proposed a kind of MIDI audio hidden information analysis method and device.
For achieving the above object, the embodiment of the invention has proposed a kind of MIDI audio hidden information analysis method on the one hand, comprises the steps:
Read in the instruction of MIDI voice-grade channel and hide dynamics parameter and the passage parameter that carries confidential information, described dynamics parameter is arranged according to the ascending order of described passage parameter resequenced;
Extract spatial domain single order and the higher order statistical moment characteristics of second order difference and the higher order statistical feature of histogram feature function of the dynamics parameter after the described ordering, form 20 dimensional feature vectors;
Described 20 dimensional feature vectors input support vector machine is trained the formation sorter, and whether input MIDI audio frequency detects described MIDI audio frequency and carries close in described sorter.
The embodiment of the invention has also proposed a kind of MIDI audio hidden information analysis device on the other hand, comprises reordering module, characteristic extracting module and training detection module.
Described reordering module is used for reading the instruction of MIDI voice-grade channel and hides dynamics parameter and the passage parameter that carries confidential information, described dynamics parameter is arranged according to the ascending order of described passage parameter resequenced;
Described characteristic extracting module is used to extract spatial domain single order and the higher order statistical moment characteristics of second order difference and the higher order statistical feature of histogram feature function of the dynamics parameter after the described ordering, forms 20 dimensional feature vectors;
Described training detection module is used for described 20 dimensional feature vectors input support vector machine is trained the formation sorter, and whether input MIDI audio frequency detects described MIDI audio frequency and carry close in described sorter.
According to method and the device that the embodiment of the invention provides, based on the latent writing detection method of the MIDI audio frequency LSB of statistical nature, can realize the latent blind Detecting of writing to multiple LSB, its steganalysis algorithm has good applicability.And have very high classification accuracy rate, through experiment, average classification accuracy rate can reach more than 95%.
The such scheme that the present invention proposes, very little to the change of existing system, can not influence the compatibility of system, and realize simple, efficient.
Aspect that the present invention adds and advantage part in the following description provide, and part will become obviously from the following description, or recognize by practice of the present invention.
Description of drawings
Above-mentioned and/or additional aspect of the present invention and advantage are from obviously and easily understanding becoming the description of embodiment below in conjunction with accompanying drawing, wherein:
Fig. 1 is the FB(flow block) according to the MIDI audio hidden information analysis method of the embodiment of the invention;
Fig. 2 is the block diagram according to the feature extraction of the embodiment of the invention;
Fig. 3 is for writing the histogrammic synoptic diagram in spatial domain according to the LSB of the embodiment of the invention is latent;
Fig. 4 conceals the synoptic diagram of writing the histogram feature function for the LSB according to the embodiment of the invention;
Fig. 5 is the FB(flow block) according to the training and the classification of the embodiment of the invention;
Fig. 6 is the comparison synoptic diagram according to the classification accuracy rate of the embodiment of the invention;
Fig. 7 writes on ROC curve under the different embedding rates for the LSB according to the embodiment of the invention conceals;
Fig. 8 is the structured flowchart according to the MIDI audio hidden information analysis device of the embodiment of the invention.
Embodiment
Describe embodiments of the invention below in detail, the example of described embodiment is shown in the drawings, and wherein identical from start to finish or similar label is represented identical or similar elements or the element with identical or similar functions.Below by the embodiment that is described with reference to the drawings is exemplary, only is used to explain the present invention, and can not be interpreted as limitation of the present invention.
In order to realize the present invention's purpose, the invention discloses a kind of MIDI audio hidden information analysis method, in conjunction with shown in Figure 1, this method comprises the steps:
S101: read in the instruction of MIDI voice-grade channel and hide dynamics parameter and the passage parameter that carries confidential information, the dynamics parameter is arranged according to the ascending order of described passage parameter resequenced.
At first pre-service is carried out in pre-service to the MIDI audio frequency.In conjunction with shown in Figure 2,, read dynamics parameter and the passage parameter of hiding secret information in the MIDI audio frequency according to given MIDI audio frequency.Most of storage is the MIDI channel instruction in the MIDI file, and dynamics parameter and passage parameter in the above-mentioned MIDI audio frequency are stored in the MIDI channel instruction.
The structure of channel instruction is as shown in table 1.Standard MIDI file has 16 MIDI passages, and the MIDI sound pitch of same passage and dynamics value distribute very regular.The MIDI dynamics is represented the dynamics that the MIDI keyboard knocks, and value is used for characterizing the intensity of note between [0,127].The LSB of MIDI audio frequency is latent to be write, and is the latent LSB position that writes on the dynamics parameter of MIDI audio frequency.If there is tremendous influence the latent LSB position that writes on other parameters to MIDI tonequality.
Trickle change will take place in the latent back dynamics parameter of writing of MIDI, can judge whether the MIDI audio frequency exists latent writing by the statistical nature of analysis dynamics parameter.Because therefore the characteristic of MIDI audio frequency subchannel storage need analyze the latent feature of writing then to the rearrangement of dynamics parameter subchannel, can reflect the latent process of writing of MIDI audio frequency LSB so better.
Suppose command M (i)=[T (i), V (i)] i ∈ 1,2 of MIDI file storage ... N.Wherein, N is the length of MIDI instruction, and T is the passage parameter of MIDI instruction, value between 1 to 16, T ∈ [1,16]; V is the dynamics parameter of MIDI instruction, value between 0 to 127, V ∈ [0,127].
When the MIDI audio frequency is carried out steganalysis, at first according to the parameter value ascending order of T to M resequence to M '=[T ' (i), V ' (i)], the dynamics V ' at subchannel extracts the latent feature of writing of LSB then.
Table 1
The MIDI instruction | ??Delta?Time | ??Event?Type?Value | ??MIDI?Channel | ??Pitch | ??Velocity |
Implication | Interval time | Event type | The MIDI passage | Sound pitch | Dynamics |
Byte length | The variable-length variable | ??4bits | ??4bits | ??1byte | ??1byte |
The better latent feature of writing of reaction when carrying out signature analysis by the dynamics parameter after according to the size of passage parameter the dynamics parameter being resequenced in the above-mentioned steps.
S102: extract spatial domain single order and the higher order statistical moment characteristics of second order difference and the higher order statistical feature of histogram feature function of the dynamics parameter after sorting, form 20 dimensional feature vectors.
1), extracts the higher order statistical moment characteristics of its spatial domain single order and second order difference according to the dynamics parameter after the ordering that obtains in the step 101.
In conjunction with shown in Figure 2, in the present embodiment, the single order and the second order difference of establishing the dynamics parameter V ' after the rearrangement are D ' and D ".Wherein, first order difference D ' and second order difference D " be:
Extract single order and second order difference D ' and D " preceding Fourth-order moment.Wherein, square is a key character of stochastic variable, and establishing X is a stochastic variable, p
x(x) probability density function (PDF) of expression X, the n rank PDF square of X can be defined as:
Because the preceding 4 rank squares that D ' and D " are discrete random variable, calculate D ' and D by following formula ",
Wherein, single order and second order difference D ' and D " preceding Fourth-order moment respectively corresponding MIDI audio frequency rearrangement back dynamics component first order difference and average, variance, the degree of bias and the kurtosis of second order difference variable, extract spatial domain 8 dimensional features altogether.
Fig. 3 shows the contrast of the spatial feature before and after latent the writing.As shown in Figure 3, same fundamental function is counted under the n value, and the normalized histogram distribution probability after latent the writing is less than the normalized histogram distribution probability that conceals before writing.
2) the higher order statistical moment characteristics of extraction histogram feature domain of function.
In above-mentioned steps, the spatial domain single order of ordering back dynamics parameter and the higher order statistical moment characteristics of second order difference have been extracted.But only use the spatial domain statistical nature, because MIDI audio distribution otherness is bigger, the feature difference of extraction is also bigger.Therefore, need further to extract the statistical nature of histogram feature domain of function.The higher order statistical feature of extracting the histogram feature function comprises: the preceding Fourth-order moment of extracting MIDI audio frequency dynamics histogram of component, single order and each fundamental function of second order difference histogram.
Specifically, in the present embodiment, the histogram distribution function of establishing the dynamics V ' after the rearrangement is H=h
V '(n), the histogram distribution function of dynamics first order difference D ' is H '=h
D '(n), second order difference D " histogram distribution function be H "=h
D "(n),
Then the fundamental function (HCF) of MIDI audio frequency dynamics histogram of component is expressed as formula H[k]:
H[k] n rank square M
(hcf) nFor:
According to above-mentioned H[k] and M
(hcf) nCalculate MIDI audio frequency dynamics histogram of component fundamental function, single order and the histogrammic preceding Fourth-order moment of second order difference, the corresponding histogram feature function average of difference, variance, the degree of bias and kurtosis, extract histogram feature domain of function 12 dimension statistical natures altogether, be used for characterizing the latent process of writing of MIDI audio frequency LSB.
Fig. 4 is the changing features of histogram feature domain of function before and after latent the writing.As shown in Figure 4, same fundamental function is counted under the n value, the absolute value H[k of the fundamental function after latent the writing] less than the latent absolute value of writing preceding fundamental function.
To sum up, in step 102, extract the 8 dimension statistical natures in spatial domain and the 12 dimension statistical natures in HCF territory, totally 20 dimension statistical natures
S103: 20 dimensional feature vectors input support vector machine is trained the formation sorter, and whether input MIDI audio frequency detects the MIDI audio frequency and carries close in sorter.
20 of extraction in the step 102 is imported support vector machine (SVM) for statistical nature train, form sorter.
Fig. 5 shows the FB(flow block) of training and classification.In conjunction with Fig. 4 and shown in Figure 5, comprise with the support vector machine training classifier: with 20 dimensional feature vectors that extract in the step 102, the input support vector machine is trained.
Specifically, training classifier after original MIDI audio frequency and year close audio feature extraction is comprised:
1) download the first MIDI resource of i from the MIDI resources bank, i is a positive integer.Carry out at MIDI dynamics component that LSB is latent to be write, carried close MIDI audio collection accordingly and form training set.
In the present embodiment, MIDI resource i is 328
2) according to the method in the step 102, original audio in the training set and a year close audio frequency are extracted spatial domain and histogram feature domain of function 20 dimension statistical natures respectively, and the tag along sort original audio is set is set to 0, carrying close audio setting is 1, then with support vector machine to training set j sample training, j is a positive integer.
In the present embodiment, sample number j is 656.
3) to the detection of input MIDI audio frequency, it is close to judge whether it carries.
In conjunction with shown in Figure 5, download test MIDI audio frequency, to the classified information that is partially submerged into wherein, another part does not embed secret information.MIDI audio frequency after the above-mentioned processing is input to the sorter that previous step trains in rapid.According to the difference training study sorter of original audio frequency with the feature of carrying close audio frequency, the sorter that trains just can detect latent the writing of MIDI audio frequency LSB, and it is close to judge whether the MIDI audio frequency carries.
Fig. 6 shows the comparison synoptic diagram of classification accuracy rate.As shown in Figure 6, the continuous lines segment table shows the classification accuracy rate of the method that present embodiment provides; Discontinuous equal length line segment is represented the classification accuracy rate of the smoothness conversion ratio method in the existing method; Discontinuous unequal length line segment is represented the classification accuracy rate of the quantity of information method of estimation in the existing method.As can be seen from the figure, the classification accuracy rate of the method that provides by present embodiment is the highest.
Fig. 7 shows that LSB is latent to write on ROC curve under the different embedding rates.1 for the embedding rate is 100% ROC curve (receiver operating characteristic, recipient's operating characteristic curve), and 2 are the ROC curve of embedding rate 10%.As shown in Figure 7, the embedding rate is 100% o'clock, and its classification accuracy rate is 100%.The embedding rate is 10% o'clock, and its classification accuracy rate is more than 90%.The steganalysis algorithm that can characterize present embodiment provides has good applicability.
The embodiment of the invention adopts support vector machine (SVM) training classifier, this sorter is based on Statistical Learning Theory, can be from available sample the feature that exists of analyzing and training, be used to predict other samples, in small sample, non-linear and higher-dimension pattern-recognition, show many distinctive advantages.
Method according to the embodiment of the invention provides based on the latent writing detection method of the MIDI audio frequency LSB of statistical nature, can realize the latent blind Detecting of writing to multiple LSB, and its steganalysis algorithm has good applicability.And have very high classification accuracy rate, through experiment, average classification accuracy rate can reach more than 95%.
The embodiment of the invention has also proposed a kind of MIDI audio hidden information analysis device, and in conjunction with shown in Figure 8, this device 100 comprises reordering module 110, characteristic extracting module 120 and training detection module 130
Wherein, reordering module 110 is used for reading the instruction of MIDI voice-grade channel and hides dynamics parameter and the passage parameter that carries confidential information, the dynamics parameter is arranged according to the ascending order of passage parameter resequenced.
Specifically, reordering module 110 at first to the MIDI audio frequency pre-service carry out pre-service.In conjunction with shown in Figure 2, according to given MIDI audio frequency, reordering module 110 reads dynamics parameter and the passage parameter of hiding secret information in the MIDI audio frequency.Most of storage is the MIDI channel instruction in the MIDI file, and dynamics parameter and passage parameter in the above-mentioned MIDI audio frequency are stored in the MIDI channel instruction.
The structure of channel instruction is as shown in table 1.Standard MIDI file has 16 MIDI passages, and the MIDI sound pitch of same passage and dynamics value distribute very regular.The MIDI dynamics is represented the dynamics that the MIDI keyboard knocks, and value is used for characterizing the intensity of note between [0,127].The LSB of MIDI audio frequency is latent to be write, and is the latent LSB position that writes on the dynamics parameter of MIDI audio frequency.If there is tremendous influence the latent LSB position that writes on other parameters to MIDI tonequality.
Trickle change will take place in the latent back dynamics parameter of writing of MIDI, can judge whether the MIDI audio frequency exists latent writing by the statistical nature of analysis dynamics parameter.Because therefore the characteristic of MIDI audio frequency subchannel storage need analyze the latent feature of writing then to the rearrangement of dynamics parameter subchannel, can reflect the latent process of writing of MIDI audio frequency LSB so better.
Suppose command M (i)=[T (i), V (i)] i ∈ 1,2 of MIDI file storage ... N.Wherein, N is the length of MIDI instruction, and T is the passage parameter of MIDI instruction, value between 1 to 16, T ∈ [1,16]; V is the dynamics parameter of MIDI instruction, is worth between 0 to 127 V ∈ [0,127].
When the MIDI audio frequency is carried out steganalysis, the module of retaking 110 at first according to the parameter value ascending order of T to M resequence to M '=[T ' (i), V ' (i)], the dynamics V ' at subchannel extracts the latent feature of writing of LSB then.The better latent feature of writing of reaction when the dynamics parameter after the dynamics parameter being resequenced according to the size of passage parameter by the above-mentioned module 110 of retaking carries out signature analysis.
Characteristic extracting module 120 is used to extract spatial domain single order and the higher order statistical moment characteristics of second order difference and the higher order statistical feature of histogram feature function of the dynamics parameter after the ordering, forms 20 dimensional feature vectors, comprising:
1) according to the dynamics parameter after the ordering that obtains in the reordering module 110, characteristic extracting module 120 is extracted the higher order statistical moment characteristics of its spatial domain single order and second order difference.
In conjunction with shown in Figure 2, in the present embodiment, the single order and the second order difference of establishing the dynamics parameter V ' after the rearrangement are D ' and D ".Wherein, first order difference D ' and second order difference D " be:
Characteristic extracting module 120 is extracted single order and second order difference D ' and D " preceding Fourth-order moment.Wherein, square is a key character of stochastic variable, and establishing X is a stochastic variable, p
x(x) probability density function (PDF) of expression X, the n rank PDF square of X can be defined as:
Because the preceding 4 rank squares that D ' and D " are discrete random variable, calculate D ' and D by following formula ",
Wherein, single order and second order difference D ' and D " preceding Fourth-order moment respectively corresponding MIDI audio frequency rearrangement back dynamics component first order difference and average, variance, the degree of bias and the kurtosis of second order difference variable, extract spatial domain 8 dimensional features altogether.
Fig. 3 shows the contrast of the spatial feature before and after latent the writing.As shown in Figure 3, same fundamental function is counted under the n value, and the normalized histogram distribution probability after latent the writing is less than the normalized histogram distribution probability that conceals before writing.
2) characteristic extracting module 120 is extracted the higher order statistical moment characteristics of histogram feature domain of function.
Characteristic extracting module 120 is extracted the spatial domain single order of ordering back dynamics parameter and the higher order statistical moment characteristics of second order difference.But only use the spatial domain statistical nature, because MIDI audio distribution otherness is bigger, the feature difference of extraction is also bigger.Therefore, characteristic extracting module 120 needs further to extract the statistical nature of histogram feature domain of function.The higher order statistical feature of extracting the histogram feature function comprises: the preceding Fourth-order moment of extracting MIDI audio frequency dynamics histogram of component, single order and each fundamental function of second order difference histogram.
Specifically, in the present embodiment, the histogram distribution function of establishing the dynamics V ' after the rearrangement is H=h
V '(n), the histogram distribution function of dynamics first order difference D ' is H '=h
D '(n), second order difference D " histogram distribution function be H "=h
D "(n),
Then the fundamental function (HCF) of MIDI audio frequency dynamics histogram of component is expressed as formula H[k]:
H[k] n rank square M
(hcf) nFor:
According to above-mentioned H[k] and M
(hcf) nCalculate MIDI audio frequency dynamics histogram of component fundamental function, single order and the histogrammic preceding Fourth-order moment of second order difference, the corresponding histogram feature function average of difference, variance, the degree of bias and kurtosis, characteristic extracting module 120 is extracted histogram feature domain of function 12 dimension statistical natures altogether, is used for characterizing the latent process of writing of MIDI audio frequency LSB.
Fig. 4 is the changing features of histogram feature domain of function before and after latent the writing.As shown in Figure 4, same fundamental function is counted under the n value, the absolute value H[k of the fundamental function after latent the writing] less than the latent absolute value of writing preceding fundamental function.
To sum up, characteristic extracting module 120 is extracted the 8 dimension statistical natures in spatial domain and the 12 dimension statistical natures in HCF territory, totally 20 dimension statistical natures.
20 of extraction in the characteristic extracting module 120 is imported support vector machine (SVM) for statistical nature train, form sorter.
Fig. 5 shows the FB(flow block) of training and classification.In conjunction with Fig. 4 and shown in Figure 5, comprise with the support vector machine training classifier: with 20 dimensional feature vectors that extract in the characteristic extracting module 120, the input support vector machine is trained.
Specifically, training classifier comprises after 120 pairs of original MIDI audio frequency of characteristic extracting module and year close audio feature extraction:
1) training detection module 130 is downloaded the first MIDI resource of i from the MIDI resources bank, and i is a positive integer.Carry out at MIDI dynamics component that LSB is latent to be write, carried close MIDI audio collection accordingly and form training set.
In the present embodiment, MIDI resource i is 328
2) according to characteristic extracting module 120, original audio in 130 pairs of training sets of training detection module and a year close audio frequency extract spatial domain and histogram feature domain of function 20 dimension statistical natures respectively, and the tag along sort original audio is set is set to 0, carrying close audio setting is 1, then with support vector machine to training set j sample training, j is a positive integer.
In the present embodiment, sample number j is 656.
3) detection of 130 pairs of inputs of training detection module MIDI audio frequency, it is close to judge whether it carries.
In conjunction with shown in Figure 5, download test MIDI audio frequency, to the classified information that is partially submerged into wherein, another part does not embed secret information.MIDI audio frequency after the above-mentioned processing is input to the sorter that trains in the previous step.According to the difference training study sorter of original audio frequency with the feature of carrying close audio frequency, the sorter that trains just can detect latent the writing of MIDI audio frequency LSB, and it is close to judge whether the MIDI audio frequency carries.
Fig. 6 shows the comparison synoptic diagram of classification accuracy rate.As shown in Figure 6, the continuous lines segment table shows the classification accuracy rate of the method that present embodiment provides; Discontinuous equal length line segment is represented the classification accuracy rate of the smoothness conversion ratio method in the existing method; Discontinuous unequal length line segment is represented the classification accuracy rate of the quantity of information method of estimation in the existing method.As can be seen from the figure, the classification accuracy rate of the method that provides by present embodiment is the highest.
Fig. 7 shows that LSB is latent to write on ROC curve under the different embedding rates.1 for the embedding rate is 100% ROC curve (receiver operating characteristic, recipient's operating characteristic curve), and 2 are the ROC curve of embedding rate 10%.As shown in Figure 7, the embedding rate is 100% o'clock, and its classification accuracy rate is 100%.The embedding rate is 10% o'clock, and its classification accuracy rate is more than 90%.The steganalysis algorithm that can characterize present embodiment provides has good applicability.
The embodiment of the invention adopts support vector machine (SVM) training classifier, this sorter is based on Statistical Learning Theory, can be from available sample the feature that exists of analyzing and training, be used to predict other samples, in small sample, non-linear and higher-dimension pattern-recognition, show many distinctive advantages.
According to the device that the embodiment of the invention provides, based on the latent writing detection method of the MIDI audio frequency LSB of statistical nature, can realize the latent blind Detecting of writing to multiple LSB, its steganalysis algorithm has good applicability.And have very high classification accuracy rate, through experiment, average classification accuracy rate can reach more than 95%.
One of ordinary skill in the art will appreciate that and realize that all or part of step that the foregoing description method is carried is to instruct relevant hardware to finish by program, described program can be stored in a kind of computer-readable recording medium, this program comprises one of step or its combination of method embodiment when carrying out.
In addition, each functional unit in each embodiment of the present invention can be integrated in the processing module, also can be that the independent physics in each unit exists, and also can be integrated in the module two or more unit.Above-mentioned integrated module both can adopt the form of hardware to realize, also can adopt the form of software function module to realize.If described integrated module realizes with the form of software function module and during as independently production marketing or use, also can be stored in the computer read/write memory medium.
The above-mentioned storage medium of mentioning can be a ROM (read-only memory), disk or CD etc.
The above only is a preferred implementation of the present invention; should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.
Claims (10)
1. a MIDI audio hidden information analysis method is characterized in that, comprises the steps:
Read in the instruction of MIDI voice-grade channel and hide dynamics parameter and the passage parameter that carries confidential information, described dynamics parameter is arranged according to the ascending order of described passage parameter resequenced;
Extract spatial domain single order and the higher order statistical moment characteristics of second order difference and the higher order statistical feature of histogram feature function of the dynamics parameter after the described ordering, form 20 dimensional feature vectors;
Described 20 dimensional feature vectors input support vector machine is trained the formation sorter, and whether input MIDI audio frequency detects described MIDI audio frequency and carries close in described sorter.
2. the method for claim 1 is characterized in that,
Described MIDI voice-grade channel instruction is M (i)=[T (i), V (i)] i ∈ 1,2 ... N, wherein, N is the length of MIDI voice-grade channel instruction; T is the passage parameter of MIDI instruction, T ∈ [1,16]; V is the dynamics parameter of MIDI instruction, V ∈ [0,127];
The parameter value ascending order of described passage parameter T is resequenced to V, obtains M '=[T ' (i), V ' is (i)],
Wherein, M ' is the channel instruction after resequencing; T ' (i) for the rearrangement after the passage parameter; V ' (i) for the rearrangement after the dynamics parameter.
3. the method for claim 1 is characterized in that, extracts the spatial domain single order of the dynamics parameter after sorting and the higher order statistical moment characteristics of second order difference, comprising:
The preceding Fourth-order moment that the spatial domain single order of dynamics parameter V ' after the described rearrangement and the difference of second order are D ' and D ", extract described D ' and D " is extracted spatial domain 8 dimensional features altogether,
Wherein,
4. method as claimed in claim 3, it is characterized in that the spatial domain single order of the dynamics parameter V ' after the described rearrangement and the difference of second order are D ' and D " the corresponding respectively spatial domain single order of the dynamics component after the described rearrangement and average, variance, the degree of bias and the kurtosis of second order difference variable of characterizing of preceding Fourth-order moment.
5. the method for claim 1, it is characterized in that, extract the higher order statistical feature of histogram feature function, comprise: the preceding Fourth-order moment of extracting MIDI audio frequency dynamics histogram of component, single order and each fundamental function of second order difference histogram, extract histogram feature domain of function 12 dimension statistical natures altogether, wherein
The fundamental function H[k of MIDI audio frequency dynamics histogram of component] be:
Histogrammic fundamental function H[k] n rank square M
(hcf) nFor:
6. a MIDI audio hidden information analysis device is characterized in that, comprises reordering module, characteristic extracting module and training detection module,
Described reordering module is used for reading the instruction of MIDI voice-grade channel and hides dynamics parameter and the passage parameter that carries confidential information, described dynamics parameter is arranged according to the ascending order of described passage parameter resequenced;
Described characteristic extracting module is used to extract spatial domain single order and the higher order statistical moment characteristics of second order difference and the higher order statistical feature of histogram feature function of the dynamics parameter after the described ordering, forms 20 dimensional feature vectors;
Described training detection module is used for described 20 dimensional feature vectors input support vector machine is trained the formation sorter, and whether input MIDI audio frequency detects described MIDI audio frequency and carry close in described sorter.
7. device as claimed in claim 6 is characterized in that, described MIDI voice-grade channel instruction is M (i)=[T (i), V (i)] i ∈ 1,2 ... N, wherein, N is the length of MIDI voice-grade channel instruction; T is the passage parameter of MIDI instruction, T ∈ [1,16]; V is the dynamics parameter of MIDI instruction, V ∈ [0,127];
Described reordering module is resequenced the parameter value ascending order of described passage parameter T to V, obtain M '=[T ' (i), V ' is (i)],
Wherein, M ' is the channel instruction after resequencing; T ' (i) for the rearrangement after the passage parameter; V ' (i) for the rearrangement after the dynamics parameter.
8. device as claimed in claim 6 is characterized in that, described characteristic extracting module is extracted the spatial domain single order of the dynamics parameter after the ordering and the higher order statistical moment characteristics of second order difference, comprising:
The preceding Fourth-order moment that the spatial domain single order of dynamics parameter V ' after the described rearrangement and the difference of second order are D ' and D ", extract described D ' and D " is extracted spatial domain 8 dimensional features altogether,
9. device as claimed in claim 8, it is characterized in that the spatial domain single order of the dynamics parameter V ' after the described rearrangement and the difference of second order are D ' and D " the corresponding respectively spatial domain single order of the dynamics component after the described rearrangement and average, variance, the degree of bias and the kurtosis of second order difference variable of characterizing of preceding Fourth-order moment.
10. device as claimed in claim 6, it is characterized in that, described characteristic extracting module is extracted the higher order statistical feature of histogram feature function, comprise: the preceding Fourth-order moment of extracting MIDI audio frequency dynamics histogram of component, single order and each fundamental function of second order difference histogram, extract histogram feature domain of function 12 dimension statistical natures altogether, wherein
The fundamental function H[k of MIDI audio frequency dynamics histogram of component] be:
Histogrammic fundamental function H[k] n rank square M
(hcf) nFor:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101884196A CN101847411B (en) | 2010-05-28 | 2010-05-28 | MIDI (Musical Instrument Digital Interface) audio hidden information analysis method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101884196A CN101847411B (en) | 2010-05-28 | 2010-05-28 | MIDI (Musical Instrument Digital Interface) audio hidden information analysis method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101847411A true CN101847411A (en) | 2010-09-29 |
CN101847411B CN101847411B (en) | 2012-05-23 |
Family
ID=42772006
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2010101884196A Expired - Fee Related CN101847411B (en) | 2010-05-28 | 2010-05-28 | MIDI (Musical Instrument Digital Interface) audio hidden information analysis method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101847411B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102063907A (en) * | 2010-10-12 | 2011-05-18 | 武汉大学 | Steganalysis method for audio spread-spectrum steganography |
CN102254561A (en) * | 2011-08-18 | 2011-11-23 | 武汉大学 | Spatial cue based audio information steganalysis method |
CN104282310A (en) * | 2014-09-26 | 2015-01-14 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
CN104681031A (en) * | 2014-12-08 | 2015-06-03 | 华侨大学 | Bit combination-based stego-detection method for least significant bits (LSB) of low-bit-rate speeches |
CN109784006A (en) * | 2019-01-04 | 2019-05-21 | 平安科技(深圳)有限公司 | Watermark insertion and extracting method and terminal device |
CN115440234A (en) * | 2022-11-08 | 2022-12-06 | 合肥工业大学 | Audio steganography method and system based on MIDI and countermeasure generation network |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7402744B1 (en) * | 2004-06-14 | 2008-07-22 | The United States Of America As Represented By The Secretary Of The Air Force | MIDI file steganography |
-
2010
- 2010-05-28 CN CN2010101884196A patent/CN101847411B/en not_active Expired - Fee Related
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7402744B1 (en) * | 2004-06-14 | 2008-07-22 | The United States Of America As Represented By The Secretary Of The Air Force | MIDI file steganography |
Non-Patent Citations (4)
Title |
---|
《IEEE SIGNAL PROCESSING LETTERS》 20050630 Andrew D. Ker Steganalysis of LSB Matching in Grayscale Images 第12卷, 第6期 2 * |
《Proceedings of the Fourth International Conference on Machine Learning and Cybernetics》 20050821 ALEXANDER ADLI et al THREE STEGANOGRAPHY ALGORITHMS FOR MIDI FILES , 2 * |
《信息安全与通信保密》 20070228 淦新富,郭立 基于主元统计的音频隐写分析 , 第2期 2 * |
《软件学报》 20040131 张涛,平西建 基于差分直方图实现LSB信息伪装的可靠检测 第15卷, 第1期 2 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102063907A (en) * | 2010-10-12 | 2011-05-18 | 武汉大学 | Steganalysis method for audio spread-spectrum steganography |
CN102063907B (en) * | 2010-10-12 | 2012-07-04 | 武汉大学 | Steganalysis method for audio spread-spectrum steganography |
CN102254561A (en) * | 2011-08-18 | 2011-11-23 | 武汉大学 | Spatial cue based audio information steganalysis method |
CN102254561B (en) * | 2011-08-18 | 2012-06-27 | 武汉大学 | Spatial cue based audio information steganalysis method |
CN104282310A (en) * | 2014-09-26 | 2015-01-14 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
CN104282310B (en) * | 2014-09-26 | 2017-05-03 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
CN104681031A (en) * | 2014-12-08 | 2015-06-03 | 华侨大学 | Bit combination-based stego-detection method for least significant bits (LSB) of low-bit-rate speeches |
CN104681031B (en) * | 2014-12-08 | 2018-03-09 | 华侨大学 | A kind of low-bit-rate speech coding least significant bit Stego-detection method based on bit combination |
CN109784006A (en) * | 2019-01-04 | 2019-05-21 | 平安科技(深圳)有限公司 | Watermark insertion and extracting method and terminal device |
CN115440234A (en) * | 2022-11-08 | 2022-12-06 | 合肥工业大学 | Audio steganography method and system based on MIDI and countermeasure generation network |
Also Published As
Publication number | Publication date |
---|---|
CN101847411B (en) | 2012-05-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101847411B (en) | MIDI (Musical Instrument Digital Interface) audio hidden information analysis method and device | |
US11828769B2 (en) | Multiple meter detection and processing using motion data | |
Lou et al. | TFDroid: Android malware detection by topics and sensitive data flows using machine learning techniques | |
CN110503961A (en) | Audio identification methods, device, storage medium and electronic equipment | |
CN104509065A (en) | Using the ability to speak as a human interactive proof | |
CN109495513A (en) | Unsupervised encryption malicious traffic stream detection method, device, equipment and medium | |
CN103730128A (en) | Audio clip authentication method based on frequency spectrum SIFT feature descriptor | |
CN108766464A (en) | Digital audio based on mains frequency fluctuation super vector distorts automatic testing method | |
CN103761965A (en) | Method for classifying musical instrument signals | |
CN105283916B (en) | Electronic watermark embedded device, electronic watermark embedding method and computer readable recording medium | |
CN110189767B (en) | Recording mobile equipment detection method based on dual-channel audio | |
CN104392161B (en) | The device authentication based on audio physical fingerprint under the conditions of displacement | |
CN103236265A (en) | MP3Stegz steganography detecting method | |
CN105340003B (en) | Speech synthesis dictionary creating apparatus and speech synthesis dictionary creating method | |
CN111933154A (en) | Method and device for identifying counterfeit voice and computer readable storage medium | |
Chuchra et al. | A deep learning approach for splicing detection in digital audios | |
CN106782612A (en) | A kind of reverse sonic boom detection method and its device | |
CN112735381B (en) | Model updating method and device | |
Liu et al. | Pseudo-zernike moments-based audio content authentication algorithm robust against feature-analysed substitution attack | |
CN113421554B (en) | Voice keyword detection model processing method and device and computer equipment | |
CN108766465A (en) | A kind of digital audio based on ENF universal background models distorts blind checking method | |
CN113259348A (en) | Heterogeneous data processing method and device, computer equipment and storage medium | |
Guo et al. | Towards the universal defense for query-based audio adversarial attacks on speech recognition system | |
CN105138696B (en) | A kind of music method for pushing and device | |
Wróbel et al. | Identification of Humans Using Hand Clapping Sounds |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120523 Termination date: 20150528 |
|
EXPY | Termination of patent right or utility model |