CN101266794A - Multiple watermark inlay and exaction method based on echo hiding - Google Patents

Abstract

The invention relates to a method of embedding and extracting multiple watermarks based on echo hiding in an audio processing method of the Information Security Technology field, comprising embedding the watermarks and extracting the watermarks two parts: in embedding the watermarks, firstly processing to divide frames for an audio carrier signal, wherein, the number of the frames depend on length of longest watermark, then introducing delayed echo in each frame audio according to elements on corresponding position of each watermark; in extracting the watermarks, firstly making use of a method of power cepstrum autocorrelation to detect the echo in each frame of the audio signal which waits for being detected, then extracting the watermarks according to delay of the echo. The invention is provided with a very strong robustness for various signal processing operations and hostile attacks probably met in complex network environment. Moreover, the original audio carrier signal need not participate to the process of extracting the watermarks, so the invention is provided with better application value.

Description

The multi-watermarking of hiding based on echo embeds and extracting method

Technical field

The present invention relates to a kind of audio-frequency processing method of field of information security technology, be specifically related to a kind of multi-watermarking of hiding based on echo and embed and extracting method.

Background technology

In today of multimedia and network technology develop rapidly and widespread use, the protection of multi-medium datas such as image, audio frequency, video is become the problem that presses for solution.The protection of multi-medium data is divided into two aspects: the one, and copyright protection, the 2nd, content integrity (authenticity) protection.Traditional encryption method protecting data has significant limitation, and emerging Information Hiding Techniques has then remedied these defectives.Information hiding is used in and is called digital watermarking on the multimedia data protection; be that relevant with content of multimedia or more incoherent beacon information are directly embedded in the middle of the content of multimedia; but do not influence the use value of raw content, and be not easy to be perceiveed by people's consciousness system or notice.Be hidden in information in the content of multimedia by these, can reach and confirm creator of content, buyer, whether perhaps identify true complete purpose.In addition, if the information of relevant multi-medium data (as the lyrics of brief description, booking method, author, song etc.) directly can be hidden in the middle of the content of multimedia, also can bring convenience to multimedia sale and use, this class watermark is called the mark watermark.

Digital watermarking system can provide the detection of duplicate to follow the trail of in the copyright electronic transaction.Before copyright was transferred the possession of, the creation of works person can embed the creation sign watermark of oneself; After works were transferred the possession of, media distribution person carried out watermark processing to the works that are stored in the server, added publisher's mark; When selling copy of works, media distribution person also will add the sale mark therein.Protection in order to make digital watermark technology provide a plurality of links for the distribution and the sale of copyright in the network environment of complexity is necessary to design the multiple digital watermark method that can embed a plurality of watermark signals in same copyright.

Find through literature search prior art, Mintzer, F. wait at " Proc.of the 1999 IEEEInternational Conference on Acoustics, Speech, and Signal Processing " (world's acoustics in 1999, voice and signal Processing proceeding) (1999,2067 to 2069 pages of the fourth phases) deliver on " If one watermark is good; are more better? " if (watermark is good, whether is the multi-watermarking meeting better?), point out that the multi-watermarking signal is generating, embedding grammar, all than single watermark signal method more requirement is arranged on the extracting method, the transparency of watermark signal and the equalization problem of robustness are more complicated.Wherein distinct issues are to guarantee that watermark signal that the back embeds is less to the influence that the watermark signal of previous embedding causes the most.In the most watermarking project, the embedded location of watermark determined by method, if in the audio carrier that has embedded watermark embed watermark once more, the watermark that embeds is for the first time destroyed.

Summary of the invention

The objective of the invention is at above-mentioned the deficiencies in the prior art; a kind of multi-watermarking method of hiding based on echo is provided; making it carry out the branch frame to the audio carrier signal handles; and introduce the echo of corresponding delay according to the element on each watermark correspondence position for it with regard to each frame; and watermark extracting need not the participation of original audio carrier signal, utilizes the creator of multi-watermarking technical protection works, the owner of copyright and the rights and interests of publisher in distribution of DAB works and sales process of medium.

The present invention is achieved by the following technical solutions, the present invention includes two steps:

1, watermark embedded part: at first the audio carrier signal is carried out the branch frame and handle, the number of frame depends on the length of the longest watermark, introduces the echo of time-delay according to the element on each watermark correspondence position for it then in each frame audio frequency;

2, watermark extracting part: the echo of at first utilizing the autocorrelative method of power cepstrum to treat in each frame of acoustic frequency signal is detected, and extracts watermark according to the delay of echo then, and watermark extracting need not the participation of original audio carrier signal.

Described watermark embeds, and is specific as follows:

The first step, suppose that the original audio carrier is s={s (i) | i=1 ..., L _s, wherein: L _sBe the contained sampling number of original audio carrier; The watermark sequence that is about to embed is respectively w ₁={ w ₁(i) | i=1 ..., L ₁, w ₂={ w ₂(i) | i=1 ..., L ₂And w ₃={ w ₃(i) | i=1 ..., L ₃, L wherein ₁, L ₂And L ₃The length of representing first, second and third watermark sequence respectively; Be d with time-delay respectively ₁, d ₀Represent binary one, " 0 " and " room " in the watermark sequence with the echo of d sampling interval.

In second step, the isometric frame with original audio carrier signal s is divided into non-overlapping copies is designated as F={f _i| i=1 ..., L}, wherein F represents the set of all frames, f _iRepresent the i frame, L represents the length of every frame, wherein L=max{L ₁, L ₂, L ₃;

The 3rd step is with i watermark sequence w of each watermark ₁(i), w ₂(i) and w ₃(i) be embedded into i frame f _iIn, specific as follows:

1. at first with w ₁(i) be embedded into f _iIn: if w ₁Item (i)=0 give f _iAdd time-delay and be d ₀The echo of individual sampling interval obtains f _i ¹If w ₁Item (i)=1 give f _iAdd time-delay and be d ₁The echo of individual sampling interval obtains f _i ¹

2. with f _i ¹Embed w as original audio ₂(i): suppose that the echo time-delay of introducing is τ in 1. ₁If, w ₂Item (i)=0 give f _i ¹Add time-delay and be d ₀+ τ ₁The echo of individual sampling interval obtains f _i ²If w ₂Item (i)=1 give f _i ¹Add time-delay and be d ₁+ τ ₁The echo of individual sampling interval obtains f _i ²

3. with f _i ²Embed w as original audio ₃(i): suppose that the echo time-delay of introducing is τ in 2. ₂If, w ₃Item (i)=0 give f _i ²Add time-delay and be d ₀+ τ ₂The echo of individual sampling interval obtains f _i ³If w ₂Item (i)=1 give f _i ²Add time-delay and be d ₁+ τ ₂The echo of individual sampling interval obtains f _i ³

In the 4th step, all are embedded with the audio frame f of watermark _i ³, i=1,2 ..., L combines, and forms to be embedded with watermark audio signal s '.

Described watermark extracting, specific as follows:

The first step, with sound signal s ' to be detected=s ' (i) | i=1 ..., L _{S '}Be divided into the isometric L frame of non-overlapping copies, be designated as $\hat{F}=\left\{{\hat{f}}_i\right|i=1,\·\·\·,L\};$

Second goes on foot, and calculates the power cepstrum coefficient of autocorrelation of each frame, is designated as C={c _i(n) | n=1 ..., L _{S '}/ L; I=1 ..., L}, wherein c _iThe power cepstrum coefficient of autocorrelation of representing the i frame;

The 3rd step: according to the power cepstrum coefficient of autocorrelation c of i frame _iSize is extracted i watermark sequence w in each watermark ₁(i), w ₂(i) and w ₃(i), specific as follows:

At first, compare c _i(d ₀), c _i(d ₁) and c _i(d) three's size, and wherein the pairing horizontal ordinate of value of maximum is designated as q _i(1), then have:

Wherein: d ₁, d ₀Represent the number of sampling interval with d, the echo of its sampling interval is represented binary one, " 0 " and " room " in the watermark sequence respectively;

Then, compare c _i(q _i(1)+d ₀), c _i(q _i(1)+d ₁) and c _i(q _i(1)+d) three's value, wherein the pairing horizontal ordinate of maximal value is designated as q _i(2), then have

For,

At last, compare c _i(q _i(2)+d ₀), c _i(q _i(2)+d ₁) and c _i(q _i(2)+d) three's value, wherein the pairing horizontal ordinate of maximal value is designated as q _i(3), then have

For,

The 4th step: all are extracted

I=..., L is combined into

All

I=1 ..., L is combined into

All

I=1 ..., L is combined into

Compared with prior art, the present invention includes following beneficial effect: the present invention can be simultaneously be embedded in the same audio carrier a plurality of watermark sequences are non-interfering, and the blind extraction of the independence that has realized watermark, may the Audio Signal Processing operation have very strong robustness to various simultaneously, therefore be well suited in the digital product electronic trading system, finishing many copyright authentications function.

Description of drawings

Fig. 1 is the power cepstrum coefficient of autocorrelation figure of watermark audio signal frame

Among the figure: (a) without any signal processing operations, (b) through Mp3 compression/de-compression, (c) through resampling, (d) through re-quantization, (e) through low-pass filtering, (f) superimposed noise, (g) amplitude amplify, (h) amplitude is dwindled, (i) time-delay, (j) superpose echo.

Embodiment

Below in conjunction with accompanying drawing embodiments of the invention are elaborated: present embodiment is being to implement under the prerequisite with the technical solution of the present invention, provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.

Present embodiment with a segment length be 90s, sampling rate be 44.1kHz, quantified precision be the classical music of 16bit/sample as the original audio carrier signal, the chaotic binary sequences that is 100 bits with three length has been carried out emulation as the multi-watermarking that is embedded into to method.Represent " 1 " and " 1 " in the watermark sequence respectively with 40 sampling interval and 60 sampling interval, the echo amplitude is 0.4.

Suppose w ₁={ w ₁(i) | i=1 ..., 100}, w ₂={ w ₂(i) | i=1 ..., 100} and w ₃={ w ₃(i) | i=1 ..., 100} is respectively three watermark sequences that are about to embedding, and s is the original audio carrier, is d with time-delay respectively ₁=40, d ₀=60 and the echo of d=100 sampling interval represent binary one, " 0 " and " room " in the watermark sequence.

The watermark embedded part comprises the steps:

The first step: at first original audio carrier signal s is divided into 100 isometric frames of non-overlapping copies, is designated as F={f _i| i=1 ..., 100};

Second step: with i watermark sequence of each watermark, i.e. w ₁(i), w ₂(i) and w ₃(i), be embedded into i frame f _iIn.With first frame is example, supposes w ₁(1)=0, w ₂(1)=1, w ₃(1)=and null, then embedding grammar is: at first, give f ₁Add time-delay and be d ₀The echo of individual sampling interval obtains f ₁ ¹Then, with f ₁ ¹As original audio,, it is d for adding time-delay ₀+ d ₁The echo of individual sampled value obtains f ₁ ²At last, with f ₁ ²As original signal,, it is d for adding time-delay ₀+ d ₁The echo of+d sampling interval is to f ₁ ³, the rest may be inferred, with w ₁(i), w ₂(i) and w ₃(i) be embedded into f _i, i=1 ..., obtain f among the L _i ³

The 3rd step: all are embedded with the audio frame f of watermark _i ³, i=1,2 ..., L combines, and forms to be embedded with watermark audio signal s '.

Watermark extracting partly comprises the steps:

The first step: with sound signal s ' to be detected=s ' (i) | i=1 ..., L _{S '}Be divided into the 100 isometric frames of non-overlapping copies, be designated as $\hat{F}=\left\{{\hat{f}}_i\right|i=1,\·\·\·,100\};$

Second step: calculate the power cepstrum coefficient of autocorrelation of each frame, be designated as C={c _i(n) | n=1 ..., L _{S '}/ 100; I=1 ..., 100}, wherein c _iThe power cepstrum coefficient of autocorrelation of representing the i frame;

The 3rd step: in order to extract i watermark sequence in each watermark, i.e. w ₁(i), w ₂(i) and w ₃(i), to the power cepstrum coefficient of autocorrelation c of i frame _iCompare:

At first, compare c _i(d ₀), c _i(d ₁)) and c _i(d) three's size, and wherein the pairing horizontal ordinate of value of maximum is designated as q _i(1), then has

For,

Then, compare c again _i(q _i(1)+d ₀), c _i(q _i(1)+d ₁) and c _i(q _i(1)+d) three's value, wherein the pairing horizontal ordinate of maximal value is designated as q _i(2), then have

For,

At last, compare c _i(q _i(2)+d ₀), c _i(q _i(2)+d ₁) and c _i(q _i(2)+d) three's value, wherein the pairing horizontal ordinate of maximal value is designated as q _i(3), then have

For,

The 4th step: all are extracted

I=1 ..., 100 are combined into

All

I=1 ..., 100 are combined into

All

I=1 ..., 100 are combined into

Experimental verification

Be embedded with watermark " 1,0, NULL " audio signal frame power cepstrum coefficient of autocorrelation as shown in Figure 1, three obvious spikes among the figure are represented the reference position of echo, work as a=0.1,0.2,, the variation tendency of the normalization amplitude of 1 o'clock 3 spike as shown in the figure.

(1) is subjected to the influence of echo amplitude

Be subjected to echo amplitude effect in order to study the present embodiment method, the amplitude of echo is set at 0.1～1.0 respectively, carry out the embedding and the extraction of multi-watermarking then, experimental result (the bit error rate that comprises signal to noise ratio snr and three watermarks, be BER_1, BER_2 and BER_3) as shown in table 1.

The performance of table 1 under different echo amplitudes relatively

(2) resist the robustness that Audio Signal Processing is operated

In order to verify that the present embodiment method resists the strong robustness of normal Audio Signal Processing operation, a frame that takes out in the sound signal that is embedded with watermark is observed.The power cepstrum coefficient of autocorrelation of this frame as described in Figure 1, three band asterisks are peak value among the figure, be P1, P2 and P3, starting point corresponding to three echoes that embed, figure (a) is the audio frequency without any signal processing operations, then, the sound signal that is embedded with watermark has been carried out the Mp3 compression/decompression processes (as figure (b)) of 32kbps respectively, 44.1-22.05-44.1kHz resampling (as figure (c)), the re-quantization of 16-8-16bits/sample (as figure (d)), by frequency is the low-pass filtering (as figure (e)) of 5kHz, intensity is the stack (as figure (f)) of the white noise of 10dB, the amplitude of 3dB is amplified (as figure (g)), the amplitude of 3dB is dwindled (as figure (h)), 100ms, 10% time-delay (as figure (i)) and 100ms, the processing of 40% echo stack (as figure (j)) is taken out same frame again and is calculated its power cepstrum coefficient of autocorrelation.

Bit error rate and signal to noise ratio (S/N ratio) that three watermarks are obtained under various normal Audio Signal Processing operations are as shown in table 2.Experimental result proves that fully the present embodiment method has good robustness to normal Audio Signal Processing operation.

This multi-watermarking method of table 2 is resisted the robustness of normal Audio Signal Processing operation

(3) resist the robustness of malicious attack

In addition, also attack software " Stirmark for Audio v02 " (a kind of professional audio watermark performance testing tool commonly used) with the audio frequency watermark of specialty the sound signal of embed watermark is attacked, extract watermark then again, experimental result is as shown in table 3.As can be seen, the present embodiment method has shown stronger robustness to most of malicious attack.

This multi-watermarking method of table 3 is resisted the robustness that " Stirmark for Audio v0.2 " attacks

Claims (5)

1, a kind of multi-watermarking of hiding based on echo embeds and extracting method, it is characterized in that, comprises two steps:

1. watermark embeds, and at first the audio carrier signal is carried out the branch frame and handles, and the number of frame depends on the length of the longest watermark, introduces the echo of time-delay according to the element on each watermark correspondence position for it then in each frame audio frequency;

2. watermark extracting, at first utilize the autocorrelative method of power cepstrum treat acoustic frequently the echo in each frame of signal detect, extract watermark according to the delay of echo then.

2, the multi-watermarking of hiding based on echo according to claim 1 embeds and extracting method, it is characterized in that described watermark embeds, and is specific as follows:

The first step, setting the original audio carrier is s={s (i) | i=1 ..., L _s, wherein: L _sBe the contained sampling number of original audio carrier; The watermark sequence that is about to embed is respectively w ₁={ w ₁(i) | i=1 ..., L ₁, w ₂={ w ₂(i) | i=1 ..., L ₂And w ₃={ w ₃(i) | i=1 ..., L ₃, L wherein ₁, L ₂And L ₃The length of representing first, second and third watermark sequence respectively; Be d with time-delay respectively ₁, d ₀Represent binary one, " 0 " and " room " in the watermark sequence with the echo of d sampling interval;

In second step, the isometric frame with original audio carrier signal s is divided into non-overlapping copies is designated as F={f _i| i=1 ..., L}, wherein F represents the set of all frames, f _iRepresent the i frame, L represents the length of every frame, wherein L=max{L ₁, L ₂, L ₃;

The 3rd step is with i watermark sequence w ₁(i), w ₂(i) and w ₃(i) be embedded into i frame f _iIn;

The 4th step combined all audio frames that are embedded with watermark, formed and was embedded with watermark audio signal.

3, the multi-watermarking of hiding based on echo according to claim 2 embeds and extracting method, it is characterized in that, and is described with i watermark sequence w ₁(i), w ₂(i) and w ₃(i) be embedded into i frame f _iIn, specific as follows:

1. at first with w ₁(i) be embedded into f _iIn: if w ₁Item (i)=0 give f _iAdd time-delay and be d ₀The echo of individual sampling interval obtains f _i ¹If w ₁Item (i)=1 give f _iAdd time-delay and be d ₁The echo of individual sampling interval obtains f _i ¹

2. with f _i ¹Embed w as original audio ₂(i): if the echo of introducing in 1. time-delay is τ ₁If, w ₂Item (i)=0 give f _i ¹Add time-delay and be d ₀+ τ ₁The echo of individual sampling interval obtains f _i ²If w ₂Item (i)=1 give f _i ¹Add time-delay and be d ₁+ τ _iThe echo of individual sampling interval obtains f _i ²

3. with f _i ²Embed w as original audio ₃(i): if the echo of introducing in 2. time-delay is τ ₂If, w ₃Item (i)=0 give f _i ²Add time-delay and be d ₀+ τ ₂The echo of individual sampling interval obtains f _i ³If w ₂Item (i)=1 give f _i ²Add time-delay and be d ₁+ τ ₂The echo of individual sampling interval obtains f _i ³

4, the multi-watermarking of hiding based on echo according to claim 1 embeds and extracting method, it is characterized in that, and described watermark extracting, specific as follows:

The first step is divided into the isometric L frame of non-overlapping copies with sound signal to be detected, is designated as $\hat{F}=\left\{{\hat{f}}_i\right|i=1,\·\·\·,L\};$

Second goes on foot, and calculates the power cepstrum coefficient of autocorrelation of each frame, is designated as C={c _i(n) | n=1 ..., L _{S '}/ L; I=1 ..., L}, wherein c _iThe power cepstrum coefficient of autocorrelation of representing the i frame;

The 3rd step: according to the power cepstrum coefficient of autocorrelation c of i frame _iSize extract i watermark sequence w in each watermark ₁(i), w ₂(i) and w ₃(i);

The 4th step: all are extracted

I=1 ..., L is combined into

All

I=1 ..., L is combined into

All

I=1 ..., L is combined into

5, the multi-watermarking of hiding based on echo according to claim 4 embeds and extracting method, it is characterized in that described power cepstrum coefficient of autocorrelation c according to the i frame _iSize extract i watermark sequence w in each watermark ₁(i), w ₂(i) and w ₃(i), specific as follows:

At first, compare c _i(d ₀), c _i(d ₁) and c _i(d) three's size, and wherein the pairing horizontal ordinate of value of maximum is designated as q _i(1), then have:

Wherein: d ₁, d ₀Represent the number of sampling interval with d, the echo of its sampling interval is represented binary one, " 0 " and " room " in the watermark sequence respectively;

Then, compare c _i(q _i(1)+d ₀), c _i(q _i(1)+d ₁) and c _i(q _i(1)+d) three's value, wherein the pairing horizontal ordinate of maximal value is designated as q _i(2), then have For,

At last, compare c _i(q _i(2)+d ₀), c _i(q _i(2)+d ₁) and c _i(q _i(2)+d) three's value, wherein the pairing horizontal ordinate of maximal value is designated as q _i(3), then have

For,

Images