CN105608661A - Shear wave domain audio watermark embedding and detecting method based on robust QR decomposition - Google Patents

Abstract

The invention discloses a method for embedding and detecting audio watermarks in the shearlet domain based on robust QR decomposition, which utilizes the good frequency domain localization characteristics, multi-directionality, translation invariance and near-optimal Sparse representation capability, and after decomposition, each sub-band audio has the same length as the original audio, plus the good numerical stability of QR decomposition, the R matrix obtained by the decomposition is an upper triangular stable matrix, and the watermark is embedded on the top of the R matrix In the triangle, combined with synchronization code technology. Experimental results show that the invention can not only resist conventional signal processing, but also effectively resist desynchronization attacks, and has good imperceptibility and robustness. In addition, it also has the advantages of simple design and easy implementation.

Description

Audio watermark embedding and detection method in shear wave domain based on robust QR decomposition

technical field

The invention belongs to the technical field of digital audio watermarking, in particular to a shear wave domain audio watermark embedding and detection method based on robust QR decomposition, which can resist conventional signal processing and effectively resist desynchronization attacks.

Background technique

Digital watermarking (Digital Watermarking) is a new technology that can protect copyright and authenticate the source and integrity in an open network environment. Audio watermarking technology is to hide a mark (watermark) with a specific meaning in a digital audio product by means of data embedding to prove the creator's ownership of his work and serve as a basis for identification and prosecution of illegal infringement. The detection and analysis of the watermark ensures the integrity and reliability of digital information, thus becoming an effective means of intellectual property protection and digital multimedia anti-counterfeiting. Digital audio watermarking is bound to be attacked in various forms, and the attacks on watermarking systems are divided into conventional attacks and desynchronization attacks. Conventional attacks include adding noise, resampling, filtering, MP3 compression, etc., which will reduce the energy of the watermark signal; desynchronization attacks include jitter attacks, amplitude changes, tone changes, clipping, time scale scaling, etc. These attack operations will cause The original watermark and the extracted watermark have synchronization errors, so the exact watermark cannot be detected.

The current digital audio watermarking methods can be roughly divided into two categories: time domain and transform domain. The time domain is to embed the watermark by modifying the time domain sampling value of the host audio signal. This type of method has poor anti-attack ability and the watermark capacity is small. The transform domain is to embed the watermark by modifying the transform domain coefficients of the host audio signal. Commonly used transforms include discrete Fourier transform (DFT), discrete cosine transform (DCT), discrete wavelet transform (DWT) and lifting wavelet transform (LWT), etc. , this type of method not only has strong anti-attack ability, large watermark capacity, but also is compatible with data compression standards. Among them, the shearlet transform in the transform domain was constructed by Guo et al. in 2007. This method can not only produce the optimal approximation, but also has a discretization implementation. Currently, it is used in image denoising, image fusion and target edge detection. Some research results have been obtained in other fields. However, most existing audio watermarking algorithms can only resist conventional signal processing, but cannot effectively resist desynchronization attacks.

Contents of the invention

The present invention aims to solve the above-mentioned technical problems existing in the prior art, and provides a shear wave domain audio watermark embedding and detection method based on robust QR decomposition that can resist conventional signal processing and effectively resist desynchronization attacks .

The technical solution of the present invention is: a kind of shear wave domain audio watermark embedding method based on robust QR decomposition, it is characterized in that carrying out according to the following steps:

Step 11: Watermark encoding: Binary watermark image Scramble encryption into a secure watermark matrix ,in

Then perform dimension reduction processing on it, that is, convert the two-dimensional watermark image into a one-dimensional binary watermark sequence:

Finally, according to the following formula, the Perform BPSK modulation mapping to get a one-dimensional {-1,1} anti-phase sequence :

;

Step 12: Segment the original audio carrier according to the size of the digital watermark and the synchronization code, and then use the time-domain audio sample statistics to embed the synchronization code, as follows:

Step 121: Put By sync code length divided into segment, each segment contain audio samples, that is

;

Step 122: Calculate the average value of

;

Step 123: Use the quantization method to embed the synchronization code, that is, for each segment , modifying its mean , to embed a one-bit synchronization code, the modification strategy is:

in, is the audio sample value before modification, is the modified audio sample value, and has

= ,

in, For the modulo operation, is the quantization step size;

Step 13: Embedding of watermark signal:

Step 131: Map the second half of the audio into the form of a two-dimensional matrix, and perform non-subsampled shearlet transform to obtain the low-frequency subband;

Step 132: Divide the low-frequency sub-band into blocks according to the length of the watermark, and perform QR decomposition on each small-block matrix to obtain an upper triangular matrix R;

Step 133: Select the elements in the first row of the R matrix for watermark embedding and modify the intensity and The size of the watermark information bits decided, the formula is as follows:

in is the quantization step size;

Step 134: by and Calculate the quantitative result and , the formula is as follows:

in , floor(.) is rounded down, ceil(.) is rounded up, is the quantization step size, and C is the element coefficient of the first row of the R matrix;

Step 135: Embedding the watermark in the coefficients of the first row of the R matrix, the formula is as follows:

where C is the coefficient for selecting the embedded watermark, is the modified coefficient, abs(.) takes the absolute value;

Step 136: Use Replace the element coefficient C in the first row of the R matrix, perform inverse QR decomposition to obtain watermarked sub-blocks, and then reconstruct all watermarked sub-blocks to obtain watermarked low-frequency subbands, and then use non-subsampled inverse shearlet Transformation, the low-frequency sub-band coefficients containing the watermark are combined with the high-frequency sub-band coefficients to obtain a two-dimensional form of audio containing a watermark, and finally restored to a one-dimensional form of audio, and then reconstructed into the original audio data segment to obtain a watermarked audio digital audio;

Step 14: Repeat steps 2 to 3 to embed synchronization code and watermark information on other audio data segments.

A shear wave domain audio watermark detection method based on robust QR decomposition corresponding to the above embedding method, characterized in that it follows the steps

Step 21: use the frame synchronization code bit-by-bit comparison method in the communication field to find the synchronization code, and use the audio data segment between two adjacent synchronization codes as the candidate audio data segment to extract the watermark according to the detected synchronization code;

Step 22: Extraction of the watermark signal, specifically as follows:

Step 221: Map the obtained watermarked audio data segment into a two-dimensional form and perform non-downsampling shearlet transform to obtain low-frequency sub-bands, divide the low-frequency sub-bands into blocks, and then perform QR decomposition on each sub-block to obtain an upper triangle matrix ;

Step 222: Select The elements in the first row of the matrix are used for watermark extraction, and the formula is as follows:

in , ceil(.) is rounded up, for the matrix The coefficients of the first row of elements of , is the quantization step size, is a modulo operation function;

Step 23: Repeat steps 21 to 22 to extract the watermark from the audio watermark data segment between each adjacent synchronization code, and use the "majority principle" to find the optimal watermark information:

;

Step 24: Watermark decoding:

Extract optimal watermark information for all Demodulate BPSK modulation to get binary sequence :

;

right Perform dimension-up processing to obtain the watermark image matrix , again Perform inverse scrambling and decryption to obtain the extracted binary image watermark :

.

The present invention utilizes the good frequency-domain localization characteristics, multi-directionality, translation invariance and near-optimal sparse representation ability of the non-subsampled shearlet transform, and the decomposed sub-band audio has the same length as the original audio, adding The upper QR decomposition has good numerical stability. The R matrix obtained by the decomposition is an upper triangular stable matrix. The watermark is embedded in the upper triangle of the R matrix, and combined with the synchronization code technology, a robust QR decomposition based Audio watermark embedding and detection method in shear wave domain. Experimental results show that the invention can not only resist conventional signal processing, but also effectively resist desynchronization attacks, and has good imperceptibility and robustness. In addition, it also has the advantages of simple design and easy implementation.

Description of drawings

FIG. 1 is a waveform diagram of original audio in an embodiment of the present invention.

Fig. 2 is a waveform diagram of audio after embedding a watermark according to an embodiment of the present invention.

detailed description

A shear wave domain audio watermark embedding method based on robust QR decomposition, characterized in that it proceeds in accordance with the following steps:

Step 11: Watermark encoding: Binary watermark image Scramble encryption into a secure watermark matrix ,in

In order to facilitate the embedding of the two-dimensional watermark image into the one-dimensional audio signal, and then perform dimension reduction processing on it, that is, convert the two-dimensional watermark image into a one-dimensional binary watermark sequence:

Finally, according to the following formula, the Perform BPSK modulation mapping to get a one-dimensional {-1,1} anti-phase sequence :

;

Step 12: Segment the original audio carrier according to the size of the digital watermark and the synchronization code, and then use the time-domain audio sample statistics to embed the synchronization code, as follows:

Step 121: Put By sync code length divided into segment, each segment contain audio samples, that is

;

Step 122: Calculate the average value of

;

Step 123: Use the quantization method to embed the synchronization code, that is, for each segment , modifying its mean , to embed a one-bit synchronization code, the modification strategy is:

in, is the audio sample value before modification, is the modified audio sample value, and has

= ,

in, For the modulo operation, is the quantization step size;

Step 13: Embedding of watermark signal:

Step 131: Map the second half of the audio into the form of a two-dimensional matrix, and perform non-subsampled shearlet transform to obtain the low-frequency subband;

Step 132: Divide the low-frequency sub-band into blocks according to the length of the watermark, and perform QR decomposition on each small-block matrix to obtain an upper triangular matrix R;

Step 133: Select the elements in the first row of the R matrix for watermark embedding and modify the intensity and The size of the watermark information bits decided, the formula is as follows:

in is the quantization step size;

Step 134: by and Calculate the quantitative result and , the formula is as follows:

in , floor(.) is rounded down, ceil(.) is rounded up, is the quantization step size, and C is the element coefficient of the first row of the R matrix;

Step 135: Embedding the watermark in the coefficients of the first row of the R matrix, the formula is as follows:

where C is the coefficient for selecting the embedded watermark, is the modified coefficient, abs(.) takes the absolute value;

Step 136: Use Replace the element coefficient C in the first row of the R matrix, perform inverse QR decomposition to obtain watermarked sub-blocks, and then reconstruct all watermarked sub-blocks to obtain watermarked low-frequency subbands, and then use non-subsampled inverse shearlet Transformation, the low-frequency sub-band coefficients containing the watermark are combined with the high-frequency sub-band coefficients to obtain a two-dimensional form of audio containing a watermark, and finally restored to a one-dimensional form of audio, and then reconstructed into the original audio data segment to obtain a watermarked audio digital audio;

Step 14: Repeat steps 2 to 3 to embed synchronization code and watermark information on other audio data segments.

A shear wave domain audio watermark detection method based on robust QR decomposition corresponding to the above embedding method, characterized in that it follows the steps

Step 21: use the frame synchronization code bit-by-bit comparison method in the communication field to find the synchronization code, and use the audio data segment between two adjacent synchronization codes as the candidate audio data segment to extract the watermark according to the detected synchronization code;

Step 22: Extraction of the watermark signal, specifically as follows:

Step 221: Map the obtained watermarked audio data segment into a two-dimensional form and perform non-downsampling shearlet transform to obtain low-frequency sub-bands, divide the low-frequency sub-bands into blocks, and then perform QR decomposition on each sub-block to obtain an upper triangle matrix ;

Step 222: Select The elements in the first row of the matrix are used for watermark extraction, and the formula is as follows:

in , ceil(.) is rounded up, for the matrix The coefficients of the first row of elements of , is the quantization step size, is a modulo operation function;

Step 23: Repeat steps 21 to 22 to extract the watermark from the audio watermark data segment between each adjacent synchronization code, and use the "majority principle" to find the optimal watermark information:

;

Step 24: Watermark decoding:

Extract optimal watermark information for all Demodulate BPSK modulation to get binary sequence :

;

right Perform dimension-up processing to obtain the watermark image matrix , again Perform inverse scrambling and decryption to obtain the extracted binary image watermark :

.

The waveform diagram of the original audio in the embodiment of the present invention is shown in Figure 1, and the waveform diagram of the audio after the watermark is embedded in the embodiment of the present invention is shown in Figure 2. It can be seen that the audio after the watermark is embedded in the embodiment of the present invention is very different from the original audio , that is, the present invention has good perceptual transparency.

Table 1 shows the comparison between the embodiment of the present invention and the prior art digital watermarking resistance to conventional signal processing.

Table 2 shows the comparison between the embodiment of the present invention and the digital watermark in the prior art on the resistance to desynchronization attacks.

Table 1 The resistance of digital watermarking to conventional signal processing

Table 2. Resistance ability of digital watermark to desynchronization attack

The result shows that the embodiment of the present invention can not only resist the conventional signal processing, but also effectively resist the desynchronization attack, and has good robustness.

Claims (2)

1. A shear wave domain audio watermark embedding method based on robust QR decomposition is characterized by comprising the following steps:

step 11: watermark coding: two-value watermark imageScrambling and encrypting into a secure watermark matrixWherein

And then, performing dimension reduction treatment on the watermark image, namely converting the two-dimensional watermark image into a one-dimensional binary watermark sequence:

finally, according to the following formula, toBPSK modulation mapping is carried out to obtain a one-dimensional reversed phase sequence of { -1,1 { (R) }：

；

Step 12: segmenting an original audio carrier according to the size of the digital watermark and the size of the synchronous code, and then embedding the synchronous code by utilizing the statistical characteristic of the time domain audio sample, wherein the method specifically comprises the following steps:

step 121: will be provided withAccording to the length of the synchronization codeIs divided intoSegments of eachComprisesAn audio sample, i.e.

；

Step 122: computingAverage value of (i), i.e.

；

Step 123: the synchronisation code is embedded by quantization, i.e. for each segmentModifying the mean value thereofWith the embedded one-bit sync code, the modification strategy is:

wherein,in order to modify the pre-audio sample values,is a modified audio sample value and has

=,

Wherein,in order to perform the operation of taking the modulus,is a quantization step size;

step 13: embedding of watermark signal:

step 131: mapping the rear half part of the audio frequency into a two-dimensional matrix form, and performing non-subsampled shear wave transformation to obtain a low-frequency sub-band;

step 132: partitioning the low-frequency sub-band according to the length of the watermark, and carrying out QR decomposition on each small block matrix to obtain an upper triangular matrix R;

step 133: selecting the first row of elements of the R matrix for watermark embedding and modifying the strengthAndis determined by the size of the watermark information bitThe decision, the formula is as follows:

whereinIs a quantization step size;

step 134: byAndcalculating a quantization resultAndthe formula is as follows:

whereinFloor (. degree.) is rounded down, ceil (. degree.) is rounded up,c is the first row element coefficient of the R matrix for the quantization step;

step 135: the watermark is embedded in the first row of coefficients of the R matrix, as follows:

where C is the coefficient selected to embed the watermark,taking the absolute value of abs (.) for the modified coefficients;

step 136: by usingReplacing the first row element coefficient C of the R matrix, carrying out inverse QR decomposition to obtain sub-blocks containing watermarks, then reconstructing all the sub-blocks containing the watermarks back to obtain low-frequency sub-bands containing the watermarks, combining the low-frequency sub-band coefficients containing the watermarks with the high-frequency sub-band coefficients by utilizing non-subsampled inverse shear wave transformation to obtain two-dimensional form audio containing the watermarks, finally reducing the two-dimensional form audio into one-dimensional form audio, and reconstructing the one-dimensional form audio into the original audio data section to obtain digital audio containing the watermarks;

step 14: and repeating the step 2 to the step 3 to embed the synchronous codes and the watermark information into other audio data segments.

2. A detection method corresponding to the method for embedding a shear-wave-domain audio watermark based on robust QR decomposition of claim 1, characterized by the following steps:

step 21: searching a synchronous code by adopting a frame synchronous code bit-by-bit comparison mode in the communication field, and taking an audio data segment between two adjacent synchronous codes as a candidate audio data segment of the watermark to be extracted according to the detected synchronous code;

step 22: the extraction of the watermark signal is as follows:

step 221: mapping the obtained audio data segment containing the watermark into a two-dimensional form, performing non-subsampled shear wave transformation to obtain low-frequency sub-bands, partitioning the low-frequency sub-bands, and performing QR decomposition on each sub-block to obtain an upper triangular matrix；

Step 222: selectingThe first row of elements of the matrix is subjected to watermark extraction, and the formula is as follows:

whereinCeil (.) is rounding up,is a matrixThe coefficients of the first row of elements of (c),in order to quantize the step size,is a modulo arithmetic function;

step 23: repeating the steps 21 to 22, extracting the watermark of the audio watermark data section between each two adjacent synchronous codes, and solving the optimal watermark information by using a majority principle:

；

step 24: watermark decoding:

extracting optimal watermark information for allDemodulating BPSK modulation to obtain binary sequence：

；

To pairPerforming dimension increasing processing to obtain a watermark image matrixThen go right againInverse scrambling and decryption are carried out to obtain the extracted binary image watermark：

。