An adaptive robust audio blind watermarking method based on SVD

Abstract

The present invention discloses an adaptive robust audio blind watermarking method based on SVD, belonging to the field of audio processing technology, comprising the following steps: encrypting a watermark image based on a logistic chaotic map to obtain an encrypted watermark sequence; embedding the encrypted watermark sequence into an original audio signal to obtain an audio signal bearing the watermark image; and decrypting the watermark sequence in the audio signal bearing the watermark image to obtain the watermark image. The present invention uses a logistic chaotic map to encrypt the watermark image and, through adaptive embedding parameters, embeds the encrypted watermark image into the original audio signal. Furthermore, the present invention can blindly extract the watermark image without requiring the original audio signal or embedding parameters, maximizing audio perceptual quality while ensuring a certain embedding capacity and resistance to conventional signal processing operations. This method solves the problems of low perceptual quality, low security, insufficient robustness, and inability to blindly extract watermark images in watermark audio methods.

Description

Technical Field

[0001] The present invention belongs to the technical field of audio processing, and in particular relates to an adaptive robust audio blind watermarking method based on SVD. Background Art

[0002] Digital watermarking embeds identifying information (such as publisher IDs, signatures, and logos) into multimedia objects in a subtle and unnoticeable manner, creating a watermarked object. Digital watermarking can be applied to various types of multimedia objects, such as images, video, and audio. Compared to image watermarking, embedding watermarks in digital audio signals is more technically challenging, primarily due to the higher sensitivity of the human auditory system compared to the visual system.

[0003] SVD technology is widely used in audio watermarking because its singular values ​​remain relatively stable even after attacks. For example, based on the QIM method, watermark data is embedded into the singular values ​​of each block in the wavelet domain through singular value decomposition through adaptive quantization, ensuring the imperceptibility of the watermark algorithm. However, this method is not robust enough to echo addition and resampling attacks, and the watermark capacity is low. For example, blind watermarking schemes based on singular value decomposition (SVD) using entropy and log-polar transform (LPT) have high embedding capacity but are not robust enough to resampling and MP3 compression attacks. Existing SVD-based algorithms only embed watermark bits into a single singular value or into the ratio of two singular values.

[0004] In summary, the shortcomings of the existing technical solutions are:

[0005] (1) Existing watermark embedding schemes are often spread spectrum schemes or quantization schemes, but spread spectrum schemes are susceptible to host signal interference, and quantization schemes are vulnerable to amplitude scaling attacks;

[0006] (2) The existing technical solutions have low payload and low watermark capacity, which cannot effectively ensure the effect of watermark information in audio data;

[0007] (3) The embedding parameters of existing technical solutions are fixed and single, and lack consideration of the characteristics of the audio itself, which exacerbates the contradiction between robustness and imperceptibility. Summary of the Invention

[0008] In response to the above-mentioned deficiencies in the prior art, the present invention provides an adaptive robust audio blind watermarking method based on SVD, which divides the original audio signal into frames, performs discrete wavelet (DWT) transform on each frame, divides the obtained DWT coefficients into two segments using a subsampling operation, then performs singular value transform (SVD), calculates the mean of the singular values ​​of the two segments, and then embeds the watermark bits through a differential embedding method. The proposed adaptive method generates embedding parameters of different sizes according to the characteristics of the original signal of each frame to reduce the degradation of perceptual quality. The present invention does not require the original signal and embedding parameters in the watermark extraction process to achieve blind detection, thereby solving the problems of low perceptual quality, low security, insufficient robustness and inability to blindly extract watermarked images in watermark audio methods.

[0009] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0010] The present invention provides an adaptive robust audio blind watermarking method based on SVD, comprising the following steps:

[0011] S1, encrypt the watermark image based on Logistic chaotic mapping to obtain the encrypted watermark sequence;

[0012] S2, embedding the encrypted watermark sequence into the original audio signal to obtain an audio signal with a watermark image;

[0013] S3. Decrypt the watermark sequence in the audio signal with the watermark image to obtain the watermark image.

[0014] The beneficial effects of the present invention are as follows: an adaptive robust audio blind watermarking method based on SVD is provided by the present invention, which improves the security of the watermarking method by encrypting the watermark image using Logistic chaotic mapping, and embeds the encrypted watermark image into the original audio signal through adaptive embedding parameters, thereby improving the perceptual quality, and can blindly extract the watermark image without the need for the original audio signal and embedding parameters, thereby achieving better maximization of perceptual quality while ensuring a certain embedding capacity and resisting conventional signal processing operations.

[0015] Furthermore, the S1 includes the following steps:

[0016] S11, constructing Logistic chaotic map;

[0017] The calculation expression of the Logistic chaotic map is as follows:

[0018]

[0019] in, Represents the first part of the Logistic chaotic map outputs, Represents the first secret key, represents the second secret key, is an integer greater than 1, Represents the first part of the Logistic chaotic map outputs, represents the number of rows of the matrix, Indicates the number of columns of the matrix;

[0020] S12, constructing encrypted binary sequence based on Logistic chaotic map;

[0021] The calculation expression of the encrypted binary sequence is as follows:

[0022]

[0023] in, represents an encrypted binary sequence, Indicates the third secret key;

[0024] S13, converting the watermark image into a one-dimensional sequence of images;

[0025] The calculation expression of the one-dimensional sequence of the image is as follows:

[0026]

[0027] in, represents a one-dimensional sequence of images, Represents the first i Bit;

[0028] S14, encrypting the one-dimensional image sequence using the encrypted binary sequence to obtain an encrypted watermark sequence;

[0029] The calculation expression of the encrypted watermark sequence is as follows:

[0030]

[0031] in, Indicates the encrypted watermark sequence i encrypted watermark bits, Represents the exclusive OR operation.

[0032] The beneficial effects of adopting the above further solution are as follows: the present invention adopts Logistic chaotic mapping to encrypt the watermark image, and uses the first key, the second key and the third key to ensure the security of the watermark image encryption.

[0033] Furthermore, the S2 includes the following steps:

[0034] S21, dividing the original audio signal into frames to obtain a plurality of frames of audio signal and the length of each frame after the framing;

[0035] The calculation expression of each frame length after framing is as follows:

[0036]

[0037] in, Indicates the length of each frame after the original audio signal is framed. Indicates the length of the original audio signal, Indicates the total number of frames. Indicates rounding operation;

[0038] S22, performing a three-level DWT transform on each frame of audio signal to obtain an approximate coefficient, a first detail coefficient, a second detail coefficient, and a third detail coefficient corresponding to each frame;

[0039] S23, selecting the approximate coefficient corresponding to each frame for watermark embedding, and obtaining the approximate coefficient vector of each frame;

[0040] S24, decomposing the approximate coefficient vector of each frame by subsampling to obtain a first subvector and a second subvector of each frame;

[0041] The calculation expressions of the first sub-vector and the second sub-vector of each frame are as follows:

[0042]

[0043] in, Indicates the The first subvector of the frame, Indicates the The second subvector of the frame, Indicates the The odd-numbered elements of the frame's approximation coefficient vector, Indicates the The even-numbered elements of the frame's approximation coefficient vector, Represents the first bit element, where , b Indicates the number of wavelet transform decomposition layers;

[0044] S25, performing SVD on the first sub-vector and the second sub-vector of each frame respectively to obtain the singular value of the first sub-vector and the singular value of the second sub-vector of each frame;

[0045] The calculation expressions of the singular values ​​of the frame vectors are as follows:

[0046]

[0047]

[0048] in, Indicates the The first subvector singular value of the frame, Indicates the The second subvector singular values ​​of the frame, Indicates the Frame No. The diagonal matrix of subvectors, Indicates the Frame No. The subvector matrix of subvectors, Indicates the Frame No. The first unitary matrix of sub-vectors, Indicates the Frame No. The transpose of the second unitary matrix of the sub-vectors;

[0049] S26. Differentially embed the encrypted watermark sequence into the first subvector singular value and the second subvector singular value of each frame based on the embedding parameter to obtain modified first singular value and second singular value of each frame;

[0050] The calculation expressions of the modified first singular value and the second singular value in each frame are as follows:

[0051]

[0052] in, Indicates the The modified first singular value in the frame, Indicates the The modified second singular value in the frame, Indicates the The element in the encrypted watermark sequence corresponding to the frame;

[0053] S27. Obtain an audio signal with a watermark image based on the modified first singular value and second singular value of each frame.

[0054] The beneficial effects of adopting the above-mentioned further scheme are as follows: the present invention divides the original audio signal into frames, performs a three-level DWT transform on each frame, and uses the approximate coefficients obtained by the DWT transform to embed the watermark, thereby ensuring the ability to resist conventional audio signal processing operations. Secondly, after decomposing and sub-sampling the approximate coefficient vector, the singular values ​​are modified using adaptive embedding parameters, taking into account the characteristics of the original audio signal and improving the perceptual quality.

[0055] Furthermore, the embedding coefficients in S26 are obtained by an adaptive quantization method;

[0056] The adaptive quantization method comprises the following steps:

[0057] A1. Construct an initial signal-to-noise ratio based on the original audio signal and the watermarked audio signal.

[0058] The calculation expression of the initial signal-to-noise ratio is as follows:

[0059]

[0060] in, represents the signal-to-noise ratio, Indicates the The square of the original audio signal of the frame, Indicates the frame original audio signal, Indicates the The audio signal after the frame is embedded with the watermark;

[0061] A2, performing a three-level DWT transformation on each frame of audio signal in the signal-to-noise ratio to obtain an approximate component signal-to-noise ratio;

[0062] The calculation expression of the approximate component signal-to-noise ratio is as follows:

[0063]

[0064] in, represents the approximate component signal-to-noise ratio, Indicates the The square of the approximate components of the original audio signal of the frame, Indicates the The approximate components of the original audio signal of the frame, Indicates the The approximate components of the audio signal after the frame is watermarked;

[0065] A3, decomposing the approximate components of the original audio signal of each frame and the approximate components of the audio signal after embedding the watermark by subsampling, and performing SVD on the results of the subsampling decomposition to obtain a singular value signal-to-noise ratio;

[0066] The calculation expression of the singular value signal-to-noise ratio is as follows:

[0067]

[0068] in, represents the singular value signal-to-noise ratio, Indicates the The square of the singular value of the subvector of the original audio signal of the frame, Indicates the The subvector singular values ​​of the original audio signal of the frame, Indicates the The subvector singular values ​​of the audio signal after the frame is embedded with the watermark;

[0069] A4. Obtain embedding coefficients based on the singular value signal-to-noise ratio;

[0070] The calculation expression of the embedding coefficient is as follows:

[0071]

[0072] in, Indicates the The square of the singular value of the first subvector of the frame, Indicates the The squares of the singular values ​​of the second subvector of the frame.

[0073] The beneficial effect of adopting the above further scheme is: in the present invention, through the adaptive quantization method, while taking into account the characteristics of the element audio signal, the signal-to-noise ratio value and the characteristics of each frame audio signal are used to obtain quantization values ​​of different sizes, and the quantization step size of each frame signal is adaptively generated according to the characteristics of each frame signal to maximize the perceptual quality.

[0074] Furthermore, the S27 includes the following sub-steps:

[0075] S271, using the modified first singular value and second singular value of each frame to replace the first singular value and second singular value in the diagonal matrix, to obtain the modified first diagonal matrix of each frame and the second diagonal matrix ;

[0076] S272, performing inverse SVD on the first diagonal matrix and the second diagonal matrix to obtain modified sub-vectors in each frame;

[0077] The calculation expression of the modified sub-vector of each frame is as follows:

[0078]

[0079] in, Indicates the Frame No. The modified sub-vector of sub-vectors, Indicates the Frame No. The modified diagonal matrix of sub-vectors;

[0080] S273, concatenating the modified sub-vectors in each frame to obtain a modified approximate coefficient vector for each frame;

[0081] The calculation expression of the modified approximate coefficient vector of each frame is as follows:

[0082]

[0083] in, Indicates the The odd-numbered elements of the approximation coefficient vector after frame modification, Indicates the The even-numbered elements of the approximation coefficient vector after frame modification, Indicates the The first subvector after the frame modification, Indicates the The second sub-vector after frame modification;

[0084] S274. Based on the modified approximate coefficient vectors of each frame, perform an inverse three-level DWT transformation on the modified approximate coefficient, the first detail coefficient, the second detail coefficient, and the third detail coefficient corresponding to each frame to obtain each modified frame.

[0085] S275 , sequentially concatenate the modified frames to obtain an audio signal with a watermark image.

[0086] The beneficial effect of adopting the above further scheme is: after the present invention modifies the singular values ​​of each frame based on the embedding parameters, the original singular values ​​are replaced to obtain the modified first diagonal matrix and second diagonal matrix in each frame, and then inverse SVD and inverse 3-level DWT changes are performed in sequence, and the modified frames are connected in series, thereby realizing the construction of an audio signal with a watermarked image that has high perceptual quality, strong security, and strong resistance to conventional audio signal processing operations.

[0087] Furthermore, the S3 includes the following steps:

[0088] S31, obtaining the first subvector singular value and the second subvector singular value of each watermark frame in the audio signal with the watermark image;

[0089] S32, calculating and obtaining the difference between the singular value of the first sub-vector and the singular value of the second sub-vector in each watermark frame;

[0090] The calculation expression of the difference between the singular value of the first sub-vector and the singular value of the second sub-vector in each watermark frame is as follows:

[0091]

[0092] in, Indicates the The difference between the singular value of the first subvector and the singular value of the second subvector in the watermark frame;

[0093] S33, extracting the encrypted watermark bits from the audio signal with the watermark image according to the watermark decryption model;

[0094] The calculation expression of the watermark decoding model is as follows:

[0095]

[0096] in, Indicates the The watermark encryption bits extracted from the watermark frame, Indicates other situations;

[0097] S34, obtaining a watermark encrypted binary sequence based on the encrypted watermark bits in the audio signal with the watermark image, the Logistic chaotic map, the first secret key, the second secret key, and the third secret key;

[0098] S35, decrypting the watermark encrypted binary sequence through the watermark decryption model to obtain a decrypted watermark sequence;

[0099] The calculation expression of the decrypted watermark sequence is as follows:

[0100]

[0101] in, represents the decrypted watermark sequence, Indicates the first Bit, Represents a watermark encrypted binary sequence;

[0102] S36, convert the decrypted watermark sequence into a size of The watermark image is obtained by using the matrix of .

[0103] The beneficial effects of adopting the above further scheme are: extracting the encrypted watermark bits of each watermark frame based on the difference between the singular values ​​of the sub-vectors of each frame of the watermark frame, and generating a watermark encrypted binary sequence based on the first secret key, the second secret key and the third secret key, and performing a one-to-one XOR operation on the watermark encrypted binary sequence and the encrypted watermark bits in the encrypted watermark bits of each watermark frame to decrypt the watermark sequence, and finally converting the decrypted binary watermark sequence into a matrix, thereby realizing blind decryption of the watermark and extraction of the watermark image.

[0104] Furthermore, the S31 includes the following steps:

[0105] S311, dividing the audio signal with the watermark image into frames to obtain a plurality of watermarked audio signals and the length of each watermarked frame after framing;

[0106] The calculation expression of each watermark frame length is as follows:

[0107]

[0108] in, Indicates the length of each watermark frame, Indicates the length of the audio signal with the watermark image;

[0109] S312, performing a three-level DWT transform on each frame signal to obtain an approximate coefficient, a first detail coefficient, a second detail coefficient, and a third detail coefficient corresponding to each watermark frame;

[0110] S313, obtaining an approximate coefficient vector of each watermark frame based on the corresponding approximate coefficient of each watermark frame;

[0111] S314, decomposing the approximate coefficient vector of each watermark frame by subsampling to obtain a first subvector and a second subvector of each watermark frame;

[0112] The calculation expressions of the first subvector and the second subvector of each watermark frame are as follows:

[0113]

[0114] in, Indicates the The first subvector of the watermark frame, Indicates the The second subvector of the watermark frame, Indicates the The odd-numbered elements of the approximate coefficient vector of the watermark frame, Indicates the The even-numbered elements of the approximate coefficient vector of the watermark frame, Represents the first bit element, where ;

[0115] S315, performing SVD on the first subvector and the second subvector of each watermark frame respectively to obtain the singular value of the first subvector and the singular value of the second subvector of each watermark frame;

[0116] The calculation expressions of the first sub-vector singular value and the second sub-vector singular value of each watermark frame are as follows:

[0117]

[0118]

[0119] in, Indicates the The first subvector singular value of the watermark frame, Indicates the The second subvector singular value of the watermark frame, Indicates the The watermark frame The diagonal matrix of subvectors, Indicates the The watermark frame A subvector matrix of subvectors.

[0120] The beneficial effect of adopting the above further scheme is: using the same method as the audio embedding process, without using the original audio signal and embedding parameters, based on framing, three-level DWT transformation, sub-vector decomposition and SVD, the first sub-vector singular value and the second sub-vector singular value of each watermark frame in the audio signal with the watermark image are obtained, providing a basis for blind decryption of the watermark sequence in the audio signal with the watermark image.

[0121] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0122] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0123] Figure 1 The figure is a flowchart of the steps of an SVD-based adaptive robust audio blind watermarking method in an embodiment of the present invention.

[0124] Figure 2 This is a sampling feature graph of the time domain difference between the original audio signal of the audio type Folk and the audio signal with the watermark image in an embodiment of the present invention.

[0125] Figure 3 This is a sampling feature graph of the time domain difference between the original audio signal of French audio type and its audio signal with a watermark image in an embodiment of the present invention. DETAILED DESCRIPTION

[0126] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0127] Singular value decomposition (SVD) is a powerful information extraction tool that can decompose the matrix in a simple way. The SVD transformation of a matrix of size is:

[0128]

[0129] in, For one The unitary matrix, express The diagonal matrix of V express The unitary matrix, T Represents the transpose operation of the matrix;

[0130] Discrete wavelet transform (DWT) is a new spectral analysis tool that discretizes the scale and translation of basic wavelets. It can examine both the frequency domain characteristics of local time domain processes and the time domain characteristics of local frequency domain processes. Therefore, even non-stationary processes can be well transformed and processed.

[0131] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides an adaptive robust audio blind watermarking method based on SVD, comprising the following steps:

[0132] S1, encrypt the watermark image based on Logistic chaotic mapping to obtain the encrypted watermark sequence;

[0133] The S1 comprises the following steps:

[0134] S11, constructing Logistic chaotic map;

[0135] The calculation expression of the Logistic chaotic map is as follows:

[0136]

[0137] in, Represents the first part of the Logistic chaotic map outputs, Represents the first secret key, represents the second secret key, is an integer greater than 1, Represents the first part of the Logistic chaotic map outputs, represents the number of rows of the matrix, Indicates the number of columns of the matrix; in this embodiment, the first key The value is 0.78, the second secret key The value of is 3.5821656;

[0138] S12, constructing encrypted binary sequence based on Logistic chaotic map;

[0139] The calculation expression of the encrypted binary sequence is as follows:

[0140]

[0141] in, represents an encrypted binary sequence, Indicates the third secret key; in this implementation, the third secret key The value of is 0.6; through the first secret key, the second secret key and the third secret key

[0142] S13, converting the watermark image into a one-dimensional sequence of images;

[0143] The calculation expression of the one-dimensional sequence of the image is as follows:

[0144]

[0145] in, represents a one-dimensional sequence of images, Represents the first i Bit;

[0146] S14, encrypting the one-dimensional image sequence using the encrypted binary sequence to obtain an encrypted watermark sequence;

[0147] The calculation expression of the encrypted watermark sequence is as follows:

[0148]

[0149] in, Indicates the encrypted watermark sequence i encrypted watermark bits, Represents the exclusive OR operation.

[0150] S2, embedding the encrypted watermark sequence into the original audio signal to obtain an audio signal with a watermark image;

[0151] The S2 comprises the following steps:

[0152] S21, dividing the original audio signal into frames to obtain a plurality of frames of audio signal and the length of each frame after the framing;

[0153] The calculation expression of each frame length after framing is as follows:

[0154]

[0155] in, Indicates the length of each frame after the original audio signal is framed. Indicates the length of the original audio signal, Indicates the total number of frames. Indicates rounding operation;

[0156] S22, performing a three-level DWT transform on each frame of audio signal to obtain an approximate coefficient, a first detail coefficient, a second detail coefficient, and a third detail coefficient corresponding to each frame;

[0157] S23, selecting the approximate coefficient corresponding to each frame for watermark embedding, and obtaining the approximate coefficient vector of each frame;

[0158] S24, decomposing the approximate coefficient vector of each frame by subsampling to obtain a first subvector and a second subvector of each frame;

[0159] The calculation expressions of the first sub-vector and the second sub-vector of each frame are as follows:

[0160]

[0161] in, Indicates the The first subvector of the frame, Indicates the The second subvector of the frame, Indicates the The odd-numbered elements of the frame's approximation coefficient vector, Indicates the The even-numbered elements of the frame's approximation coefficient vector, Represents the first bit element, where , b Indicates the number of wavelet transform decomposition layers;

[0162] S25, performing SVD on the first sub-vector and the second sub-vector of each frame respectively to obtain the singular value of the first sub-vector and the singular value of the second sub-vector of each frame;

[0163] The calculation expressions of the singular values ​​of the frame vectors are as follows:

[0164]

[0165]

[0166] in, Indicates the The first subvector singular value of the frame, Indicates the The second subvector singular values ​​of the frame, Indicates the Frame No. The diagonal matrix of subvectors, Indicates the Frame No. The subvector matrix of subvectors, Indicates the Frame No. The first unitary matrix of sub-vectors, Indicates the Frame No. The transpose of the second unitary matrix of the sub-vectors;

[0167] S26. Differentially embed the encrypted watermark sequence into the first subvector singular value and the second subvector singular value of each frame based on the embedding parameter to obtain modified first singular value and second singular value of each frame;

[0168] The calculation expressions of the modified first singular value and the second singular value in each frame are as follows:

[0169]

[0170] in, Indicates the The modified first singular value in the frame, Indicates the The modified second singular value in the frame, Indicates the The elements in the encrypted watermark sequence corresponding to the frame; it is worth noting that different embedding coefficients corresponding to each frame are provided by adaptive quantization.

[0171] The embedding coefficients in S26 are obtained by an adaptive quantization method;

[0172] The adaptive quantization method is different from the fixed quantization parameter method. It avoids the situation where the fixed quantization parameter does not take the characteristics of the audio signal into consideration. Instead, it uses the signal-to-noise ratio value and the characteristics of each frame of audio signal to obtain quantization values ​​of different sizes, thereby improving the perceived quality.

[0173] The adaptive quantization method comprises the following steps:

[0174] A1. Construct an initial signal-to-noise ratio based on the original audio signal and the watermarked audio signal.

[0175] The calculation expression of the initial signal-to-noise ratio is as follows:

[0176]

[0177] in, represents the signal-to-noise ratio, Indicates the The square of the original audio signal of the frame, Indicates the frame original audio signal, Indicates the The audio signal after the frame is embedded with the watermark;

[0178] A2, performing a three-level DWT transformation on each frame of audio signal in the signal-to-noise ratio to obtain an approximate component signal-to-noise ratio;

[0179] The calculation expression of the approximate component signal-to-noise ratio is as follows:

[0180]

[0181] in, represents the approximate component signal-to-noise ratio, Indicates the The square of the approximate components of the original audio signal of the frame, Indicates the The approximate components of the original audio signal of the frame, Indicates the The approximate components of the audio signal after the frame is watermarked;

[0182] A3, decomposing the approximate components of the original audio signal of each frame and the approximate components of the audio signal after embedding the watermark by subsampling, and performing SVD on the results of the subsampling decomposition to obtain a singular value signal-to-noise ratio;

[0183] The calculation expression of the singular value signal-to-noise ratio is as follows:

[0184]

[0185] in, represents the singular value signal-to-noise ratio, Indicates the The square of the singular value of the subvector of the original audio signal of the frame, Indicates the The subvector singular values ​​of the original audio signal of the frame, Indicates the The subvector singular values ​​of the audio signal after the frame is embedded with the watermark;

[0186] According to the initial signal-to-noise ratio, approximate component signal-to-noise ratio and singular value signal-to-noise ratio, in order to correctly extract the watermark, it is necessary to ensure that: ;

[0187] That is, substituting the singular value signal-to-noise ratio is equivalent to: ;

[0188] Simplifying, we get: ;

[0189] pass Control the noise of the watermark and adaptively generate the quantization step size of each frame signal according to the characteristics of each watermark frame signal to maximize the perceptual quality. In the embedding process, it is specifically reflected as the embedding coefficient.

[0190] A4. Obtain embedding coefficients based on the singular value signal-to-noise ratio;

[0191] The calculation expression of the embedding coefficient is as follows:

[0192]

[0193] in, Indicates the The square of the singular value of the first subvector of the frame, Indicates the The square of the singular value of the second sub-vector of the frame. In this embodiment, the singular value signal-to-noise ratio The value is 30.

[0194] S27. Obtain an audio signal with a watermark image based on the modified first singular value and second singular value of each frame.

[0195] The S27 includes the following sub-steps:

[0196] S271. Replace the first singular value and the second singular value in the diagonal matrix with the modified first singular value and the second singular value of each frame to obtain the modified first diagonal matrix and the second diagonal matrix of each frame;

[0197] S272, performing inverse SVD on the first diagonal matrix and the second diagonal matrix to obtain modified sub-vectors in each frame;

[0198] The calculation expression of the modified sub-vector of each frame is as follows:

[0199]

[0200] in, Indicates the Frame No. The modified sub-vector of sub-vectors, Indicates the Frame No. The modified diagonal matrix of sub-vectors;

[0201] S273, concatenating the modified sub-vectors in each frame to obtain a modified approximate coefficient vector for each frame;

[0202] The calculation expression of the modified approximate coefficient vector of each frame is as follows:

[0203]

[0204] in, Indicates the The odd-numbered elements of the approximation coefficient vector after frame modification, Indicates the The even-numbered elements of the approximation coefficient vector after frame modification, Indicates the The first subvector after the frame modification, Indicates the The second sub-vector after frame modification;

[0205] S274. Based on the modified approximate coefficient vectors of each frame, perform an inverse three-level DWT transformation on the modified approximate coefficient, the first detail coefficient, the second detail coefficient, and the third detail coefficient corresponding to each frame to obtain each modified frame.

[0206] S275 , sequentially concatenate the modified frames to obtain an audio signal with a watermark image.

[0207] like Figure 2 and Figure 3 As shown, Figure 2 The original audio signal of type Folk, Figure 3 The original audio signal of French is Figure 2 and Figure 3 The time domain representations of different types of original audio signals and their watermarked counterparts are shown, along with the differences between them. The two figures show that the waveforms of the original and watermarked signals are very similar, with minimal differences that are virtually indistinguishable to the human eye. This demonstrates that the watermark embedded in the original audio signal, implemented using this invention, exhibits excellent transparency.

[0208] The evaluation index table for the imperceptibility of audio signals with watermarked images includes several subjective (SDG) and objective (ODG) evaluation indicators, as shown in Table 1:

[0209] Table 1

[0210] SDG ODG describe quality 0 0 Imperceptible superior -1 -1 Perceptible but noisy good -2 -2 There is slight noise generally -3 -3 Noise Difference -4 -4 A lot of noise Very bad

[0211] The encrypted watermark sequence is embedded into the original audio signal using the method of the present invention to obtain the audio signal with the watermark image. SNR , subjective and objective evaluation indicators are used for evaluation, and the evaluation results are shown in Table 2:

[0212] Table 2

[0213]

[0214] Table 2 shows that the signal-to-noise ratio (SNR) values ​​for all types of original audio signals are greater than 20 dB, reaching a maximum of 26.9154 dB, meeting the requirements of the International Federation of the Phonographic Industry (IFPI). Furthermore, the objective evaluation indexes for all types of watermarked audio signals are greater than -1, indicating that the watermarked audio signals obtained using the method of the present invention are highly imperceptible. Furthermore, the subjective evaluation indexes are all close to 0, indicating that it is difficult to distinguish between the original audio and the watermarked audio.

[0215] The audio signal with the watermark image obtained by the present invention still has a large number of NC values ​​close to 1 under the conditions of noise addition, resampling, re-normalization, echo addition, amplitude +10%, amplitude -10%, amplitude +20%, amplitude -20%, MP3 compression 128kbps, MP3 compression 64kbps and shearing 3000. The extracted watermark image can still be recognized in its original shape. The bit error rate of almost all attacks is 0, and the maximum bit error rate is less than 1%. The watermark image embedded in the audio signal implemented by this scheme has good robustness.

[0216] S3. Decrypt the watermark sequence in the audio signal with the watermark image to obtain the watermark image.

[0217] The S3 comprises the following steps:

[0218] S31, obtaining the first subvector singular value and the second subvector singular value of each watermark frame in the audio signal with the watermark image;

[0219] The S31 includes the following steps:

[0220] S311, dividing the audio signal with the watermark image into frames to obtain a plurality of watermarked audio signals and the length of each watermarked frame after framing;

[0221] The calculation expression of each watermark frame length is as follows:

[0222]

[0223] in, Indicates the length of each watermark frame, Indicates the length of the audio signal with the watermark image;

[0224] S312, performing a three-level DWT transform on each frame signal to obtain an approximate coefficient, a first detail coefficient, a second detail coefficient, and a third detail coefficient corresponding to each watermark frame;

[0225] S313, obtaining an approximate coefficient vector of each watermark frame based on the corresponding approximate coefficient of each watermark frame;

[0226] S314, decomposing the approximate coefficient vector of each watermark frame by subsampling to obtain a first subvector and a second subvector of each watermark frame;

[0227] The calculation expressions of the first subvector and the second subvector of each watermark frame are as follows:

[0228]

[0229] in, Indicates the The first subvector of the watermark frame, Indicates the The second subvector of the watermark frame, Indicates the The odd-numbered elements of the approximate coefficient vector of the watermark frame, Indicates the The even-numbered elements of the approximate coefficient vector of the watermark frame, Represents the first bit element, where ;

[0230] S315, performing SVD on the first subvector and the second subvector of each watermark frame respectively to obtain the singular value of the first subvector and the singular value of the second subvector of each watermark frame;

[0231] The calculation expressions of the first sub-vector singular value and the second sub-vector singular value of each watermark frame are as follows:

[0232]

[0233]

[0234] in, Indicates the The first subvector singular value of the watermark frame, Indicates the The second subvector singular value of the watermark frame, Indicates the The watermark frame The diagonal matrix of subvectors, Indicates the The watermark frame A subvector matrix of subvectors.

[0235] S32, calculating and obtaining the difference between the singular value of the first sub-vector and the singular value of the second sub-vector in each watermark frame;

[0236] The calculation expression of the difference between the singular value of the first sub-vector and the singular value of the second sub-vector in each watermark frame is as follows:

[0237]

[0238] in, Indicates the The difference between the singular value of the first subvector and the singular value of the second subvector in the watermark frame;

[0239] S33, extracting the encrypted watermark bits from the audio signal with the watermark image according to the watermark decryption model;

[0240] The calculation expression of the watermark decoding model is as follows:

[0241]

[0242] in, Indicates the The watermark encryption bits extracted from the watermark frame, Indicates other situations;

[0243] S34, obtaining a watermark encrypted binary sequence based on the encrypted watermark bits in the audio signal with the watermark image, the Logistic chaotic map, the first secret key, the second secret key, and the third secret key;

[0244] S35, decrypting the watermark encrypted binary sequence through the watermark decryption model to obtain a decrypted watermark sequence;

[0245] The calculation expression of the decrypted watermark sequence is as follows:

[0246]

[0247] in, represents the decrypted watermark sequence, Indicates the first Bit, Represents a watermark encrypted binary sequence;

[0248] S36, convert the decrypted watermark sequence into a size of The watermark image is obtained by using the matrix of .

[0249] The present invention does not require the original audio signal during the watermark extraction process, that is, the method is blind and does not require embedding parameters.

[0250] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. An adaptive robust audio blind watermarking method based on SVD, characterized in that: The steps include: S1, encrypt the watermark image based on Logistic chaotic mapping to obtain the encrypted watermark sequence; S2, embedding the encrypted watermark sequence into the original audio signal to obtain an audio signal with a watermark image; The S2 comprises the following steps: S21, dividing the original audio signal into frames to obtain a plurality of frames of audio signal and the length of each frame after the framing; The calculation expression of each frame length after framing is as follows: in, Indicates the length of each frame after the original audio signal is framed. Indicates the length of the original audio signal, Indicates the total number of frames. Indicates rounding operation; S22, performing a three-level DWT transform on each frame of audio signal to obtain an approximate coefficient, a first detail coefficient, a second detail coefficient, and a third detail coefficient corresponding to each frame; S23, selecting the approximate coefficient corresponding to each frame for watermark embedding, and obtaining the approximate coefficient vector of each frame; S24, decomposing the approximate coefficient vector of each frame by subsampling to obtain a first subvector and a second subvector of each frame; The calculation expressions of the first sub-vector and the second sub-vector of each frame are as follows: in, Indicates the The first subvector of the frame, Indicates the The second subvector of the frame, Indicates the The odd-numbered elements of the frame's approximation coefficient vector, Indicates the The even-numbered elements of the frame's approximation coefficient vector, Represents the first bit element, where , b represents the number of wavelet transform decomposition layers; S25, performing SVD on the first sub-vector and the second sub-vector of each frame respectively to obtain the singular value of the first sub-vector and the singular value of the second sub-vector of each frame; The calculation expressions of the first sub-vector singular value and the second sub-vector singular value of each frame are as follows: in, Indicates the The first subvector singular value of the frame, Indicates the The second subvector singular value of the frame, Indicates the Frame No. The diagonal matrix of subvectors, Indicates the Frame No. The subvector matrix of subvectors, Indicates the Frame No. The first unitary matrix of sub-vectors, Indicates the Frame No. The transpose of the second unitary matrix of the sub-vectors; S26. Differentially embed the encrypted watermark sequence into the first subvector singular value and the second subvector singular value of each frame based on the embedding parameter to obtain modified first singular value and second singular value of each frame; The calculation expressions of the modified first singular value and second singular value of each frame are as follows: in, Indicates the The modified first singular value in the frame, Indicates the The modified second singular value in the frame, Indicates the The element in the encrypted watermark sequence corresponding to the frame, Represents embedded parameters; S27, obtaining an audio signal with a watermark image based on the modified first singular value and second singular value of each frame; S3. Decrypt the watermark sequence in the audio signal with the watermark image to obtain the watermark image.

2. The SVD-based adaptive robust audio blind watermarking method according to claim 1, characterized in that The S1 comprises the following steps: S11, constructing Logistic chaotic map; The calculation expression of the Logistic chaotic map is as follows: in, The first part of the Logistic chaotic map outputs, Represents the first secret key, represents the second secret key, is an integer greater than 1, The first part of the Logistic chaotic map outputs, represents the number of rows of the matrix, Indicates the number of columns of the matrix; S12, constructing encrypted binary sequence based on Logistic chaotic map; The calculation expression of the encrypted binary sequence is as follows: in, represents an encrypted binary sequence, Indicates the third secret key; S13, converting the watermark image into a one-dimensional sequence of images; The calculation expression of the one-dimensional sequence of the image is as follows: in, represents a one-dimensional sequence of images, Represents the i-th position in the one-dimensional sequence of the image; S14, encrypting the one-dimensional image sequence using the encrypted binary sequence to obtain an encrypted watermark sequence; The calculation expression of the encrypted watermark sequence is as follows: in, represents the i-th encrypted watermark bit in the encrypted watermark sequence, Represents the exclusive OR operation.

3. The SVD-based adaptive robust audio blind watermarking method according to claim 2, characterized in that: The embedding parameters in S26 are obtained by an adaptive quantization method; The adaptive quantization method comprises the following steps: A1. Construct an initial signal-to-noise ratio based on the original audio signal and the watermarked audio signal. The calculation expression of the initial signal-to-noise ratio is as follows: in, represents the signal-to-noise ratio, Indicates the The square of the original audio signal of the frame, Indicates the frame original audio signal, Indicates the The audio signal after the frame is embedded with the watermark; A2, performing a three-level DWT transformation on each frame of audio signal in the signal-to-noise ratio to obtain an approximate component signal-to-noise ratio; The calculation expression of the approximate component signal-to-noise ratio is as follows: in, represents the approximate component signal-to-noise ratio, Indicates the The square of the approximate components of the original audio signal of the frame, Indicates the The approximate components of the original audio signal of the frame, Indicates the The approximate components of the audio signal after the frame is watermarked; A3, decomposing the approximate components of the original audio signal of each frame and the approximate components of the audio signal after embedding the watermark by subsampling, and performing SVD on the results of the subsampling decomposition to obtain a singular value signal-to-noise ratio; The calculation expression of the singular value signal-to-noise ratio is as follows: in, represents the singular value signal-to-noise ratio, Indicates the The square of the singular value of the subvector of the original audio signal of the frame, Indicates the The subvector singular values ​​of the original audio signal of the frame, Indicates the The subvector singular values ​​of the audio signal after the frame is embedded with the watermark; A4. Based on the singular value signal-to-noise ratio, the embedding parameters are obtained; The calculation expression of the embedding parameter is as follows: in, Indicates the The square of the singular value of the first subvector of the frame, Indicates the The squares of the singular values ​​of the second subvector of the frame.

4. The SVD-based adaptive robust audio blind watermarking method according to claim 3, characterized in that: The S27 includes the following sub-steps: S271. Replace the first singular value and the second singular value in the diagonal matrix with the modified first singular value and the second singular value of each frame to obtain the modified first diagonal matrix and the second diagonal matrix of each frame; S272, performing inverse SVD on the first diagonal matrix and the second diagonal matrix to obtain modified sub-vectors in each frame; The calculation expression of the modified sub-vector in each frame is as follows: in, Indicates the Frame No. The modified sub-vector of sub-vectors, Indicates the Frame No. The modified diagonal matrix of sub-vectors; S273, concatenating the modified sub-vectors in each frame to obtain a modified approximate coefficient vector for each frame; The calculation expression of the modified approximate coefficient vector of each frame is as follows: in, Indicates the The odd-numbered elements of the approximation coefficient vector after frame modification, Indicates the The even-numbered elements of the approximation coefficient vector after frame modification, Indicates the The first subvector after the frame modification, Indicates the The second sub-vector after frame modification; S274. Based on the modified approximate coefficient vectors of each frame, perform an inverse three-level DWT transformation on the modified approximate coefficient, the first detail coefficient, the second detail coefficient, and the third detail coefficient corresponding to each frame to obtain each modified frame. S275 , sequentially concatenate the modified frames to obtain an audio signal with a watermark image.

5. The SVD-based adaptive robust audio blind watermarking method according to claim 4, characterized in that: The S3 comprises the following steps: S31, obtaining the first subvector singular value and the second subvector singular value of each watermark frame in the audio signal with the watermark image; S32, calculating and obtaining the difference between the singular value of the first sub-vector and the singular value of the second sub-vector in each watermark frame; The calculation expression of the difference between the singular value of the first sub-vector and the singular value of the second sub-vector in each watermark frame is as follows: in, Indicates the The difference between the singular value of the first subvector and the singular value of the second subvector in the watermark frame; S33, extracting the encrypted watermark bits from the audio signal with the watermark image according to the watermark decryption model; The calculation expression of the watermark decryption model is as follows: in, Indicates the The watermark encryption bits extracted from the watermark frame, Indicates other situations; S34, obtaining a watermark encrypted binary sequence based on the encrypted watermark bits in the audio signal with the watermark image, the Logistic chaotic map, the first secret key, the second secret key, and the third secret key; S35, decrypting the watermark encrypted binary sequence through the watermark decryption model to obtain a decrypted watermark sequence; The calculation expression of the decrypted watermark sequence is as follows: in, represents the decrypted watermark sequence, Indicates the first Bit, Represents a watermark encrypted binary sequence; S36, convert the decrypted watermark sequence into a size of The watermark image is obtained by using the matrix of .

6. The SVD-based adaptive robust audio blind watermarking method according to claim 5, characterized in that: The S31 includes the following steps: S311, dividing the audio signal with the watermark image into frames to obtain a plurality of watermarked audio signals and the length of each watermarked frame after framing; The calculation expression of each watermark frame length is as follows: in, Indicates the length of each watermark frame, Indicates the length of the audio signal with the watermark image; S312, performing a three-level DWT transform on each frame signal to obtain an approximate coefficient, a first detail coefficient, a second detail coefficient, and a third detail coefficient corresponding to each watermark frame; S313, obtaining an approximate coefficient vector of each watermark frame based on the corresponding approximate coefficient of each watermark frame; S314, decomposing the approximate coefficient vector of each watermark frame by subsampling to obtain a first subvector and a second subvector of each watermark frame; The calculation expressions of the first subvector and the second subvector of each watermark frame are as follows: in, Indicates the The first subvector of the watermark frame, Indicates the The second subvector of the watermark frame, Indicates the The odd-numbered elements of the approximate coefficient vector of the watermark frame, Indicates the The even-numbered elements of the approximate coefficient vector of the watermark frame, Represents the first bit element, where ; S315, performing SVD on the first subvector and the second subvector of each watermark frame respectively to obtain the singular value of the first subvector and the singular value of the second subvector of each watermark frame; The calculation expressions of the first sub-vector singular value and the second sub-vector singular value of each watermark frame are as follows: in, Indicates the The first subvector singular value of the watermark frame, Indicates the The second subvector singular value of the watermark frame, Indicates the The watermark frame The diagonal matrix of subvectors, Indicates the The watermark frame A subvector matrix of subvectors.

Images