Wavelet packet domain adaptive quantization digital audio watermark embedding and extracting method based on Fibonacci sequence and particle swarm optimization
By introducing wavelet packet domain adaptive quantization method with Fibonacci sequence and particle swarm optimization in digital audio watermarks, the problem of insufficient adaptability and universality in the prior art is solved, and robustness and transparency are achieved effective balance of sex and resistance to multiple attack methods.
Patent Information
- Application Number
- CN202410985504.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-07-23
AI Technical Summary
Existing digital audio watermarking algorithms lack adaptability, are difficult to effectively weigh robustness and transparency, and are not universal, so they cannot adapt to the performance requirements of different music types.
Adaptive quantization digital audio watermarking method based on Fibonacci sequence and particle swarm optimization is adopted. Through mess encryption and wavelet packet transformation, embedded watermark information is adaptively quantized, and embedded parameters are tuned using particle swarm optimization algorithm. Achieve a balance of adaptability and robustness and transparency to different music types.
Improves the transparency and robustness of watermarks, can maintain a high signal-to-noise ratio under different music types, resists multiple attack methods, and the audio embedded in the watermark is almost indistinguishable from the original audio, significantly improving the adaptability of the algorithm. and universality.
Smart Images

Figure CN119049487A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of digital audio watermarking, and in particular relates to a wavelet packet domain adaptive quantization digital audio watermark embedding and extraction method based on Fibonacci sequence and particle swarm optimization. Background Art
[0002] With the rapid development of the Internet and multimedia technology, digital multimedia information has become an important source of information for people in their daily lives, and along with it come data security issues and copyright issues of electronic publications. As an effective means to solve such problems, digital watermarking technology has received widespread attention from researchers and has developed rapidly. It has been widely used in the fields of copyright protection and content authentication.
[0003] In the field of digital audio watermarking, most of the existing watermarking algorithms have good performance. Since the transparency and robustness of the watermark are a pair of contradictory factors, how to balance the relationship between the two has always been a key issue in the research of digital audio watermarking algorithms. In addition, as the audio signal changes, the parameters involved in the algorithm may need to be adjusted accordingly to adapt to different audio signals. Common watermarking algorithms all involve the concept of embedding strength, and there are corresponding parameters to control the embedding strength to solve the above two problems. For example, the quantization-based digital audio watermarking algorithm adjusts the embedding strength of the watermark by controlling the quantization step size, thereby balancing the transparency and robustness of the watermark and adapting to different audio signals.
[0004] Currently, many algorithms do not have good adaptability and cannot effectively balance the relationship between robustness and transparency, requiring manual adjustment of algorithm parameters. In addition, some existing adaptive algorithms are not universal and are all optimized for a single parameter. For example, for common quantization-based adaptive algorithms, the main goal is to find the optimal quantization step size, and the ability to maintain the same or similar performance for different music types is not significant enough.
[0005] The patent application document with the publication number CN106611601A discloses an audio watermarking method based on DWT-SVD and particle swarm optimization. In the process of audio watermark embedding, the original audio signal is firstly processed into blocks, and each block obtained after the block division is subjected to SVD decomposition, and then a vector is constructed for the maximum value in the matrix obtained after the decomposition, and the obtained vector is subjected to three-level wavelet decomposition, and then the watermark is embedded into the selected embedding position by the quantization modulation method in the approximate component after the wavelet decomposition, and then the peak signal-to-noise ratio and the normalized correlation coefficient are used to evaluate the imperceptibility and robustness of the audio signal after the watermark is embedded, and the particle swarm optimization algorithm is used to find the optimal watermark embedding strength. However, since the wavelet transform resolution adopted by this method is not fine enough, the quantization step value in the quantization modulation method is fixed, and the particle swarm optimization algorithm adopted only performs a single parameter optimization on the quantization step, it has the disadvantages of the same degree of modification for coefficients of different sizes, lack of adaptability, and insufficient compromise effect between the imperceptibility and robustness of the watermark. Summary of the invention
[0006] In view of the defects and problems existing in the prior art, the purpose of the present invention is to provide a wavelet packet domain adaptive quantization digital audio watermark embedding and extraction method based on the Fibonacci sequence and particle swarm optimization. By introducing the Fibonacci sequence, it is possible to perform different degrees of quantization modification on wavelet packet domain coefficients of different sizes, thereby achieving the adaptability of the coefficient modification degree, reducing the impact of watermark information embedding on the original audio signal, thereby improving the transparency of the watermark; by introducing the particle swarm optimization algorithm, it is possible to tune multiple parameters involved in the embedding method of the present invention according to the result of the objective function obtained when the watermark is extracted under attack from the audio signal, so that the watermark can adaptively balance the robustness and transparency of the watermark for different music.
[0007] In order to achieve the above object, the technical solution adopted by the present invention is:
[0008] A wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization comprises the following steps:
[0009] Step 1, watermark information preprocessing: scrambling encryption and dimensionality reduction processing of watermark information;
[0010] Step 2, selecting the watermark embedding position: the watermark embedding position is selected in the low- and medium-frequency region of the audio carrier signal;
[0011] Step 3, embedding watermark information: adaptively embedding the watermark information preprocessed in step 1 into the watermark embedding position in the low-frequency region of the audio carrier signal selected in step 2 to obtain an audio signal containing the watermark;
[0012] Step 4, optimizing embedding parameters: Based on the particle swarm optimization algorithm, multiple embedding parameters involved in the watermarked audio signal in step 3 are optimized to obtain the final watermarked audio signal.
[0013] In step 1, the watermark information is the original binary watermark image information, and the original binary watermark image is M0={m0(i, j), 0≤i<X, 0≤j<Y}, where X and Y are the width and height of the original binary watermark image, respectively, and m0(i, j)∈{0,1} represents the gray value of the pixel point in the i-th row and j-th column of the original binary watermark image; the scrambling encryption and dimensionality reduction processing includes the following steps:
[0014] Step 1.1, perform Arnold scrambling on the original binary watermark image M0, and use the number of iterations and control parameters selected during Arnold scrambling as the key K1. After the Arnold scrambling is completed, the scrambled binary watermark image M1 = {m1(i, j), 0≤i<X, 0≤j<Y} is obtained;
[0015] Step 1.2, perform dimensionality reduction operation on the scrambled binary watermark image M1 obtained in step 1.1 to obtain a one-dimensional binary sequence w0 = {w(i), 0≤i<MN};
[0016] Step 1.3, generate the Logistic chaotic sequence, the generation method is shown in formula 1:
[0017] x i+1 =μx i (1-x i ) (Formula 1)
[0018] Among them, x i Indicates the current state, μ is the control parameter, 0<x i <1, when 3.569946<μ≤4, the Logistic map is in a chaotic state; select x0 and μ that make the Logistic map in a chaotic state as the key K2;
[0019] Step 1.4, based on the Logistic chaotic sequence obtained in step 1.3, use formula 2 to generate a binary chaotic sequence v;
[0020]
[0021] Among them, δ is the decision threshold, 0<δ<1, and v is the generated binary chaotic sequence;
[0022] Step 1.5, perform an XOR operation on the one-dimensional binary sequence w0 obtained in step 1.2 and the binary chaotic sequence v generated in step 1.4 to obtain the final watermark sequence w to be embedded.
[0023] The frequency range of the mid-low frequency region in step 2 is within 500 Hz to 3000 Hz.
[0024] The step 3 specifically includes:
[0025] Step 3.1, normalize and frame the original audio signal, the frame length is N, and each frame of the audio signal is embedded with 1-bit binary watermark information;
[0026] Step 3.2: Perform 6-level discrete wavelet packet transform (DWPT) on each frame of audio signal in step 3.1. According to the frequency interval selected in step 2, select a suitable sub-band to embed the watermark sequence w. Assume that the length of the sub-band coefficient is L.
[0027] Then the relationship between L and N is shown in formula 3;
[0028]
[0029] Step 3.3, select n coefficients on the sub-band obtained in step 3.2 for watermark embedding operation, the value of n is an odd number less than the length of the sub-band coefficient, and n is at least 1; when n>1, watermark information will be embedded redundantly, that is, the same watermark information is embedded in the n coefficient positions;
[0030] Step 3.4, construct the Fibonacci sequence Fib0 = {0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, ...}, divide the Fibonacci sequence Fib0 as a whole by a constant to make the positions of all wavelet packet domain coefficients in the sequence as evenly distributed as possible, and obtain the Fibonacci sequence Fib used in the adaptive quantization watermark embedding process; Assume that there are two adjacent Fibonacci numbers F in the Fibonacci sequence Fib k and F k+1 , and the wavelet packet domain coefficient is greater than F k and less than F k+1 , then the position of the wavelet packet domain coefficient in the Fibonacci sequence Fib is the index k;
[0031] Step 3.5, define the quantization step size S, assuming that the wavelet packet domain coefficients of the current frame are d = {d0, d1, d2, ..., d L-1}, the watermark information embedded in the current frame is w(j)∈{0,1}, and the n coefficients described in step 3.3 in the current frame are modified according to the embedding strategy of formula 4 and formula 5 to adaptively quantize the embedded watermark information, with one of the n coefficients d i Take as an example, and the other (n-1) coefficients are similar. The embedding strategy of formula 4 and formula 5 is:
[0032]
[0033] d i =t i S (Formula 5)
[0034] Among them, t i is the coefficient d i The position in the quantization interval, L fib is the length of the Fibonacci sequence Fib in step 3.4, fib i Yes i The position in Fib, that is, when d i Greater than the element Fib in Fib k and less than Fib k+1 The index k when the modified coefficient is 0≤a≤1 is used to control the intensity of the phenomenon that "coefficients of different sizes change to different degrees according to their positions in Fib"; l and r are the bottom factors, 0≤l≤0.5, 0.5≤r≤1, respectively, which are used to ensure that the modified coefficients of d i In the range [l, r] of the corresponding quantization interval;
[0035] Step 3.6, performing an inverse discrete wavelet packet transform (IDWPT) on each frame of audio after the watermark information is embedded in step 3.5, and restoring the audio frame to a time domain signal;
[0036] Step 3.7, all the audio frames obtained in step 3.6 are stitched together in order, and denormalized to obtain an audio signal containing a watermark.
[0037] In step 4, the optimization of the embedding parameters is achieved by using a particle swarm optimization algorithm, using the signal-to-noise ratio (SNR) as an indicator of transparency, and using the normalized correlation coefficient (NC) as an indicator of robustness, to optimize the quantization step size S, the control factor a, and the bottom factor l and r in step 3.5; the particle speed and position update formulas in the particle swarm optimization algorithm are as follows:
[0038] v i (t+1)=wv i (t)+c1r1(p best (t)-x i (t))+c2r2(g best (t)-x i (t)) (Formula 6)
[0039] x i (t+1)=xi (t)+v i (t+1) (Formula 7)
[0040] In Formula 6 and Formula 7, v i (t) and x i (t) are the velocity and position of each particle i at time t, p best is the optimal position of each particle, i.e., its optimal solution in the search process, g best is the global best position of the entire particle swarm, that is, the optimal solution of the entire particle swarm, w represents the inertia weight, which is used to adjust the search ability for the solution space, c1 and c2 represent learning factors, which are called cognitive parameters and social parameters respectively, r1 and r2 are random numbers uniformly distributed in [0, 1], which can increase the randomness of the search process; the position of each particle can represent a set of parameters in step 3.5, namely, the quantization step size S, the control factor a, and the bottom factor l and r;
[0041] The specific steps include:
[0042] Step 4.1, initialize the population and set the relevant parameters of the particle swarm optimization algorithm;
[0043] Step 4.2, according to the population initialized in step 4.1, perform the particle swarm optimization algorithm iteration process to find the optimal watermark embedding parameters, namely, the quantization step size S, the control factor a, and the floor factors l and r, and embed the watermark into the audio according to the optimal watermark embedding parameters to obtain the final audio signal containing the watermark.
[0044] The step 4.2, during the particle swarm optimization algorithm iteration, includes the following sub-steps:
[0045] Step 4.2.1, update the particle speed and position in each iteration;
[0046] Step 4.2.2, according to the updated particle position obtained in step 4.2.1, update the individual optimal solution and the global optimal solution;
[0047] Step 4.2.3, according to the current particle position, obtain the corresponding quantization step size S, control factor a and floor factors l and r, embed the watermark information in the audio signal, and calculate the SNR value of the audio after the watermark information is embedded under the current parameters;
[0048] Step 4.2.4, attacking the watermarked audio signal after embedding the watermark information obtained in step 4.2.3;
[0049] Step 4.2.5, extracting the watermark from the attacked audio signal obtained in step 4.2.4, and comparing it with the watermark information originally embedded in step 4.2.3, to obtain the NC value between the watermark information extracted under different attacks and the original watermark information;
[0050] Step 4.2.6, construct the objective function based on the SNR value obtained in step 4.2.3 and the NC value obtained in step 4.2.5. Use the objective function to calculate the fitness of each particle and evaluate the population in each iteration. When a satisfactory solution is found or the maximum number of iterations is reached, the optimal solution is considered to be found. The objective function is shown in Formula 8:
[0051]
[0052] Among them, λ1 and λ2 are weight factors, which respectively represent the importance of transparency and robustness to the objective function. The larger the λ1, the more inclined to transparency, and the larger the λ2, the more inclined to robustness. R represents the number of attack types, r i is the weight factor of each attack. When the algorithm considers the same weight for different attacks, r i A constant can be taken;
[0053] Step 4.2.7, taking the optimal solution obtained in step 4.2.6 as the final selected embedding parameter, embedding the watermark information according to step 3 to obtain the final audio signal containing the watermark.
[0054] The extraction method based on the wavelet packet domain adaptive quantization digital audio watermark embedding method based on the Fibonacci sequence and particle swarm optimization includes the following steps:
[0055] Step 5, audio preprocessing: normalize and frame the audio signal to be watermarked, the frame length is N, the same as in step 3.1 of the embedding method, each frame of audio signal contains 1 bit of binary watermark information, and perform 6-level DWPT transformation on each frame of audio signal;
[0056] Step 6, select the watermark extraction position: select the same sub-band as in step 3.2 of the embedding method, and select n coefficients in the sub-band with the same positions as in step 3.3 of the embedding method;
[0057] Step 7, extracting watermark information: performing watermark extraction operation on the n coefficients in the sub-band described in step 6, taking one of the n coefficients d′ i Take it as an example, and the other (n-1) coefficients are extracted in the same way as shown in Formula 9:
[0058]
[0059] Step 8, counting and determining the watermark information in the current frame: after extracting the watermark information in step 7, each frame of audio signal obtains n binary watermark information; counting the n binary watermark information, obtaining the number zc of "0" and the number oc of "1" therein, and using the method shown in formula 10 to determine the watermark information contained in the current audio frame;
[0060]
[0061] Step 9, sequentially splicing the watermark information: sequentially splicing the watermark information contained in all the audio frames obtained in step 8 to obtain the binary watermark sequence w′ finally extracted from the audio signal to be extracted;
[0062] Step 10, decryption and recovery: Use the key K1 obtained in the watermark information preprocessing step 1.1 of the embedding method, the key K2 obtained in step 1.3, and the binary chaotic sequence v obtained in step 1.4 to decrypt and recover the binary watermark sequence w′ extracted in step 9. When decrypting, first use the key K2 to perform Logistic scrambling recovery on the binary watermark sequence w′, and perform an XOR operation on the binary watermark sequence w′ and the binary chaotic sequence v to obtain the decrypted and recovered binary sequence w″; then, the decrypted and recovered binary sequence w″ is dimensionally upgraded to obtain a binary image M′; finally, use the key K1 to perform Arnold scrambling recovery on the binary image M′ to obtain the final decrypted and recovered watermark image M″.
[0063] Compared with the prior art, the present invention has the following beneficial effects:
[0064] (1) The present invention ensures the confidentiality and security of watermark information during transmission and extraction by scrambling and encrypting the watermark information.
[0065] (2) The present invention utilizes the high flexibility and high resolution characteristics of wavelet packet transform, combined with the analysis of psychoacoustic model, to embed the watermark in the low- and medium-frequency region to which the human ear is relatively insensitive, thereby ensuring the transparency of the watermark.
[0066] (3) The present invention introduces the Fibonacci sequence to guide the modification strategy of wavelet packet domain coefficients of different sizes, so that the degree of modification is adapted to the size of the coefficients, reducing the impact of watermark information embedding on the original audio signal, thereby improving the transparency of the algorithm.
[0067] (4) The present invention introduces a particle swarm optimization algorithm to find the optimal watermark embedding parameters and optimize multiple parameters involved in the embedding method. On the one hand, it can adaptively select the appropriate quantization step size for different audio signals without manual intervention. On the other hand, through the mutual coordination and optimization of multiple parameters, it can well balance the relationship between watermark robustness and transparency.
[0068] (5) The watermark of the present invention can resist many attack methods, such as adding Gaussian noise, resampling, re-quantization, low-pass filtering, MP3 compression, etc. The watermark extracted after the attack is clear and highly recognizable. The watermark has good transparency. The audio containing the watermark after embedding the watermark information has almost no difference in auditory perception from the original audio before embedding the watermark information. The signal-to-noise ratio between the two can be maintained at a high level under different music types.
[0069] (6) The algorithm of the present invention can adaptively adjust relevant parameters for different music types, effectively balancing the relationship between watermark robustness and transparency, and the algorithm has good adaptability and universality.
[0070] In summary, the present invention, with the help of scrambling encryption technology, utilizes the advantages of high flexibility and high resolution of wavelet packet transform, takes into account the psychoacoustic model, and introduces the Fibonacci sequence and particle swarm optimization algorithm to well ensure the security, transparency, adaptability and universality of digital audio watermarking. The watermark can effectively resist various attack methods and can adaptively balance the contradictory relationship between robustness and transparency. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Figure 1 It is a flow chart of a wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization provided by an embodiment of the present invention.
[0072] Figure 2 It is a flow chart of an extraction method based on a wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization provided by an embodiment of the present invention.
[0073] Figure 3 It is the original binary watermark image provided by the embodiment of the present invention.
[0074] Figure 4 This is a scrambled binary watermark image provided by an embodiment of the present invention.
[0075] Figure 5 It is a schematic diagram of sub-band frequency intervals after 6-level wavelet packet transformation of an audio signal provided by an embodiment of the present invention.
[0076] Figure 6 It is a schematic diagram of a modification strategy of coefficients in a quantization interval in an embedding method provided in an embodiment of the present invention.
[0077] Figure 7 It is a flow chart of optimizing embedding parameters based on a particle swarm optimization algorithm in an embedding method provided in an embodiment of the present invention.
[0078] Figure 8It is a schematic diagram of the change of the objective function value of the global optimal solution in the particle swarm optimization algorithm provided by an embodiment of the present invention with the iteration process.
[0079] Fig. 9 It is an audio waveform diagram before and after watermark embedding provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0080] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0081] The present invention aims to solve the problems that many digital audio watermark algorithms are not adaptive, cannot well balance the relationship between robustness and transparency, and have poor universality when processing different music types.
[0082] like Figure 1 As shown, a wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization includes the following steps:
[0083] Step 1, watermark information preprocessing: scrambling encryption and dimensionality reduction processing of watermark information;
[0084] Further, in step 1, the watermark information is the original binary watermark image information, and the original binary watermark image is M0={m0(i, j), 0≤i<X, 0≤j<Y}, wherein X and Y are the width and height of the original binary watermark image, respectively, and m0(i, j)∈{0,1} represents the gray value of the pixel point in the i-th row and j-th column of the original binary watermark image; the scrambling encryption and dimensionality reduction processing includes the following steps:
[0085] Step 1.1, perform Arnold scrambling on the original binary watermark image M0, and use the number of iterations and control parameters selected during Arnold scrambling as the key K1. After the Arnold scrambling is completed, the scrambled binary watermark image M1 = {m i (i, j), 0≤i<X, 0≤j<Y};
[0086] Step 1.2, perform dimensionality reduction operation on the scrambled binary watermark image M1 obtained in step 1.1 to obtain a one-dimensional binary sequence w0 = {w(i), 0≤i<MN};
[0087] Step 1.3, generate the Logistic chaotic sequence, the generation method is shown in formula 1;
[0088] x i+1 =μx i (1-x i ) (Formula 1)
[0089] Among them, x i Indicates the current state, μ is the control parameter, 0<x i <1, when 3.569946<μ≤4, the Logistic map is in a chaotic state; select x0 and μ that make the Logistic map in a chaotic state as the key K2;
[0090] Step 1.4, based on the Logistic chaotic sequence obtained in step 1.3, use formula 2 to generate a binary chaotic sequence v;
[0091]
[0092] Among them, δ is the decision threshold, 0<δ<1, and v is the generated binary chaotic sequence;
[0093] Step 1.5, perform an XOR operation on the one-dimensional binary sequence w0 obtained in step 1.2 and the binary chaotic sequence v generated in step 1.4 to obtain the final watermark sequence w to be embedded.
[0094] The key K1 and the key K2 together constitute the only correct decryption key for watermark extraction.
[0095] Step 2, select the watermark embedding position: According to the analysis of the psychoacoustic model, comprehensively consider the sensitivity of the human ear to different frequencies and the impact of MP3 compression on audio and watermarks, and consider that the digital audio watermark will be more robust to common attacks when embedded in the low-frequency area than when embedded in the high-frequency area. The selected watermark embedding position is in the low- to medium-frequency area of the audio carrier signal, and the frequency range is between 500Hz and 3000Hz, with 1500Hz being the best frequency under normal circumstances.
[0096] Step 3, embedding watermark information: adaptively embedding the watermark information preprocessed in step 1 into the watermark embedding position in the audio carrier signal selected in step 2 to obtain an audio signal containing the watermark;
[0097] The step 3 specifically includes:
[0098] Step 3.1, normalize and frame the original audio signal, the frame length is N, and each frame of the audio signal is embedded with 1-bit binary watermark information;
[0099] Step 3.2, perform 6-level discrete wavelet packet transform (DWPT) on each frame of audio signal in step 3.1, and select a suitable sub-band to embed the watermark sequence w according to the frequency interval selected in step 2; assuming that the length of the sub-band coefficient is L, the relationship between L and N is shown in Formula 3;
[0100]
[0101] Step 3.3, select n coefficients on the sub-band obtained in step 3.2 for watermark embedding operation, the value of n is an odd number less than the length of the sub-band coefficient, and n is at least 1; when n>1, watermark information will be embedded redundantly, that is, the same watermark information is embedded in the n coefficient positions;
[0102] Step 3.4, construct the Fibonacci sequence Fib0 = {0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, ...}, divide the Fibonacci sequence Fib0 as a whole by a constant, so that the positions of all wavelet packet domain coefficients in the sequence are distributed as evenly as possible, and obtain the Fibonacci sequence Fib used in the adaptive quantization watermark embedding process; Assume that there are two adjacent Fibonacci numbers F in the Fibonacci sequence Fib k and F k+1 , and the wavelet packet domain coefficient is greater than F k and less than F k+1 , then the position of the wavelet packet domain coefficient in the Fibonacci sequence Fib is the index k;
[0103] Step 3.5, define the quantization step size S, assuming that the wavelet packet domain coefficients of the current frame are d = {d0, d1, d2, ..., d L-1}, the watermark information embedded in the current frame is w(j)∈{0,1}, and the n coefficients described in step 3.3 in the current frame are modified according to the embedding strategy of formula 4 and formula 5 to adaptively quantize the embedded watermark information, with one of the n coefficients d i Take as an example, and the other (n-1) coefficients are similar. The embedding strategy of formula 4 and formula 5 is:
[0104]
[0105] d i =t i S (Formula 5)
[0106] Among them, t i is the coefficient d i The position in the quantization interval, L fib is the length of the Fibonacci sequence Fib in step 3.4, fib i Yes i The position in Fib, that is, when d i Greater than the element Fib in Fib k and less than Fib k+1The index k when a is used; a is the control factor, 0≤a≤1, which is used to control "coefficients of different sizes will change to different degrees according to their positions in Fib"
[0107] The intensity of this phenomenon; l and r are the bottom factors, 0≤l≤0.5, 0.5≤r≤1, which are used to ensure that the modified d i Within the range [l, r] of the corresponding quantization interval, the robustness of the algorithm is guaranteed;
[0108] Step 3.6, performing an inverse discrete wavelet packet transform (IDWPT) on each frame of audio after the watermark information is embedded in step 3.5, and restoring the audio frame to a time domain signal;
[0109] Step 3.7, all the audio frames obtained in step 3.6 are stitched together in order, and denormalized to obtain an audio signal containing a watermark.
[0110] Step 4, optimizing embedding parameters: Based on the particle swarm optimization algorithm, multiple embedding parameters involved in the watermarked audio signal in step 3 are optimized to obtain the final watermarked audio signal.
[0111] Further, in step 4, the optimization of the embedding parameters is achieved by using a particle swarm optimization algorithm, using the signal to noise ratio (SNR) as an indicator of transparency, and using the normalized correlation coefficient (NC) as an indicator of robustness, to optimize the quantization step size S, the control factor a, and the bottom factor l and r in step 3.5; the particle speed and position update formulas in the particle swarm optimization algorithm are as follows:
[0112] v i (t+1)=wv i (t)+c1r1(p best (t)-x i (t))+c2r2(g best (t)-x i (t)) (Formula 6)
[0113] x i (t+1)=x i (t)+v i (t+1) (Formula 7)
[0114] In Formula 6 and Formula 7, v i (t) and x i (t) are the velocity and position of each particle i at time t, p bestis the optimal position of each particle, i.e., its optimal solution in the search process, g best is the global best position of the entire particle swarm, that is, the optimal solution of the entire particle swarm, w represents the inertia weight, which is used to adjust the search ability for the solution space, c1 and c2 represent learning factors, which are called cognitive parameters and social parameters respectively, r1 and r2 are random numbers uniformly distributed in [0, 1], which can increase the randomness of the search process; the position of each particle can represent a set of parameters in step 3.5, namely, the quantization step size S, the control factor a, and the bottom factor l and r;
[0115] The specific steps include:
[0116] Step 4.1, initialize the population and set the relevant parameters of the particle swarm optimization algorithm, including the number of population particles, inertia weight, cognitive parameters, social parameters, number of iterations, etc.;
[0117] Step 4.2, according to the population initialized in step 4.1, perform the particle swarm optimization algorithm iteration process to find the optimal watermark embedding parameters, namely, the quantization step size S, the control factor a, and the floor factors l and r, and embed the watermark into the audio according to the optimal watermark embedding parameters to obtain the final audio signal containing the watermark.
[0118] Furthermore, the step 4.2 of the particle swarm optimization algorithm iteration process includes the following sub-steps:
[0119] Step 4.2.1, update the particle speed and position in each iteration;
[0120] Step 4.2.2, according to the updated particle position obtained in step 4.2.1, update the individual optimal solution and the global optimal solution;
[0121] Step 4.2.3, according to the current particle position, obtain the corresponding quantization step size S, control factor a and floor factors l and r, embed the watermark information in the audio signal, and calculate the SNR value of the audio after the watermark information is embedded under the current parameters;
[0122] Step 4.2.4, attacking the watermarked audio signal after embedding the watermark information obtained in step 4.2.3. In the embodiment of the present invention, the attack means adopted include adding Gaussian white noise, bandpass filtering, resampling, re-quantization and MP3 compression;
[0123] Step 4.2.5, extracting the watermark from the attacked audio signal obtained in step 4.2.4, and comparing it with the watermark information originally embedded in step 4.2.3, to obtain the NC value between the watermark information extracted under different attacks and the original watermark information;
[0124] Step 4.2.6, construct the objective function based on the SNR value obtained in step 4.2.3 and the NC value obtained in step 4.2.5. Use the objective function to calculate the fitness of each particle and evaluate the population in each iteration. When a satisfactory solution is found or the maximum number of iterations is reached, the optimal solution is considered to be found. The objective function is shown in Formula 8:
[0125]
[0126] Among them, λ1 and λ2 are weight factors, which respectively represent the importance of transparency and robustness to the objective function. The larger the λ1, the more inclined to transparency, and the larger the λ2, the more inclined to robustness. R represents the number of attack types, r i is the weight factor of each attack. When the algorithm considers the same weight for different attacks, r i A constant can be taken;
[0127] Step 4.2.7, taking the optimal solution obtained in step 4.2.6 as the final selected embedding parameter, embedding the watermark information according to step 3 to obtain the final audio signal containing the watermark.
[0128] like Figure 2 As shown, an embodiment of the present invention provides a wavelet packet domain adaptive quantization digital audio watermark extraction method based on Fibonacci sequence and particle swarm optimization, comprising the following steps:
[0129] Step 5, audio preprocessing: normalize and frame the audio signal to be watermarked, the frame length is N, the same as in step 3.1 of the embedding method, each frame of audio signal contains 1 bit of binary watermark information, and perform 6-level DWPT transformation on each frame of audio signal;
[0130] Step 6, select the watermark extraction position: select the same sub-band as in step 3.2 of the embedding method, and select n coefficients in the sub-band with the same positions as in step 3.3 of the embedding method;
[0131] Step 7, extracting watermark information: performing watermark extraction operation on the n coefficients in the sub-band described in step 6, taking one of the n coefficients d′ i Take it as an example, the other (n-1) coefficients are extracted in the same way as shown in Formula 9;
[0132]
[0133] Step 8, counting and determining the watermark information in the current frame: after extracting the watermark information in step 7, n binary watermark information can be obtained for each frame of audio signal; these n binary watermark information are counted to obtain the number of "0" zc and the number of "1" oc, and the method shown in formula 10 is used to determine the watermark information contained in the current audio frame;
[0134]
[0135] Step 9, sequentially splicing the watermark information: sequentially splicing the watermark information contained in all the audio frames obtained in step 8, to obtain the binary watermark sequence w′ finally extracted from the audio signal to be extracted;
[0136] Step 10, decryption and recovery: Use the key K1 obtained in the watermark information preprocessing step 1.1 of the embedding method, the key K2 obtained in step 1.3, and the binary chaotic sequence v obtained in step 1.4 to decrypt and recover the binary watermark sequence w′ extracted in step 9. When decrypting, first use the key K2 to perform Logistic scrambling recovery on the binary watermark sequence w′, and perform an XOR operation on the binary watermark sequence w′ and the binary chaotic sequence v to obtain the decrypted and recovered binary sequence w″. Then, the decrypted and recovered binary sequence w″ is dimensionally upgraded to obtain a binary image M′; finally, use the key K1 to perform Arnold scrambling recovery on the binary image M′ to obtain the final decrypted and recovered watermark image M″.
[0137] The present invention discloses a wavelet packet domain adaptive quantization digital audio watermark embedding and extraction method based on Fibonacci sequence and particle swarm optimization. The present invention first normalizes and frames the audio carrier signal, then performs discrete wavelet packet transform to obtain a number of sub-bands, and then selects sub-bands with appropriate frequency intervals according to the psychoacoustic model to embed and extract watermark information. The present invention utilizes the Fibonacci sequence to perform different degrees of quantitative modification on wavelet packet domain coefficients of different sizes. The core idea is to perform a smaller degree of modification on smaller coefficients and a larger degree of modification on larger coefficients, thereby achieving the degree of coefficient modification adaptation, reducing the impact of watermark information embedding on the original audio signal, thereby improving the transparency of the watermark. The present invention adopts a particle swarm optimization algorithm, and according to the result of the objective function obtained when the audio signal is attacked to extract the watermark, multiple parameters involved in the present invention are tuned, so that the watermark can adaptively balance the robustness and transparency of the watermark for different music. The digital audio watermark embedding and extraction method provided by the present invention is a very effective method for embedding watermark information in audio, and has outstanding advantages: it has good universality, can adaptively balance the contradictory relationship between the robustness and transparency of the watermark, and has good transparency while being able to resist various attack methods.
[0138] like Figure 3 As shown, the original binary watermark image M0 provided by the embodiment of the present invention is a 32×32 binary image, and what is shown in the figure is the effect after enlargement.
[0139] like Figure 4As shown, the scrambled binary watermark image M1 provided by the embodiment of the present invention is obtained in step 1.1. Figure 3 The original binary watermark image M0 shown is subjected to Arnold scrambling to obtain a scrambled binary watermark image M1.
[0140] In the embodiment of the present invention, after step 1, the length of the final watermark sequence w to be embedded is 1024 bits.
[0141] like Figure 5 As shown in FIG. 1 , a schematic diagram of the sub-band frequency range after the 6-level wavelet packet transform of an audio signal provided by an embodiment of the present invention is shown. Taking an audio signal with a sampling rate of 44.1kHz as an example, it can be known from the sampling theorem that the sampling frequency is at least twice the highest frequency of the signal, so the highest frequency of the signal that can be represented is 22.05kHz. After the 6-level wavelet packet transform of the audio signal is performed, a total of 64 sub-bands can be obtained, and the frequency range represented by each sub-band is as follows: Figure 5 As shown in the figure, the frequency ranges represented by sub-bands 4, 5, 6, etc. are all in the mid-low frequency region and are relatively close to 1500Hz, so they can be preferentially selected as sub-bands. According to the frequency domain characteristics of different types of audio signals, the selection position can be appropriately adjusted to reduce the impact of the watermark on the audio signal, or multiple sub-bands can be selected to embed the watermark at the same time to obtain stronger robustness. When selecting, it is necessary to weigh the relationship between the transparency and robustness of the watermark. It is necessary to select an area that is relatively imperceptible to the human ear as much as possible, and to fully consider the robustness of the watermark under compression attacks. The selection of watermark embedding positions for audio signals at other sampling rates is similar. The basic idea is to try to ensure that the selected interval is close to the mid-low frequency region.
[0142] like Figure 6 As shown, a schematic diagram of the modification strategy of the coefficients in the embedding method provided in an embodiment of the present invention in the quantization interval is intended to better illustrate and explain the modification strategy in Formula 4, and embodies the 5 modification rules in Formula 4 in the quantization interval.
[0143] Figure 6 The bold part in the middle represents a complete quantization interval. A total of three complete intervals are shown, and the three intervals represent the watermark information "1", "0", and "1" respectively. The unit length of the quantization interval is 1. Each interval represents a watermark information "0" or "1", and the information is arranged alternately. The black dot in the figure is the midpoint of the interval, and the red dot is the position of the floor factors l and r. In the same interval, the area between l and r is within the floor interval, and the area outside l to r is outside the floor interval. Assuming that the quantization interval where the current coefficient is located is the "0" interval, the bold interval in the figure represents the quantization interval where the current coefficient is located. Combined with Figure 6 , the description and explanation of the five modification rules in formula 4 are as follows:
[0144] Rule 1, that is, the quantization interval of the current coefficient is the same as the watermark information to be embedded (the watermark information is "0"), and the position of the current coefficient in the quantization interval is within the guaranteed interval. Refer to the orange line 4 and orange line 5 in the figure. At this time, the modification strategy is to move the coefficient closer to the midpoint of the quantization interval according to its position in Fib, and the value smaller than the midpoint of the interval becomes larger, and the value larger than the midpoint of the interval becomes smaller.
[0145] Rule 2, that is, the quantization interval of the current coefficient is the same as the watermark information to be embedded (the watermark information is "0"), and the position of the current coefficient in the quantization interval is located on the left side outside the minimum interval. Refer to the green line 3 in the figure. At this time, the modification strategy is to modify to the minimum factor l.
[0146] Rule 3, that is, the quantization interval of the current coefficient is the same as the watermark information to be embedded (the watermark information is "0"), and the position of the current coefficient in the quantization interval is located on the right side outside the minimum interval. Refer to the blue line 7 in the figure. At this time, the modification strategy is to modify to the minimum factor r.
[0147] Rule 4, that is, the quantization interval where the current coefficient is located is different from the watermark information to be embedded (the watermark information is "1"), and the current coefficient is located on the left side of the midpoint of the quantization interval. Refer to the red line 1 in the figure. At this time, the modification strategy is to move the coefficient to between the floor factor r and the midpoint of the adjacent interval on the left according to the position of the coefficient in Fib. If the position of the coefficient in Fib is later, or the control factor a is larger, the position after the move is closer to the midpoint of the adjacent interval on the left, which is reflected in the figure that the displacement distance of red line 6 is greater than the displacement distance of red line 1.
[0148] Rule 5, that is, the quantization interval where the current coefficient is located is different from the watermark information to be embedded (the watermark information is "1"), and the current coefficient is located to the right of the midpoint of the quantization interval. Refer to the black line 8 in the figure. At this time, the modification strategy is to move the coefficient to between the floor factor l and the midpoint of the adjacent interval on the right according to the position of the coefficient in Fib. If the position of the coefficient in Fib is later, or the control factor a is larger, the position after the move is closer to the midpoint of the adjacent interval on the right, which is reflected in the figure that the displacement distance of black line 8 is greater than the displacement distance of black line 2.
[0149] In the embodiment of the present invention, the original audio signal uses the audio data in the GTZAN data set, which are all WAV format audio files with a sampling frequency of 22.05kHz and a length of 30 seconds. Compared with the common audio with a sampling frequency of 44.1kHz, the audio with a sampling frequency of 22.05kHz has fewer samples in the same time, and the difficulty of watermark embedding will be higher. Therefore, the use of 22.05kHz audio as the embedding object in the embodiment of the present invention can better prove the effectiveness and universality of the present invention.
[0150] like Figure 7As shown, an embedding parameter optimization flow chart based on a particle swarm optimization algorithm provided in an embodiment of the present invention, in the embedding parameter optimization step of the embedding method in an embodiment of the present invention, the parameter settings of the particle swarm optimization algorithm are shown in Table 1.
[0151] Table 1 Particle swarm optimization algorithm parameter setting table
[0152]
[0153] like Figure 8 As shown in FIG. 1 , a schematic diagram of the change of the objective function value of the global optimal solution in the particle swarm optimization algorithm provided by the embodiment of the present invention with the iteration process is shown. Since the particle swarm optimization algorithm used in the embodiment of the present invention is used to solve the minimum value, Figure 8 The objective function value curve shown in is the result of taking a negative value based on Formula 8. It can be seen that the curve basically converges to a stable value after the number of iterations reaches 50. Therefore, considering the time factor comprehensively, in the embodiment of the present invention, in the step of optimizing the embedding parameters of the embedding method, the maximum number of iterations of the particle swarm optimization algorithm is 50.
[0154] like Fig. 9 As shown in the figure, the audio waveforms before and after the watermark is embedded in the embodiment of the present invention. The audio signal shown in the figure is the 8th music of the pop type in the GTZAN data set. The waveforms in the figure are the original audio signal, the watermarked audio signal and the difference signal waveform from top to bottom. Obviously, the difference before and after the watermark is embedded is almost indistinguishable from the waveform, indicating that the watermark has little effect on the original audio signal waveform.
[0155] To prove the effectiveness and universality of the present invention, the transparency and robustness of the watermark are tested and evaluated in the embodiments of the present invention. The SNR and Mean Opinion Score (MOS) are used to evaluate the transparency of the watermark. The SNR value can objectively evaluate the degree of difference between the watermarked audio and the original audio from a mathematical perspective. Generally, the higher the SNR value, the better the watermark transparency. The MOS score can better subjectively evaluate the audio quality from a human perspective. The higher the MOS score, the better the watermark transparency. The NC and Bit Error Rate (BER) are used to evaluate the robustness of the watermark. The higher the NC value, the better the watermark robustness, and the lower the BER value, the better the watermark robustness. During the test, the embedding method of the present invention was first used to embed watermarks in ten types of music, and then the watermarked audio was processed by the following attack methods: (A) adding 30dB Gaussian white noise; (B) adding 20dB Gaussian white noise; (C) resampling, the sampling was first downsampled from 22.05kHz to 11.025kHz, and then upsampled to 22.05kHz; (D) requantization, the sampling accuracy was first quantized from 16 bits to 8 bits, and then quantized to 16 bits; (E) low-pass filtering, low-pass filtering with a cutoff frequency of 4kHz; (F) MP3 compression, first MP3 compression with a bit rate of 128kbps, and then decompressed to restore to WAV format; (G) MP3 compression, first MP3 compression with a bit rate of 64kbps, and then decompressed to restore to WAV format. Finally, the watermark was extracted and evaluated using the extraction method of the present invention. The test results of transparency and robustness are shown in Tables 2 and 3, respectively, where the BER in Table 3 is a percentage result.
[0156] According to the transparency test results in Table 2, it can be seen that the present invention has good transparency. The SNR values of the ten types of songs are all above 36dB, with an average value of 38.10dB, which is much higher than the 20dB required by IFPI. This objectively shows that the present invention has good transparency. The MOS scores of the ten types of songs are all 4.7 and above, with an average value of 4.83. According to the MOS judgment standard, the audio quality after embedding the watermark is close to excellent, which subjectively shows that the present invention has good transparency and is not easily detected by the human ear. In addition, the SNR values and MOS scores between different music types are relatively close, which verifies the universality of the present invention in terms of transparency.
[0157] Table 2 Transparency test results
[0158]
[0159] Table 3 Robustness test results
[0160]
[0161]
[0162] According to the robustness test results in Table 3, it can be seen that for different music types, the NC values of the watermark under several attack methods are all higher than 0.96, with an average value of 0.993, and the BER is less than 2.2%, with an average value of 0.396%. The NC value under the two attack methods of adding 30dB Gaussian white noise and re-quantization is 1, indicating that the present invention is very robust to these two attack methods, and is relatively less robust to MP3 compression attacks at 64kbps, with an average BER of 1.432%. The test results show that the present invention can resist a variety of attack methods and has good universality in terms of robustness.
[0163] The above results and analysis are obtained by balancing transparency and robustness. In the objective function of the particle swarm optimization algorithm, the weight of each attack method is the same. In actual application scenarios, the parameters can be adjusted appropriately according to the needs, so that the algorithm focuses more on transparency or robustness, and achieves results that better meet the needs of the scenario.
[0164] The above whole process includes the whole process of the wavelet packet domain adaptive quantization digital audio watermark embedding and extraction method based on Fibonacci sequence and particle swarm optimization proposed by the present invention. The above is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement and improvement made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention shall be covered within the protection scope of the present invention.
Claims
1. A wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization, characterized in that: It includes the following steps: Step 1, preprocessing of watermark information: performing scrambling encryption and dimensionality reduction on the watermark information; Step 2, selecting the watermark embedding position: selecting the watermark embedding position in the mid-low frequency region of the audio carrier signal; Step 3, embedding the watermark information: adaptively embedding the preprocessed watermark information in Step 1 into the watermark embedding position in the mid-low frequency region of the audio carrier signal selected in Step 2 to obtain the watermarked audio signal; Step 4, optimizing the embedding parameters: based on the particle swarm optimization algorithm, optimizing multiple embedding parameters involved in the watermarked audio signal in Step 3 to obtain the final watermarked audio signal.
2. According to claim 1, a wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization is characterized in that: In Step 1, the watermark information is the original binary watermark image information. The original binary watermark image is M0 = {m0(i, j), 0 ≤ i < X, 0 ≤ j < Y}, where X and Y are the width and height of the original binary watermark image respectively, and m0(i, j) ∈ {0, 1} represents the gray value of the pixel at the i-th row and j-th column of the original binary watermark image; the scrambling encryption and dimensionality reduction processing include the following steps: Step 1.1, performing Arnold scrambling on the original binary watermark image M0, and taking the iteration times and control parameters selected during Arnold scrambling as the key K1 together. After Arnold scrambling is completed, the scrambled binary watermark image M1 = {m1(i, j), 0 ≤ i < X, 0 ≤ j < Y} is obtained; Step 1.2, performing dimensionality reduction operation on the scrambled binary watermark image M1 obtained in Step 1.1 to obtain a one-dimensional binary sequence w0 = {w(i), 0 ≤ i < MN}; Step 1.3, generating a Logistic chaotic sequence, and the generation method is shown in Formula 1: x i+1 =μx i (1-x i ) (Formula 1) Among them, x i represents the current state, μ is the control parameter, 0 <x i <1, when 3.569946<μ≤4, the Logistic map is in a chaotic state; select x0 and μ when the Logistic map is in a chaotic state as the key K2; Step 1.4, according to the Logistic chaotic sequence obtained in Step 1.3, generating a binary chaotic sequence v using Formula 2; where δ is the decision threshold, 0 < δ < 1, and v is the generated binary chaotic sequence; Step 1.5, performing exclusive OR operation on the one-dimensional binary sequence w0 obtained in Step 1.2 and the binary chaotic sequence v generated in Step 1.4 to obtain the final watermark sequence w to be embedded.
3. According to claim 1, a wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization is characterized in that: The frequency range of the mid-low frequency region in Step 2 is within 500 Hz to 3000 Hz.
4. According to claim 1, a wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization is characterized in that: Step 3 specifically includes: Step 3.1, normalizing and framing the original audio signal, with the frame length being N, and embedding 1-bit binary watermark information in each frame of the audio signal; Step 3.2, performing 6-level discrete wavelet packet transform (Discrete Wavelet Packet Transform, DWPT) on each frame of the audio signal in Step 3.1, and selecting a suitable sub-band to embed the watermark sequence w according to the frequency range selected in Step 2; assuming the length of the sub-band coefficient is L, the relationship between L and N is shown in Formula 3; Step 3.3, selecting n coefficients on the sub-band obtained in Step 3.2 for watermark embedding operation, where the value of n is an odd number less than the length of the sub-band coefficient, and the minimum value of n is 1; when n > 1, redundant watermark information will be embedded at this time, that is, the same watermark information is embedded at n coefficient positions. Step 3.4, construct the Fibonacci sequence Fib0 = {0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, ...}, divide the Fibonacci sequence Fib0 as a whole by a constant to make the positions of all wavelet packet domain coefficients in the sequence as evenly distributed as possible, and obtain the Fibonacci sequence Fib used in the adaptive quantization watermark embedding process; Assume that there are two adjacent Fibonacci numbers F in the Fibonacci sequence Fib k and F k+1 , and the wavelet packet domain coefficient is greater than F k and less than F k+1 , then the position of the wavelet packet domain coefficient in the Fibonacci sequence Fib is the index k; Step 3.5, define the quantization step size S, assuming that the wavelet packet domain coefficients of the current frame are d = {d0, d1, d2, ..., d L-1 }, the watermark information embedded in the current frame is w(j)∈{0,1}, and the n coefficients described in step 3.3 in the current frame are modified according to the embedding strategy of formula 4 and formula 5 to adaptively quantize the embedded watermark information, with one of the n coefficients d i Take as an example, and the other (n-1) coefficients are similar. The embedding strategy of formula 4 and formula 5 is: d i =t i S (Formula 5) Among them, t i is the coefficient d i The position in the quantization interval, Lf ib is the length of the Fibonacci sequence Fib in step 3.4, fib i Yes i The position in Fib, that is, when d i Greater than the element Fib in Fib k and less than Fib k+1 The index k when the modified d is 0≤a≤1; a is the control factor, 0≤a≤1, which is used to control the intensity of the phenomenon that "coefficients of different sizes change to different degrees according to their positions in Fib"; l and r are the bottom factors, 0≤l≤0.5, 0.5≤r≤1, which are used to ensure that the modified d i In the range [l,r] of the corresponding quantization interval; Step 3.6, performing an inverse discrete wavelet packet transform (IDWPT) on each frame of audio after the watermark information is embedded in step 3.5, and restoring the audio frame to a time domain signal; Step 3.7, all the audio frames obtained in step 3.6 are stitched together in order, and denormalized to obtain an audio signal containing a watermark.
5. The wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization according to claim 1 is characterized in that: In step 4, the optimization of embedding parameters is achieved by using a particle swarm optimization algorithm, using the signal to noise ratio (SNR) as an indicator of transparency, and the normalized correlation coefficient (NC) as an indicator of robustness, to optimize the quantization step size S, the control factor a, and the bottom factor l and r in step 3.5; the particle speed and position update formula in the particle swarm optimization algorithm are as follows: v i (t + 1)= wv i (t)+ c1r1(p best (t)- x i (t))+ c2r2(g best (t)- x i (t)) (Equation 6) x i (t+1)=x i (t)+v i (t+1) (Formula 7) In Formula 6 and Formula 7, v i (t) and x i (t) are the velocity and position of each particle i at time t, p best is the optimal position of each particle, i.e., its optimal solution in the search process, g best is the global best position of the entire particle swarm, that is, the optimal solution of the entire particle swarm, w represents the inertia weight, which is used to adjust the search ability for the solution space, c1 and c2 represent learning factors, respectively called cognitive parameters and social parameters, r1 and r2 are random numbers uniformly distributed in [0,1], which can increase the randomness of the search process; the position of each particle can represent a set of parameters in step 3.5, namely, the quantization step size S, the control factor a, and the bottom factor l and r; The specific steps include: Step 4.1, initialize the population and set the relevant parameters of the particle swarm optimization algorithm; Step 4.2, according to the population initialized in step 4.1, perform the particle swarm optimization algorithm iteration process to find the optimal watermark embedding parameters, namely, the quantization step size S, the control factor a, and the floor factors l and r, and embed the watermark into the audio according to the optimal watermark embedding parameters to obtain the final audio signal containing the watermark.
6. The wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization according to claim 5 is characterized in that: The step 4.2, during the particle swarm optimization algorithm iteration, includes the following sub-steps: Step 4.2.1, update the particle speed and position in each iteration; Step 4.2.2, according to the updated particle position obtained in step 4.2.1, update the individual optimal solution and the global optimal solution; Step 4.2.3, according to the current particle position, obtain the corresponding quantization step size S, control factor a and floor factors l and r, embed the watermark information in the audio signal, and calculate the SNR value of the audio after the watermark information is embedded under the current parameters; Step 4.2.4, attacking the watermarked audio signal after embedding the watermark information obtained in step 4.2.3; Step 4.2.5, extracting the watermark from the attacked audio signal obtained in step 4.2.4, and comparing it with the watermark information originally embedded in step 4.2.3, to obtain the NC value between the watermark information extracted under different attacks and the original watermark information; Step 4.2.6, construct the objective function based on the SNR value obtained in step 4.2.3 and the NC value obtained in step 4.2.
5. Use the objective function to calculate the fitness of each particle and evaluate the population in each iteration. When a satisfactory solution is found or the maximum number of iterations is reached, the optimal solution is considered to be found. The objective function is shown in Formula 8: Among them, λ1 and λ2 are weight factors, which respectively represent the importance of transparency and robustness to the objective function. The larger the λ1, the more inclined to transparency, and the larger the λ2, the more inclined to robustness. R represents the number of attack types, r i is the weight factor of each attack. When the algorithm considers the same weight for different attacks, r i A constant can be taken; Step 4.2.7, taking the optimal solution obtained in step 4.2.6 as the final selected embedding parameter, embedding the watermark information according to step 3 to obtain the final audio signal containing the watermark.
7. An extraction method based on the wavelet packet domain adaptive quantization digital audio watermark embedding method based on Fibonacci sequence and particle swarm optimization according to any one of claims 1 to 6, characterized in that: The following steps are involved: Step 5, audio preprocessing: normalize and frame the audio signal to be watermarked, the frame length is N, the same as in step 3.1 of the embedding method, each frame of audio signal contains 1 bit of binary watermark information, and perform 6-level DWPT transformation on each frame of audio signal; Step 6, select the watermark extraction position: select the same sub-band as in step 3.2 of the embedding method, and select n coefficients in the sub-band with the same positions as in step 3.3 of the embedding method; Step 7, extracting watermark information: performing watermark extraction operation on the n coefficients in the sub-band described in step 6, taking one of the n coefficients d' i Take it as an example, and the other (n-1) coefficients are extracted in the same way as shown in Formula 9: Step 8, counting and determining the watermark information in the current frame: after extracting the watermark information in step 7, n binary watermark information is obtained for each frame of audio signal; The n binary watermark information is counted to obtain the number of "0" zc and the number of "1" oc, and the method shown in formula 10 is used to determine the watermark information contained in the current audio frame; Step 9, sequentially concatenating the watermark information: sequentially concatenating the watermark information contained in all the audio frames obtained in step 8 to obtain a binary watermark sequence w' finally extracted from the audio signal to be extracted; Step 10, decryption and recovery: Use the key K1 obtained in the watermark information preprocessing in step 1.1 of the embedding method, the key K2 obtained in step 1.3, and the binary chaotic sequence v obtained in step 1.4 to decrypt and recover the binary watermark sequence w' extracted in step 9. When decrypting, first use the key K2 to perform Logistic scrambling recovery on the binary watermark sequence w', and perform an XOR operation on the binary watermark sequence w' and the binary chaotic sequence v to obtain the decrypted and recovered binary sequence w"; then, the decrypted and recovered binary sequence w" is dimensionally upgraded to obtain a binary image M'; finally, use the key K1 to perform Arnold scrambling recovery on the binary image M' to obtain the final decrypted and recovered watermark image M".
Citation Information
Patent Citations
Image enhancement method based on adaptive immunity genetic algorithm
CN104123706A
Audio watermarking method based on DWT-SVD and PSO (Particle Swarm Optimization)
CN106611601A
Fusion domain blind watermarking method based on graph transformation and particle swarm optimization algorithm
CN115510404A
Audio watermark embedding and extracting method
CN116895279A
Dynamic range compression of ultrasound images
US20180278944A1
Cited By
Dark watermark generation system and generation method thereof
CN119293764A
A dark watermark generation system and its generation method
CN119293764B
