CN113096671B - A method and system for reversible information hiding of large-capacity audio files - Google Patents

Abstract

The invention discloses a reversible information hiding method and a reversible information hiding system for a high-capacity audio file, which comprise an embedding process: dividing different audio points of the initial file into a cross set and a point set according to positions; respectively constructing a first cross set prediction error matrix and a first point set prediction error matrix by using local consistency characteristics between adjacent audio frequency points and adopting a target audio point amplitude prediction algorithm; dividing data to be hidden into two equal halves, a first part being embedded in a cross set and a second part being embedded in a dot set; reversibly constructing a cross set embedded audio sequence and a point set embedded audio sequence according to an error prediction algorithm; and finally combining to form the secret audio file. The invention embeds the secret information into the audio carrier file for many times based on the characteristic that the extension sequences are mutually orthogonal, most elements of different extension sequences are mutually offset in the information embedding process, and the audio fidelity capability of the file is improved while the embedding capacity of the reversible information of the carrier audio is increased.

Description

A method and system for reversible information hiding of large-capacity audio files

technical field

The invention relates to the technical field of information hiding, in particular to a reversible information hiding method and system for large-capacity audio files.

Background technique

The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

Information hiding realizes functions such as digital steganography, secret communication, and copyright protection by embedding secret information into the carrier. Reversible information hiding is one of the important branches. While ensuring the complete extraction of the embedded secret information, it can also restore the original carrier signal losslessly. Therefore, in military, medical and other fields that require high original images or audio and video files, copyright authentication and covert communication algorithms based on reversible information hiding have a wide range of application requirements. There are two standards for reversible information hiding, namely, the embedding capacity and the distortion rate. An excellent reversible information hiding algorithm requires a higher information embedding capacity while keeping the carrier file with a lower degree of distortion.

In recent years, with the advancement of the digital society, massive audio data has been spread on the Internet, and the development of audio information hiding technology has become more and more mature. The secret information is usually embedded in the LSB (least significant bit) of the carrier sample in the form of a bit stream, the distortion of the carrier audio is almost negligible perceptually, and the zero distortion of the carrier audio data can be achieved by reversible information hiding.

The reversible information hiding (RDH) algorithm has been greatly developed in recent years, but the reversible information hiding error prediction and information embedding algorithms are becoming more and more complex, and the computational cost is getting higher and higher. At the same time, in most cases, audio quality degrades rapidly as the payload increases.

SUMMARY OF THE INVENTION

In order to solve the above problems, the present invention proposes a reversible information hiding method and system for large-capacity audio files. According to the principle of Code Division Multiple Access (CDMA), different orthogonal sequences are used to represent the information to be embedded and embedded in the carrier audio; Based on the linear independence of orthogonal vectors, covert information can be overlapped and embedded without interfering with each other, thus enabling the embedding of large-capacity information. At the same time, since most elements of different extension sequences cancel each other during overlapping embedding, high-capacity reversible information embedding is achieved while reducing audio file distortion, thereby effectively improving the reversible information hiding performance of audio files.

In some embodiments, the following technical solutions are adopted:

A reversible information hiding method for large-capacity audio files, including the embedding process:

Divide the different audio points of the initial file into fork sets and point sets according to their positions;

Using the local consistency feature between adjacent audio points, the target audio point amplitude prediction algorithm is used to construct the first fork set prediction error matrix and the first point set prediction error matrix respectively;

The data to be hidden is divided into two equal halves, the first part is embedded in the fork set, and the second part is embedded in the point set;

According to the error prediction algorithm, the fork-set ciphered audio sequence and the point-set ciphered audio sequence are reversibly constructed; finally, the ciphered audio file is formed by merging.

Further, the extraction process is also included:

Divide the encrypted audio files into fork sets and point sets;

According to the target error prediction algorithm, construct the second point set prediction error matrix and the second fork set prediction error matrix respectively;

Recover the attachment information embedded in the point set prediction error matrix and the cross set prediction error matrix respectively, extract the secret information embedded in the error matrix according to the code division multiple access reversible information embedding algorithm, and restore the original amplitude of the target audio point;

The extracted secret information is connected to reconstruct the hidden secret information sequence, and the original audio file is restored losslessly by using the original amplitude matrix of the restored target audio point.

In other embodiments, the following technical solutions are adopted:

A reversible information hiding system for large-capacity audio files, comprising:

A device for dividing different audio points in the initial file into fork sets and point sets according to their positions;

A device for constructing a first fork set prediction error matrix and a first point set prediction error matrix respectively by utilizing the local consistency feature between adjacent audio points and adopting the target audio point amplitude prediction algorithm;

A means for dividing the data to be hidden into equal halves, the first part is embedded in the fork set, and the second part is embedded in the point set;

A device for reversibly constructing a fork-set embedded cipher audio sequence and a point-set cipher audio sequence according to an error prediction algorithm; and finally merging to form a ciphered audio file.

Further, it also includes:

means for dividing the encrypted audio file into fork sets and point sets;

A device for constructing the second point set prediction error matrix and the second cross set prediction error matrix respectively according to the target error prediction algorithm;

A device for restoring the attachment information embedded in the point set prediction error matrix and the cross set prediction error matrix respectively, extracting the secret information embedded in the error matrix according to the code division multiple access reversible information embedding algorithm, and restoring the original amplitude value of the target audio point;

The device is used to connect the extracted secret information to reconstruct the hidden secret information sequence, and use the restored original amplitude matrix of the target audio point to restore the original audio file losslessly.

In other embodiments, the following technical solutions are adopted:

A terminal device, which includes a processor and a computer-readable storage medium, where the processor is used to implement various instructions; the computer-readable storage medium is used to store a plurality of instructions, the instructions are suitable for being loaded by the processor and executing the above-mentioned large capacity A reversible information hiding method for audio files.

In other embodiments, the following technical solutions are adopted:

A computer-readable storage medium stores a plurality of instructions, wherein the instructions are adapted to be loaded by a processor of a terminal device and execute the above-mentioned method for reversible information hiding of large-capacity audio files.

Compared with the prior art, the beneficial effects of the present invention are:

The present invention is based on a code division multiple access audio reversible information hiding method, firstly designing a target audio point prediction algorithm to form a prediction error matrix of the original audio file, and then using different orthogonal expansion sequences to realize the reversible information hiding of secret data; In the process, not only can the embedded confidential information be extracted losslessly, but also the original audio can be completely restored. Based on the mutual orthogonality of the extension sequences, the secret information is embedded into the audio carrier file for many times, and most of the elements of different extension sequences cancel each other during the information embedding process, which increases the reversible information embedding capacity of the carrier audio and improves the file. Audio fidelity capability.

Description of drawings

Fig. 1(a)-(f) respectively represent 6-segment test audio prediction error histograms in the embodiment of the present invention;

2 is a flowchart of a method for reversible information hiding of large-capacity audio files in the embodiment of the present invention;

3(a)-(f) respectively represent the embedding rate (BPP) and signal-to-noise ratio (SNR) of 6-segment audio files under different embedding strengths in the embodiment of the present invention;

4(a)-(f) respectively represent the embedding rate (BPP) and signal-to-noise ratio (SNR) of 6-segment audio files under extension sequences of different lengths in the embodiment of the present invention;

Fig. 5 is the SNR value of different audio files under 4 kinds of embedding rates in the embodiment of the present invention;

Figures 6(a)-(f) respectively show the test results under different embedding rates for 6 representative test audios in the database in the embodiment of the present invention.

Detailed ways

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components and/or combinations thereof.

Embodiments of the invention and features of the embodiments may be combined with each other without conflict.

Example 1

In one or more embodiments, a reversible information hiding method for large-capacity audio files is disclosed, including an embedding process and an extraction process, wherein the embedding process specifically includes:

Divide different audio points in the initial file into fork sets and point sets according to their positions;

Using the local consistency feature between adjacent audio points, the target audio point amplitude prediction algorithm is used to construct the first fork set prediction error matrix and the first point set prediction error matrix respectively;

The data to be hidden is divided into two equal halves, the first part is embedded in the fork set, and the second part is embedded in the point set;

According to the error prediction algorithm, the fork-set ciphered audio sequence and the point-set ciphered audio sequence are reversibly constructed; finally, the ciphered audio file is formed by merging.

The extraction process specifically includes:

Divide the encrypted audio files into fork sets and point sets;

According to the target error prediction algorithm, construct the second point set prediction error matrix and the second fork set prediction error matrix respectively;

Recover the attachment information embedded in the point set prediction error matrix and the cross set prediction error matrix respectively, extract the secret information embedded in the error matrix according to the code division multiple access reversible information embedding algorithm, and restore the original amplitude of the target audio point;

The extracted secret information is connected to reconstruct the hidden secret information sequence, and the original audio file is restored losslessly by using the original amplitude matrix of the restored target audio point.

The method disclosed in this embodiment will be described in detail below.

1. Audio Reversible Information Hiding Based on Code Division Multiple Access (CDMA)

Audio signal is a typical one-dimensional non-stationary signal, its autocorrelation function and mean function will change with time, and the signal has the characteristics of gradual change. The audio signal exhibits random noise characteristics in some time periods, periodic signal characteristics in other time periods, or a mixture of the two. In the actual processing process, the audio signal is generally first segmented segment by segment, and then the segment samples are processed. According to the continuous correlation characteristics of audio signals, the reversible information hiding and code division multiple access are combined, and the consistent correlation between adjacent audio samples is fully utilized, and the watermark information is embedded in the sample vector, which can realize the reversible information hiding of audio files.

Code Division Multiple Access (CDMA) is a spectrum spreading algorithm used for secure information transmission and channel multiplexing in wireless communications. The CDMA-based communication system is considered to be a very secure communication system. It transmits the information to be sent through a predetermined orthogonal spreading sequence carrier, and embeds the transmitted signal into different orthogonal spreading sequences for transmission. Only the spread sequence can correctly decode and extract the transmitted secret information, so the system has very high security; and based on the spread spectrum transmission characteristics of code division multiple access, the secret information can be transmitted in the same channel without interfering with each other, which greatly saves money frequency resources.

Since the reversible information hiding system realizes the transmission of secret information based on the carrier audio, the reversible information hiding system based on code division multiple access can be analogous to a covert communication system, in which the carrier is the open transmission channel, and the secret information is the content of the encoded transmission. In the reversible information hiding system based on code division multiple access, the rows (columns) of the Walsh Hardmard matrix are generally used to generate orthogonal spreading sequences, and the elements of the orthogonal sequences are composed of "1" or "-1". The number of "1" and "-1" is equal, so it also has the characteristics of zero mean. In the process of reversible information hiding transmission based on code division multiple access, an orthogonal spreading sequence is allocated to the sender, the binary bit "1" is represented by the sequence itself, and the binary bit "0" represents the corresponding negative sequence. The intersection characteristic and 0-mean feature can realize the superposition transmission of secret information. At the same time, in the process of information superposition transmission, the elements of different sequences cancel each other, which ensures the audio reversible information hiding system based on code division multiple access to achieve better security. and higher reversible information hiding performance.

1.1 Embedding information

Assuming the embedded secret information W=[b ₁ ,b ₂ ,b ₃ ,...,b _k ]( _bi ∈{1,0}, 1≤i≤n}, based on formula (1), the secret can be Information b _i is transformed into embedded watermark information ω _i .

ω _i = -cosπb _i (1)

The transformed secret information becomes W=[ω ₁ ,ω ₂ ,ω ₃ ,...,ω _k ](ω _i ∈{1,-1}, 1≤i≤k}, in the Walsh Hadamard matrix Select k rows to generate mutually orthogonal spreading sequences S _i ～S _k , and construct the spreading sequence matrix E:

Let the length of the sequence S _i ={t _i1 ,t _i2 ,t _i3 ,...t _il } (1≤i≤k) be l (even), where the number of "1" and "-1" is equal , the spreading sequence has the characteristics of zero mean and mutual orthogonality.

Let A be an initial audio sequence of length N, select the target samples in A to construct a vector matrix X:

如式(5)所示。Wherein, the vector vector V _j =[u ₁ , u ₂ , u ₃ , . . . , _{u l} ] (1≤j≤k), the length l is the same as the length of the extension sequence Si. According to formula (4), the watermark information is embedded into the carrier vector V _j , and the vector matrix after embedding the information

As shown in formula (5).

是嵌入信息后的第j个载体向量，ω₁,ω₂,…,ω_k是嵌入的秘密信息，S₁,S₂,…,S_k是正交扩展序列，k为嵌入信息位数，每个扩展序列都和一位嵌入信息相关。in formula (5)

is the j-th vector vector after embedding information, ω ₁ , ω ₂ ,…,ω _k is the embedded secret information, S ₁ , S ₂ ,…,S _k is the orthogonal expansion sequence, k is the number of bits of embedded information, Each spreading sequence is associated with a bit of embedded information.

When a large amount of watermark information (such as n·k bits) needs to be embedded, W _n =[ω ₁₁ ,ω ₁₂ ,ω ₃ ,····,ω _1k ,ω ₂₁ ,ω ₂₂ ,····,ω _2k ,·· ·· _ωn1 , _ωn2 ,····, _ωnk ], the watermark information sequence can be segmented and embedded in groups according to the form shown by W _M , as shown in equation (6). At the same time, based on Due to the orthogonality of the propagation sequence, the transformed watermark information sequence W _n can also be embedded into the same vector matrix X through multi-level embedding to achieve large-capacity reversible information embedding, as shown in equation (7).

为多次嵌入后的向量矩阵，n为嵌入的次数，n·k为嵌入信息的比特数，α为正整数，代表嵌入过程中的嵌入强度，α越大，可以嵌入的向量个数越多，嵌入容量越高，对样本的改变程度也会越大，相对应的也会带给音频较大的失真。where E ₁ ～E ₁ is the corresponding extended sequence matrix for each embedding,

is the vector matrix after multiple embeddings, n is the number of times of embedding, n·k is the number of bits of embedded information, α is a positive integer, representing the embedding strength in the embedding process, the larger the α, the more vectors that can be embedded , the higher the embedding capacity is, the greater the degree of change to the sample will be, and the corresponding audio will be distorted.

1.2 Extract information

嵌入的水印信息可以通过计算向量矩阵

和扩展序列矩阵E的内积提取出来，如式(8)所示：Let the vector matrix after embedding information be

The embedded watermark information can be calculated by the vector matrix

and the inner product of the extended sequence matrix E is extracted, as shown in formula (8):

When the ith secret information needs to be extracted, it can be calculated by the following formula (9)

的计算结果也是正整数，所以表达式

的正负由ω_i决定，由于ω_i的取值只有1和-1两种可能，因此当满足条件

嵌入的信息可以通过公式(10)进行提取：As shown in formula (9), α is a positive integer,

The result of the calculation is also a positive integer, so the expression

The positive or negative of ω _i is determined by ω i. Since there are only two possibilities for the value of ω _i , 1 and -1, so when the condition is satisfied

The embedded information can be extracted by formula (10):

Formulas (9) and (10) can be further transformed to obtain formula (11) to determine whether the carrier has information embedded.

将嵌入到载体向量V_j中的秘密信息ω_i提取出来，因为扩展序列S_i具有零均值的特点，

运算等同于计算载体向量V_j中相邻元素之间的差值的和。因此嵌入强度α的值越大、音频数据越平滑，可以嵌入到载体音频中的信息就越多。It can be seen from the above analysis that when formula (10) is satisfied, it can be obtained by

Extract the secret information ω _i embedded in the vector vector _{V j} , because the extended sequence Si has the characteristic of zero mean,

The operation is equivalent to computing the sum of the differences between adjacent elements in the vector vector _Vj . Therefore, the larger the value of the embedding strength α, the smoother the audio data, and the more information that can be embedded in the carrier audio.

时，载体向量V_j不满足可逆信息嵌入条件，为了区分不可嵌入信息向量，可以将一个伪信息位嵌入到不满足条件的向量中，在解码阶段通过提取向量中的伪位，判断一个向量是否嵌入了水印信息，具体处理如表1所示：On the other hand, if

When the carrier vector V _j does not meet the reversible information embedding condition, in order to distinguish the non-embeddable information vector, a pseudo-information bit can be embedded into the vector that does not meet the condition. In the decoding stage, the pseudo-bit in the vector is extracted to determine whether a vector is The watermark information is embedded, and the specific processing is shown in Table 1:

Table 1 for the vector processing that does not meet the conditions

1.3 Original Audio File Recovery

According to the orthogonal characteristic of the CDMA embedding sequence, after extracting the original watermark information from the embedded information carrier, the original carrier vector matrix X can be obtained by formula (11), and the original carrier audio can be reconstructed.

This algorithm utilizes the orthogonal characteristic of embedding vector to realize multi-level information superposition embedding to increase the reversible information embedding capability of carrier audio. Since the elements in the embedding vector Si only _consist of "1" and "-1", most of the elements in the embedding vector cancel each other out in the multi-level information embedding process. Therefore, even in the case of increasing the data embedding capacity, this algorithm can A high listening quality can still be obtained.

1.4 CDMAA specific calculation example

令扩展序列矩阵

嵌入强度α＝1。待嵌入的秘密信息W＝[1,0]，根据公式(1)得出将秘密信息转化为对应位水印信息序列W＝[1,-1]进行嵌入。在嵌入过程中，先计算

和

的值，满足条件

待嵌信息ω₁＝1，通过

嵌入到向量V₁中，嵌入信息后的向量变为

同理，可将水印信息ω₂＝-1嵌入到向量V₂中，修改后的向量矩阵

二次嵌入信息后的载体向量变为A′＝[16,14,17,15,15,13,12,16]。当接收方获取嵌入信息的载体向量后，通过sign(<V_j·S_i>)分别提取出嵌入到两个向量中的水印信息1和-1，并采用

计算出向量V，构建出原始向量矩阵

得到初始样本序列A＝[15,15,16,16,14,14,13,15]。Set the audio carrier sample A = [15, 15, 16, 16, 14, 14, 13, 15], and construct a vector matrix based on it

Let the extended sequence matrix

Embedded strength α=1. The secret information to be embedded is W=[1,0], according to formula (1), it is obtained that the secret information is converted into the corresponding bit watermark information sequence W=[1,-1] for embedding. In the embedding process, first calculate

and

value, which satisfies the condition

The information to be embedded ω ₁ =1, through

Embed into vector V ₁ , the vector after embedding information becomes

Similarly, the watermark information ω ₂ =-1 can be embedded into the vector V ₂ , the modified vector matrix

The vector vector after the secondary embedding information becomes A′=[16, 14, 17, 15, 15, 13, 12, 16]. When the receiver obtains the carrier vector of the embedded information, it extracts the watermark information 1 and -1 embedded in the two vectors through sign(<V _j · S _i >), and uses

Calculate the vector V and construct the original vector matrix

The initial sample sequence A=[15, 15, 16, 16, 14, 14, 13, 15] is obtained.

嵌入向量矩阵

将W₁嵌入到向量矩阵X中，经过嵌入后向量矩阵转变为

再将W₂嵌入到向量矩阵

中，最终得到嵌秘向量矩阵

在解码端，嵌入的水印信息和原始向量矩阵可以按照如下方法进行恢复(如表2所示)：When the amount of information to be embedded is large and multi-level embedding is required, the embedded information can be superimposed according to the following method. Assuming that the secret information sequence to be embedded is W=[1,0,1,1], the secret information is converted into the corresponding bit information sequence W=[1,-1,1,1] by formula (1), and the to-be-embedded information is constructed. information matrix

Embedding vector matrix

Embed W ₁ into the vector matrix X, after embedding, the vector matrix is transformed into

Then embed W ₂ into the vector matrix

, and finally get the embedded secret vector matrix

At the decoding end, the embedded watermark information and the original vector matrix can be restored as follows (as shown in Table 2):

Table 2 Detailed information extraction stages of multi-level information extraction

嵌入两位比特信息后，初始向量V₁中只有两个元素值发生改变，两个扩展序列的第一个和第四个元素经过两次信息嵌入后相互抵消；对于初始向量V₂与嵌入后的

两个扩展序列的第二个和第四个元素经过两次信息嵌入后相互抵消。实际上，由于不同扩展序列的大部分元素在多级信息嵌入中是相互抵消的，使得本方法能够在大容量可逆信息嵌入的同时，将失真抑制在较低水平。In this process, for the initial vector V ₁ and the embedded

After embedding two bits of information, only two element values in the initial vector V ₁ change, and the first and fourth elements of the two extended sequences cancel each other after two information embeddings; for the initial vector V ₂ and the embedded of

The second and fourth elements of the two spreading sequences cancel each other after two information embeddings. In fact, since most elements of different extension sequences cancel each other in multi-level information embedding, this method can suppress distortion at a low level while embedding large-capacity reversible information.

In the reversible information hiding method for large-capacity audio files disclosed in this embodiment, a carrier vector is first constructed;

时，可以在向量V_j中嵌入信息。由于嵌入向量中1和-1个数相同且分布均匀，计算

等于对向量V_j中相邻元素的差值进行求和，所以，载体向量V_j中的元素值越相近，

会更小。即载体向量中的元素值越近似，满足嵌入条件的载体向量越多，信息嵌入生成的音频失真越小，音频文件的嵌入能力越强。It can be seen from the above analysis that when the vector vector meets the conditions

, information can be embedded in the vector V _j . Since the number of 1 and -1 in the embedded vector is the same and evenly distributed, computing

is equal to summing the difference values of adjacent elements in the vector V _j , so the closer the element values in the vector vector V _j are,

will be smaller. That is, the more similar the element values in the carrier vector, the more carrier vectors that satisfy the embedding conditions, the smaller the audio distortion generated by the information embedding, and the stronger the embedding ability of the audio file.

Therefore, according to the time-series correlation of audio files, using the local consistency characteristics between adjacent audio points, the target audio point amplitude prediction algorithm is used to construct the target audio point amplitude prediction error carrier matrix, and the carrier vector is formed to realize the audio frequency. File reversible information embedding can effectively improve the reversible information hiding performance of audio files. According to the characteristics of the audio file, the numerical values between adjacent audio sample points have strong similarity. In this embodiment, the method of nearest neighbor mean prediction is used to predict the target audio point, resulting in a prediction error, as shown in formula (12):

where x′ _i is the predicted value of the sample point x _i , and x _i-1 and x _i+1 are the samples on the left and right side of the sample point x _i , so the prediction error d _j =x′ _j -x _j , due to the audio Adjacent sample points are closely related, and this algorithm can generate a small prediction error, and finally form a steep histogram closely distributed around the "0" value (as shown in Figure 1). It is guaranteed that the vector elements constructed based on the prediction error matrix have smaller numerical values and consistent similarity characteristics, which effectively improves the performance of reversible information hiding of the carrier audio files.

Figure 1(a)-(f) are the prediction error histograms of Clip27, Clip40, Clip51, Clip58, Clip66 and Clip70, respectively. From the histograms, it can be clearly observed that the prediction errors based on this algorithm are closely distributed around the "0" value Therefore, the carrier matrix generated based on the error prediction algorithm realizes reversible information embedding, which can maintain a relatively low audio distortion rate while achieving higher information embedding capacity.

embedding process

The schematic diagram of the embedding and extraction process of information is shown in Figure 2, and the description of the embedding process and extraction process is as follows:

1) Divide the audio file content into two groups in order: odd-numbered audio points (point set) and even-numbered audio points (fork set). According to the close correlation of adjacent audio points in the audio file, the mean prediction algorithm is firstly used to predict the even-numbered audio points by using the odd-numbered points on both sides of them, and the error between the actual amplitude of the audio point and the prediction is calculated, and the cross-set error is generated. matrix; thereafter, the same strategy is used to predict odd audio points and generate a point set prediction error matrix. The data to be hidden is divided into two equal halves, the first part is embedded in the fork set and the second part is embedded in the point set.

2) The sample sequence in the fork set is divided into an embedded area E and a reserved area S, and each audio point is used to carry secret information only in the reserved embedded area E, and each audio word in the reserved reserved area S is used. The last three bits of the section (the least significant bit LSB) store additional information in the information embedding process, such as embedding sequence, embedding strength, and the quantity of embedding information. The length of the information in the data storage area S is preset to 4096 binary bits, of which the first 12 bits are used to represent the length of the storage area S and the identifier is set to determine the location of the storage area, and the other bits are used to store additional information. The last three bits of original information of each audio byte in the S area are combined with the secret information to form the information stream to be embedded.

3) Similarly, the same strategy is used to embed another part of the secret information to be embedded into the error matrix corresponding to the point set.

4) The auxiliary information required by the reversible information hiding algorithm based on code division multiple access is embedded into the reserved area through LSB replacement.

5) Based on the error matrix after the embedded information and the value of the target audio point, the embedded secret audio sequence is reversibly constructed according to the error prediction algorithm. According to the error prediction sequence, firstly construct the fork set embedded cipher audio sequence, then construct the point set cipher audio sequence, and finally merge to form the ciphered audio file.

Among them, the value of the target audio point is: based on the prediction algorithm, the predicted value of the current target sample is obtained by using the linear combination of the amplitudes of the audio samples before and after, and the predicted value is the amplitude obtained by the prediction algorithm, which is similar to the actual amplitude of the sample. , using the error value generated by the predicted value and the true value to embed the data.

Using the point set to predict the samples of the fork set, the predicted sample value is obtained, the prediction error is calculated, and the secret information is embedded into the prediction error through the reversible information hiding algorithm based on CDMA, and the prediction error changed after the embedded information is obtained. The prediction error is added to the predicted value to obtain the sample value after the embedded information, and then the cross-set embedded secret audio sequence is obtained.

Extraction process

After receiving the encrypted audio file, the receiver extracts the embedded secret information losslessly and restores the original audio file according to the following steps:

1) First, the embedded audio samples are divided into fork sets and point sets. According to the reverse order of information embedding, firstly, the data in the cross set and the target audio points in the point set are used to predict the target sample points according to the target error prediction algorithm to construct a prediction error matrix. Extract the last three (LSB) values of audio points in the reserved area in the prediction error matrix of the point set, and restore the attached information embedded in the information. According to the additional information, the secret information embedded in the error matrix is extracted according to the code division multiple access reversible information embedding algorithm, and the original amplitude value of the target audio point is restored.

2) According to the same method, extract the secret information embedded in the fork set, and restore the original amplitude value of the target audio point.

3) Using the recovered data in the point set and the fork set, connect the extracted secret information to reconstruct the hidden secret information sequence, and use the restored original audio point amplitude matrix to restore the original audio file losslessly.

The method of this embodiment is verified and analyzed by experiments below.

This experiment uses a test database containing 70 standard audio files. The sampling frequency of the audio files is 44.1KHz and the number of sampling bits is 16. The test files include synthetic signals, single instrument, vocal music, language and other types. The audio files with different audio signal characteristics are selected for experimental results, and the detailed performance of the proposed algorithm has been verified. And two parameters of embedding strength and extended sequence length are used for performance evaluation.

1. Experimental results of different embedding strengths

The experiment first selects 6 test audios with different audio characteristics including Clip27, Clip40, Clip51, Clip58, Clip66 and Clip70 to verify the algorithm performance. The selected 6 audio segments are single-instrumental treble, single-instrumental bass, Chinese voice, solo instrument, orchestral music and pop music, so the feasibility of the proposed algorithm can be verified from different aspects. The experimental results are described by the SNR-BPP relationship curve. The signal-to-noise ratio (SNR) and the embedding rate (BPP) are the capacity standard parameters to measure the audio distortion and reversible information embedding respectively, which is beneficial to compare the performance with other algorithms. Figure 3(a)-(f) respectively show the relationship between the signal-to-noise ratio (SNR) and the embedding rate (BPP) of the 6-segment audio files under different embedding strengths. Due to the different lengths of the audio files selected in the experiment, the shortest is 18 seconds and the longest is 2 minutes and 17 seconds. At the same time, due to the inconsistency of different audio types, the waveform trends are also quite different. In order to better verify the performance of the algorithm, This paper adaptively selects reversible embedding information with different embedding strengths for different audio files, and examines the relationship between the embedding rate and the signal-to-noise ratio of audio files when different audio files have different embedding strengths α.

As can be seen from the 6 graphs in Fig. 3(a)-(f), as the embedding rate increases, the SNR of different audio files decreases accordingly. As shown in Figure 3(a), when the embedding capacity is 0.1 BPP, the SNR corresponding to the embedding strength α=14 is 40.12 dB, and the corresponding SNR is 38.14 dB when the embedding strength α=16. The experimental results show that: under the same embedding capacity, with the increase of the embedding strength α, the SNR decreases, and the quality of the audio file decreases significantly. This is because with the increase of the embedding strength, there are more audio points that can be moved in and adjusted, and a large amount of information can be embedded into the audio file at one time; at the same time, the degree of modification of the original audio file by the information embedding increases, and the The distortion of the file is obvious. However, when the embedding strength is small, there are not many carrier vectors that meet the reversible information embedding conditions, and the capacity of a single information embedding is small. In order to increase the embedding capacity, the information needs to be embedded multiple times. In the process of multiple information embedding, based on the characteristics of code division multiple access, the elements of different orthogonal sequences cancel each other in the process of superimposing and embedding. With the increase of the embedding capacity, the quality degradation curve of the original audio file becomes slower, ensuring the original audio file. Distortion is reduced. Therefore, in the process of reversible information embedding of audio files based on code division multiple access, when the embedding capacity is large, the appropriate reversible information embedding strength is used to perform multiple overlapping information embedding, which can make the embedded encrypted file obtain better reversible information embedding. performance, while ensuring that the original audio file has less distortion while a large amount of secret information is embedded. As shown in Fig. 3(b), when the embedding strength α=4 and the embedding capacity is 0.1BPP, the SNR is 53.12dB, and when the embedding capacity increases to 0.2BPP, the SNR of the carrier audio drops to 47.52dB; however, when the embedding strength When it is 6, the SNR corresponding to the embedded capacity of the carrier audio is 46.1dB at 0.1BPP, and the corresponding SNR is 44.8dB at 0.2BPP. In this case, multiple embeddings with smaller embedding strength can achieve better reversible information embedding performance.

2. Experimental results of extended sequences of different lengths

Spreading sequences of different lengths also have an impact on the performance of the algorithm. As shown in Figure 4(a), when the sequence length is 2, the maximum capacity of a single information embedding is 0.13bpp; but when the sequence length is 4, the maximum capacity of a single embedding is only 0.087bpp, and the propagation of shorter lengths The modification of the sequence to the audio carrier file is small (when the length of the extended sequence is 2, the value of two audio prediction error points needs to be modified to embed 1-bit information; and when the length of the extended sequence is 4, the embedded 1-bit information needs to be modified. value of the four prediction error points). At the same time, according to the close correlation between adjacent audio points, the shorter carrier vector is easier to satisfy the condition of reversible information embedding of the code division multiple access algorithm than the long carrier vector. In order to deeply explore the correlation between adjacent audio points, this embodiment explores the influence of extended sequences of different lengths on the performance of the algorithm by constructing 4 vector vectors of different lengths.

It can be seen from Figure 4(a)-(f) that when the information embedding capacity is not large, the embedding performance of the short extension sequence is obviously better than that of the longer extension sequence. When the length of the extension sequence is 8 (1 × 8 carrier audio prediction error points), the embedding performance drops significantly compared to extended sequences of length 2. The reasons are analyzed as follows: on the one hand, 8 samples need to be modified for each embedded 1 bit of information using the extended sequence of length 8; The length of the condition is 1 × 8 and the number of vector vectors is small, so when the embedding capacity is not high, the SNR-BPP curve is lower than that of the remaining three extended sequences with shorter lengths. However, as the payload increases, multi-level superposition information embedding is required to expand the embedding capacity, most elements of different extension sequences cancel each other during the reversible information embedding process, and the difference in embedding performance of extension sequences with different lengths gradually decreases.

It can be seen from Fig. 4(a)-(f) that with the increase of the embedding capacity, the difference of the signal-to-noise ratio (SNR) of the embedded dense carrier with extended sequence lengths 4 and 2 gradually decreases, especially for the audio files Clip 40 and Clip 66 , when the embedding capacity is greater than 0.9, the embedding performance of the extended sequence of length 4 is better than that of the extended sequence of length 2. This is because the correlation between 4-bit adjacent audio points in the audio file is still relatively strong, and there are many carrier vectors that satisfy the reversible information embedding condition of CDMA, and the elements in the carrier vector cancel each other after multiple embeddings. The number is better than the extended sequence of length 2. Therefore, in the process of large-capacity reversible information embedding, the optimal reversible information embedding performance can generally be obtained when the length of the vector vector is 4. However, when the lengths of the extension sequences are 6 and 8, as shown in Fig. 4(a)-(f), the carrier audio SNR drops rapidly with the increase of the embedding capacity, which is much lower than that of the extension sequences of length 4. Embedded performance. This is due to the characteristics of one-dimensionality and instantaneous mutation of audio files, so that longer extension sequences need to correspond to larger embedding strengths to meet the conditions of reversible information embedding. With the increase of embedding capacity, longer extension sequences are in The embedded information will cause the original audio to be greatly deformed, and the audio quality will decrease significantly.

In order to further explore the degree of audio file distortion of the reversible information hiding algorithm based on code division multiple access under high embedding capacity, this paper tests all 70 audio files in the database when the embedding capacity is 1.0BPP, 1.1BPP, 1.2BPP and 1.3BPP. Embedding distortion rate, the SNR values of different audio files under 4 embedding rates are shown in Figure 5.

It can be seen from Figure 5 that when the embedding rate is greater than 1.0BPP, the algorithm in this paper can still ensure that the original audio file has a higher SNR value. For most audio files, even when the embedded capacity reaches 1.3BPP, the sound distortion can still be kept low. This is because in the process of multi-level embedding of the reversible information hiding algorithm based on code division multiple access, due to the orthogonality of the spreading sequence, most of the elements in the embedding vector cancel each other out, which locks the increase of the embedding capacity and reduces the degree of the SNR curve. Gradually transformed, the original audio file retains high sonic fidelity even while bulk reversible information is embedded.

3. Compare with other methods

In order to further evaluate the performance superiority of this method, this algorithm is compared with the other two latest algorithms for reversible information hiding of high-performance audio files. In this embodiment, the above two latest algorithms are referred to as Ref. [30] algorithm and Ref.[32] algorithm; among them, the Ref.[30] algorithm uses the audio samples on the left and right sides to predict the value of the target audio point, and uses the histogram translation technology to embed effective information, the algorithm consumes a short time, especially in high embedding The quality of the original audio file degrades rapidly when the capacity is increased. The Ref.[32] algorithm first calculates the distance between the prediction point and the target audio point, and determines the prediction coefficient according to the distance to reduce the value of the prediction error of the target audio point, and then uses the difference expansion algorithm to achieve reversible information embedding. The information hiding ability of the target audio sample is improved, but when the embedding capacity is large, the audio file will still be seriously distorted.

In this experiment, SNR-BPP is still used as the evaluation standard to measure the reversible information hiding performance. Test results at rate:

As can be seen from Figure 6(a)-(f), the reversible information hiding performance of this scheme does not have much advantage compared with the other two algorithms in the case of low load. However, with the increase of payload, the superiority of this algorithm for large-capacity information hiding is also shown. In the audio files Clip27, Clip40, Clip66 and Clip70, when the embedding rate is greater than 0.38bpp, 0.56bpp, 0.39bpp and 0.40bpp, respectively, the SNR value of this algorithm begins to surpass the other two reversible information hiding algorithms for audio files. For the audio files Clip51 and Clip58, when the embedding rate reaches 0.95bpp and 0.79bpp, the performance of this algorithm begins to surpass the other two algorithms. The experimental results show that the performance of this algorithm is significantly better than other algorithms in large-capacity reversible information hiding. The reasons are as follows: On the one hand, when the embedding load is small, the algorithm needs to modify at least two samples to embed one information bit, so the performance of the algorithm is average compared to the other two algorithms under low embedding capacity. However, with the increase of the embedding capacity, the algorithm can use multi-level embedding to expand the embedding capacity, most of the elements of different extension sequences cancel each other, and the distortion of the original audio file is significantly reduced compared with other algorithms. Therefore, the present scheme has a higher signal-to-noise ratio (SNR) in the case of medium to high embedding capacity. On the other hand, as shown in Figure 1(c) and Figure 1(d), since the audio Clip51 and Clip58 change drastically, their prediction error histograms are evenly distributed on the left and right, and a large embedding strength is required to realize the secret information. Embedding, information embedding leads to large distortion of the original file, so the performance of this algorithm can gradually surpass the other two algorithms after multiple times of information superposition and embedding, which is consistent with the above theoretical analysis.

Embodiment 2

In one or more embodiments, a reversible information hiding system for large-capacity audio files is disclosed, including:

Divide the different audio points of the initial file into fork sets and point sets according to their positions;

A device for constructing a first fork set prediction error matrix and a first point set prediction error matrix respectively by utilizing the local consistency feature between adjacent audio points and adopting the target audio point amplitude prediction algorithm;

A means for dividing the data to be hidden into equal halves, the first part is embedded in the fork set, and the second part is embedded in the point set;

A device for reversibly constructing a fork-set embedded cipher audio sequence and a point-set cipher audio sequence according to an error prediction algorithm; and finally merging to form a ciphered audio file.

Also includes:

means for dividing the encrypted audio file into fork sets and point sets;

A device for constructing the second point set prediction error matrix and the second cross set prediction error matrix respectively according to the target error prediction algorithm;

A device for restoring the attachment information embedded in the point set prediction error matrix and the cross set prediction error matrix respectively, extracting the secret information embedded in the error matrix according to the code division multiple access reversible information embedding algorithm, and restoring the original amplitude value of the target audio point;

The device is used to connect the extracted secret information to reconstruct the hidden secret information sequence, and use the restored original amplitude matrix of the target audio point to restore the original audio file losslessly.

The specific working process of the above device adopts the method disclosed in the first embodiment, and is not repeated here.

Embodiment 3

In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor executing the The program implements the reversible information hiding method for a large-capacity audio file in the first embodiment. For brevity, details are not repeated here.

It should be understood that, in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate array FPGA or other programmable logic devices , discrete gate or transistor logic devices, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory may include read-only memory and random access memory and provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.

In the implementation process, each step of the above-mentioned method can be completed by a hardware integrated logic circuit in a processor or an instruction in the form of software.

The method for reversible information hiding of a large-capacity audio file in the first embodiment may be directly embodied in the execution of a hardware processor, or executed in a combination of hardware and software modules in the processor. The software modules may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art. The storage medium is located in the memory, and the processor reads the information in the memory, and completes the steps of the above method in combination with its hardware. To avoid repetition, detailed description is omitted here.

Those of ordinary skill in the art can realize that the unit, that is, the algorithm step of each example described in conjunction with this embodiment, can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementations should not be considered beyond the scope of this application.

Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they do not limit the scope of protection of the present invention. Those skilled in the art should understand that on the basis of the technical solutions of the present invention, those skilled in the art do not need to pay creative work. Various modifications or deformations that can be made are still within the protection scope of the present invention.

Claims (7)

1. a large-capacity audio file reversible information hiding method, is characterized in that, comprises embedding process: Divide the different audio points of the initial file into fork sets and point sets according to their positions; The fork set is an even-numbered audio point, and the point set is an odd-numbered audio point; Using the local consistency feature between adjacent audio points, the target audio point amplitude prediction algorithm is used to construct the first fork set prediction error matrix and the first point set prediction error matrix respectively; The data to be hidden is divided into two equal halves, the first part is embedded in the fork set, and the second part is embedded in the point set; According to the error prediction algorithm, the fork set embedded secret audio sequence and the point set embedded secret audio sequence are reversibly constructed; finally, the secret audio file is formed by merging; Constructing a fork set prediction error matrix, specifically: using the odd-numbered points on both sides of the even-numbered audio points to predict the even-numbered audio points, calculating the error between the actual amplitude of the even-numbered audio points and the prediction, and generating a first fork set error matrix; Constructing a point set prediction error matrix, specifically: using the even-numbered points on both sides of the odd-numbered audio points to predict the odd-numbered audio points, calculating the error between the actual amplitude of the odd-numbered audio points and the prediction, and generating the first point set error matrix; According to the error prediction algorithm, the fork set-embedded audio sequence is reversibly constructed, specifically: The sample sequence in the fork set is divided into an embedded area E and a reserved area S, and each audio point is used to carry secret information only in the reserved embedded area E, while the back of each audio point in the reserved save area S is used to carry secret information. The additional information in the information embedding process is stored in the three bits; the last three original information of each audio point in the S area is formed with the secret information to form the information stream to be embedded; The auxiliary information required by the code division multiple access-based reversible information hiding algorithm is embedded in the reserved area S through LSB replacement; Based on the error matrix after embedding information and the value of the target audio point, according to the error prediction algorithm, the cross-set embedded cipher audio sequence is reversibly constructed. 2. a kind of large-capacity audio file reversible information hiding method as claimed in claim 1, is characterized in that, adopts the method of nearest neighbor mean value prediction to predict target audio frequency point, produces prediction error, is specially:

Among them, x' _i is the predicted value of the sample point x _i , and x _i-1 and x _i+1 are the samples on the left and right sides of the sample point x _i , so the prediction error d _i =x' _{i -xi} . 3. a kind of large-capacity audio file reversible information hiding method as claimed in claim 1, is characterized in that, Also includes the extraction process: Divide the encrypted audio files into fork sets and point sets; According to the target error prediction algorithm, construct the second point set prediction error matrix and the second fork set prediction error matrix respectively; Recover the attachment information embedded in the point set prediction error matrix and the cross set prediction error matrix respectively, extract the secret information embedded in the error matrix according to the code division multiple access reversible information embedding algorithm, and restore the original amplitude of the target audio point; The extracted secret information is connected to reconstruct the hidden secret information sequence, and the original audio file is restored losslessly by using the original amplitude matrix of the restored target audio point. 4. a large-capacity audio file reversible information hiding system, is characterized in that, comprises: A device for dividing different audio points in the initial file into fork sets and point sets according to their positions; The fork set is an even-numbered audio point, and the point set is an odd-numbered audio point; A device for constructing a first fork set prediction error matrix and a first point set prediction error matrix respectively by utilizing the local consistency feature between adjacent audio points and adopting the target audio point amplitude prediction algorithm; A means for dividing the data to be hidden into equal halves, the first part is embedded in the fork set, and the second part is embedded in the point set; For reversibly constructing the fork set embedded encrypted audio sequence and the point set embedded encrypted audio sequence according to the error prediction algorithm; finally merge to form the device of the encrypted audio file; Constructing a fork set prediction error matrix, specifically: using the odd-numbered points on both sides of the even-numbered audio point to predict the even-numbered audio point, calculating the error between the actual amplitude value of the even-numbered audio point and the prediction, and generating the first fork set error matrix; Constructing a point set prediction error matrix, specifically: using the even-numbered points on both sides of the odd-numbered audio points to predict the odd-numbered audio points, calculating the error between the actual amplitude of the odd-numbered audio points and the prediction, and generating the first point set error matrix; According to the error prediction algorithm, the fork set-embedded audio sequence is reversibly constructed, specifically: The sample sequence in the fork set is divided into an embedded area E and a reserved area S, and each audio point is used to carry secret information only in the reserved embedded area E, while the back of each audio point in the reserved save area S is used to carry secret information. The additional information in the information embedding process is stored in the three bits; the last three original information of each audio point in the S area is formed with the secret information to form the information stream to be embedded; The auxiliary information required by the code division multiple access-based reversible information hiding algorithm is embedded in the reserved area S through LSB replacement; Based on the error matrix after embedding information and the value of the target audio point, according to the error prediction algorithm, the cross-set embedded cipher audio sequence is reversibly constructed. 5. a kind of large-capacity audio file reversible information hiding system as claimed in claim 4, is characterized in that, also comprises: means for dividing the encrypted audio file into fork sets and point sets; A device for constructing the second point set prediction error matrix and the second fork set prediction error matrix respectively according to the target error prediction algorithm; A device for restoring the attachment information embedded in the point set prediction error matrix and the cross set prediction error matrix respectively, extracting the secret information embedded in the error matrix according to the code division multiple access reversible information embedding algorithm, and restoring the original amplitude value of the target audio point; The device is used to connect the extracted secret information to reconstruct the hidden secret information sequence, and use the restored original amplitude matrix of the target audio point to restore the original audio file losslessly. 6. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; the computer-readable storage medium is used to store a plurality of instructions, wherein the instructions are suitable for being loaded by the processor And execute the reversible information hiding method for large-capacity audio files described in any one of claims 1-3. 7. A computer-readable storage medium, wherein a plurality of instructions are stored, wherein the instructions are adapted to be loaded by the processor of the terminal device and execute the large-capacity audio file according to any one of claims 1-3 Reversible information hiding method.

Images