Early use of block DCT watermarking technology, their watermarking scheme is to use a key to randomly select some blocks of the image, slightly change a triplet in the middle frequency domain to hide the binary sequence information. This method is robust to lossy compression and low pass filtering. Cox et al. [proposed the famous digital watermarking technology based on image global transformation, which performs discrete cosine transform (DCT) on the whole image. Barni et al. Proposed a DCT watermarking algorithm based on HVS masking characteristics. Scanned the DCT coefficients according to Zig – Zag rearrange for one dimensional vector, leaving the vector L don’t modify a coefficient, starting on the first L a coefficient M behind a coefficient modification to embed watermark. Ji-wu huang et al In DC and AC components of DCT coefficient, on the basis of qualitative and quantitative analysis, points out that the DC component than AC component is more suitable for embedding the watermark, The watermark embedded with DC component has better robustness, and an adaptive algorithm using DC component is proposed.
Digital Watermarking technology is to directly embed some marking information into Digital carriers (including multimedia, documents, software, etc.) or indirectly represent (modify the structure of a specific area), and does not affect the use value of the original carrier, and is not easy to detect and modify again. But it can be identified and identified by the manufacturer. These hidden information in the carrier can be used to confirm the creators and buyers of the content, transmit secret information or judge whether the carrier is tampered with. Digital watermarking is an important research direction of information hiding technology. Digital watermarking is an effective way to realize copyright protection and an important branch of information hiding technology. In this paper, several important algorithms of digital watermarking are briefly introduced, and an algorithm based on discrete wavelet transform (DWT) is proposedaudioDigital watermarking model and some experimental results are given.
1 Audio Watermarking
The main application of audio digital watermarking is concealed communication and copyright protection. Covert communication focuses on information concealment and data embedding capacity while copyright protection emphasizes robustness. At present, most of the watermarking techniques used in audio digital copyright protection are limited to non-compression domain, including time domain and transform domain. The time domain algorithm mainly includes LSB algorithm and echo algorithm, while the transform domain algorithm mainly adopts DCT, DFT and DWT.
The more important algorithms of audio embedding digital watermarking are as follows:
(1) LsB-LESat Significant Bit embedding is one of the simplest embedding methods. Any watermark can be converted into a binary stream. Each sample of the audio file is also represented by a binary value. In this way, the watermark can be embedded in the audio signal by replacing the least significant bit (usually the least significant) of each sample value with a binary bit representing the watermark. If the audio signal is regarded as the channel of watermark transmission, and the watermark is regarded as the signal transmitted in the channel, ideally, the channel capacity will be 1Kbps/kHz, that is, the sampling rate and the bit rate are numerically equivalent. Pseudo random sequences can be generated by pseudo random sequence generators. When the pseudo-random sequence generator has a fixed structure, different initial values will generate different pseudo-random sequences, so the receiver and receiver only need to secretly transmit an initial value as the key instead of the whole pseudo-random sequence value. In order to enhance the robustness of watermark, it can be considered to add watermark to the high frequency component of audio data.
LSB method is simple, large data capacity and high security. The disadvantage is the poor robustness of counter signal processing.
(2) Spread Sprectrum Encoding. This method encodes the flow of information by distributing the encoded data over as much spectrum as possible. Direct sequence spread spectrum coding (DSSS) is commonly used, which is usually combined with the excellent performance of m sequence encoding and decoding. In order to take advantage of the masking effect of HAS, it is generally necessary to perform several levels of filtering on the sequence used, and watermark detection is combined with correlation hypothesis testing. This method is robust to MP3 audio encoding, PCM quantization and additional noise. Spread spectrum method has good anti-interference performance, strong concealment, small interference, easy to realize code division multiple access, digital-analog compatibility.
(3) phase coding. The human auditory system is insensitive to absolute phase and sensitive to relative phase. The reference phase representing watermark information is used to replace the absolute phase of the original audio segment, and the other audio segments are adjusted to keep the relative phase unchanged. The coding steps are described as follows:
⑥ According to the modified phase matrix and the original amplitude matrix, IDFT inverse transformation is carried out to generate audio signal containing watermark.
(4) Echo hiding. The watermark data is embedded into the audio signal by introducing echo, which makes use of another feature of HAS: the backward shielding function of audio signal in the time domain, that is, the weak signal is shielded after the disappearance of the strong signal, and it will continue to act within 50ms ~ 200ms after the disappearance of the strong signal without being detected by human ear.
Because echo hiding is to embed watermark information into the carrier data as the environment rather than random noise, it has satisfactory robustness to some lossy compression algorithms.
(5) Transform domain algorithm: transform domain algorithm has many advantages that spatial domain algorithm does not have, among which the most prominent point is the robustness of its algorithm. Transform domain algorithms include discrete Fourier transform (DFT), discrete cosine transform (DCT) and discrete wavelet transform (DWT). For the first two methods, a lot of research HAS been done at home and abroad. The basic idea is to transform the original audio data in a certain frequency domain combined with the auditory characteristics of HAS, and then change the corresponding transformation coefficients to embed the watermark. The algorithm in this paper is based on the third transform, namely discrete wavelet transform. Here’s a quick look at DWT technology.
DWL algorithm uses the original audio of Daubechies-4 wavelet basis to carry out wavelet decomposition of L level, reserving the differential components of the former L-1 level and processing the detail components of L level and embedding watermark. One of the characteristics of this algorithm is that the watermark signal is placed in the low-frequency part where the energy of speech signal is most concentrated.
Human auditory model
The response of the human auditory system to input signals is frequency based, and the difference in pitch corresponds to the change in frequency. Figure 1 shows the sensitivity of the human ear as a function of frequency. The figure shows the lowest sound intensity that can be heard by the human ear, which is exactly the reciprocal of the audio sensitivity for each different frequency. As can be seen from Figure 1, human ear is most sensitive to the frequency around 3kHz, while for the frequency too high (20kHz) and too low (20Hz), human ear sensitivity will decrease.
According to this feature, it can be reasonably expected that the quality of the original audio will not be damaged if the watermark is embedded in the appropriate high-frequency or low-frequency components of the audio data. This can be verified by subsequent experiments.
Three numerical method
This algorithm adopts DWT, including watermark embedding, watermark detection and watermark attack. The working principle of watermarking is shown in Figure 2. Watermark detection requires raw audio data.
3.1 Watermark embedding algorithm
(1) Scramble the image to be embedded with watermark. This algorithm simply uses pseudorandom algorithm to eliminate the correlation of data. (2) The original audio data are decomposed into multi-scale one-dimensional data, and the low-frequency coefficients and three-layer high-frequency coefficients are extracted respectively. In order to obtain better robustness, the watermark data is embedded into the third layer high-frequency component of the audio data. (3) Embed watermark data according to the formula Vw (I) =V (I) + (α+ E) ×W (I). Where V (I) is the audio data bit, W (I) is the watermark data bit, Vw (I) is the audio data bit after embedding the watermark, α is the watermark embedding strength, and e value is the correction value of 10-20. Through experiments, it is found that the watermark effect is ideal when α value is 0.004. (4) Perform IDWT transformation on the audio embedded with watermark data, that is, get the audio data containing watermark.
3.2 Watermark detection algorithm
(1) Multi-scale one-dimensional decomposition is carried out on the audio data containing watermark, and its high-frequency component coefficients of three layers are extracted.
(2) The detection algorithm is the reverse process of the embedded algorithm, requiring original audio data to participate in the detection, expressed as:
W (I) = (Vw (I) -v (I))/(α+e), where the values of α and e are consistent with those determined in the embedding algorithm.
(3) W (I) obtained in step (2) is the extracted one-dimensional watermark information sequence, which is processed in ascending dimension to obtain the two-dimensional image form. The result is the watermark detected in the output.
The original image embedded with the watermark and the extracted watermark after the first embedding are shown in FIG. 3 and FIG. 4 respectively.
4 Part tests
In order to test the performance of this watermarking system, various kinds of attacks are carried out on watermarked audio data, and some experimental results are given here.
Definition:
Nc is used to measure the similarity between the extracted watermark image and the original watermark image. The similarity between the watermark directly extracted from the unattacked audio with watermark image and the original watermark image is as high as 0.9998.
(1) Forced selection experiment. The testers who did not know the exact original audio signal were asked to identify the original audio by playing the original audio and the watermarked audio respectively. According to the conclusion of L. Bononey et al. [5], if the proportion of type ii audio is roughly the same as that of the original audio, it can be considered that the embedding of watermark does not cause significant difference in human ear perception. In the experiment, 8 students from the same lab were randomly selected. By embedding watermarks in different WAV files and asking randomly, 53.4% of the respondents thought that the original audio had better sound quality. The watermark embedded in the system does not cause significant change in the quality of the original audio.
(2) cut off the audio n/10 of all data (n= 1,2,3……) , the length of the original audio data bit is slightly more than 40 000, and the clip starts from the 20 000 bit.
According to Figure 5, Figure 6 and Figure 7, it can be seen that obvious watermark patterns can still be extracted after about one-third of the audio content is cut out. If the shear part is too much, the watermark cannot be detected satisfactorily. However, this result is acceptable because a one-third cut rate will also result in a large loss of carrier audio data.
(3) MP3 compression. At present, it is a common audio processing technology to compress and encode audio signals with MP3. Its goal is to reduce the amount of audio data without affecting the quality of original audio signals. Different bit rates correspond to different MP3 compression ratios. In this test, an audio clip containing watermark is first compressed at a bit rate of 96Kbps (the compression ratio is 7.4:1), and then the corresponding decoding process is carried out. The detected watermark image is shown in Figure 8.
5 conclusion
In recent years, the research on the embedding and detection of audio watermarking in transform domain is developing rapidly, and the discrete wavelet analysis (DWT) is one of the hotspots of the whole digital watermarking system. Experiments show that the algorithm has good concealment performance, hardly weakens the quality of original audio, and has certain ability to resist shear attack and other attacks. In order to further improve the robustness of the algorithm, further consideration should be given to how to utilize more HAS features as well as the position and intensity of watermark embedding. Considering the practicability of the algorithm, the capacity of embedding watermark should be increased. These are all areas for further improvement.
clear all; clc; key=35; Orignalmark=double(imread('suda64.bmp')); % Read 64*64 watermark image [wrow,wcol]=size(Orignalmark); if wrow~=wcol error('wrow~=wcol error'); End % - test key key is beyond the scope of -- -- -- -- -- -- -- -- -- n = check_arnold (wrow); if (key+1)>n error('arnold key error'); end Arnoldw=arnold(Orignalmark,wrow,key); % Arnold transform watermark image [X,fs,bits]= wavRead ('laile.wav'); %X=imread('lena.bmp'); figure; Subplot (2,1,1); %imshow(X),title(' original image '); %X=double(X); plot(X); % Display audio file waveform title(' original audio signal '); %sound(X,fs,bits); %pause; % watermark embedding -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- [c] l = wavedec (X, 2, 'db3'); 2= appCoef (c, L,'db4',2); The low and high frequency coefficients cd2= DETCoef (C, L,2); cd1=detcoef(c,l,1); lca=length(ca2); % low frequency length blocksize=fix(lca/(wrow*wcol)); Water_vector = 0 0 0 0 0 0 0 0 0 0 0 0 % Convert the scrambled watermark to a one-dimensional wlength=wrow*wcol; % watermark length a=0.25; % quantization step j=1; for i=1:wlength Block=ca2(j:j+blocksize-1); [U,S,V]=svd(double(Block)); Cc = floor (S/a (1, 1)); If (Arnoldw (I) = = 1) % embedded in an odd number of times the if (mod (cc, 2) = = 0) cc = cc + 1; End of S (1, 1) = a * cc; End the if (Arnoldw (I) = = 0) % embedded even times the if (mod (cc, 2) = = 1) cc = cc + 1; End of S (1, 1) = a * cc; end Blockw=U*S*V'; Ca2 (j:j+blocksize-1)=Blockw; j=j+blocksize; end c1=[ca2',cd2',cd1']';Copy the code
