Digital watermarking is a technique used for preventing illegal copying and tampering of data. It is similar to the paper watermarking. It is mainly used for two purposes: authentication of data and copyrighting. Using copyrighting we can trace illegal copies of digital data and certify original owners. Also in case of dispute over a data the watermark can be used to prove the ownership. In watermarking the original data is superimposed with a signal containing the copyrighting information. This signal is not perceivable and thus it does not interfere with the listeners perception of the data.
Digital watermarking is the process of embedding copyrighting information into digital media frame, such as text, image, audio and video. They are preferred to be imperceptible to the end user.
The need for watermarking arises because of the inherent ease with which digital data can be copied and manipulated.
Digital watermarking find application in areas like copyrighting of information, authentication of data, tracing of illegal copies etc.
Audio watermarking refers to applying watermarking on audio signals. The knowledge of the human auditory system i.e. psychoacoustics is of great importance in audio watermarking. This aspect of audio watermarking necessitates it to be treated as a separate topic.
Current watermarking techniques are mainly concerned with spread spectrum type of watermarking.
HISTORY OF WATERMARKING
Information hiding is a very old art. By the sixteenth and seventeenth centuries, there had arisen a large literature on steganography and many of the methods depended on novel means of hiding information. Watermarking became an area of intense research in the 90â„¢s. Early work in this area mainly concentrated on the watermarking for digital images. By 1996 the first papers on audio watermarking appeared in the literature. The initial attempts were in adapting the existing image watermarking algorithms into audio signals. The spread spectrum based scheme was the most popular one.
In the earlier watermarking schemes no attention was given to the utilization of the knowledge of the host signal. In second generation schemes the knowledge of the host signal was made use of to improve the system performance. Psychoacoustic shaping is a good example of this.
GENERAL REQUIREMENTS OF A WATERMARKING SYSTEM
An ideal watermarking system should have the following characteristics.
1 The watermark should be embedded in the host audio material itself, not in any syntactically identifiable part of the audio material (e.g. header). Else, the watermark can be easily removed or manipulated using purely syntax-based schemes.
2 The perceived quality of the watermarked signal should not deteriorate (beyond certain limit), from that of the host.
3 Any audio player, that can play the host signal, should be able to play the watermarked signal as well.
4 The scheme should ensure that there would be no miss (watermark is not detected when there is one) and no false alarm (a watermark is detected when there is none.
5 The scheme should ensure that there would be no false alarm across watermarks. That is, when tested for a watermark different from the one that is embedded in the signal, the detector must give a negative answer.
6 The watermarking should be robust to normal signal processing operations on the signal like scaling, translation, spectral filtering, compression and decompression chains, D/A-A/D conversion chains.
7 Further, some degree of robustness to intentional tampering is also expected. Such attacks, if successful in tampering the WM, should also tamper the signal so that the quality is deteriorated to render the audio material useless.
METHODS OF AUDIO WATERMARKING
1. ECHO HIDING
Echoes of the signal are generated, scaled and added to the host signal. The position of the echoes are determined by the copyright information being added. The echoes must be placed in the coloration region and hence psychoacoustics is made use of.
2. QUANTISATION INDEX MODULATION
The host signal is quantized using a quantiser from an ensemble of quantisers. Each user can be assigned a different quantiser. The reconstruction points are used to decode the watermark.
3. FEATURE BASED WATERMARKING TECHNIQUE
Certain features of the audio signal are imperceivably modified to embed the watermark. Time scale modification is one such technique. In this the distance between the silent points in the host audio is slightly modulated with watermark.
4. SPREAD SPECTRUM BASED WATERMARKING
In spread spectrum watermarking, copyright information is modulated using pseudo random digital sequence. This is equivalent to spreading the information in the frequency domain. Hence the name. This modulated sequence is then added to the audio signal at an imperceptibly small strength to generate the watermarked signal.
SPREAD SPECTRUM BASED WATERMARKING.
The psychoacoustic properties is made use of in the spread spectrum based watermarking scheme. The properties of the human auditory system is known as psychoacoustic properties.
The important psychoacoustic properties are
1. ABSOLUTE THRESHOLD OF HEARING
It characterizes the amount of energy needed in a pure tone such that it can be detected by a listener in a noiseless environment. It is a nonlinear function of frequency.
2. CRITICAL BANDS
The inner ear consists of a bank of highly overlapping band pass filters. The magnitude responses are asymmetrical and level dependant. Cochlear filter pass bands are of non uniform bandwidth. The bandwidth increases with frequency. The bandwidth frequency relationship is given by the equation
It refers to the process where one sound is rendered inaudible because of the presence of another sound.
SIMULTANEOUS (SPECTRAL )MASKING
The presence of a strong noise or tone masker creates an excitation of sufficient strength on the basilar membrane at the critical band location to block effectively, the detection of a weaker signal.
NONSIMULTANEOUS (TEMPORAL )MASKING
For masking of finite duration, nonsimultaneous masking occurs both prior to the masker onset as well as after masker removal.
THE WATERMARK EMBEDDING SCHEME
The watermark embedding scheme modifies the original audio signal, which is represented as a 16-bit sample sequence sampled at 44100 Hz mono
The watermark embedding scheme consists of the following operations.
1. Temporal analysis of the audio signal.
2. m sequence generation
3. Shaping of the m sequence.
4. Cyclic shifting and inclusion of the information payload.
5. Watermark embedding
1. TEMPORAL ANALYSIS OF THE AUDIO SIGNAL
The host audio is analyzed in the time domain, where a maximum or a minimum is determined in the block of audio signal that has the length of 7.6 ms .The goal of the temporal analysis is to place the watermark inside the raw audio signal without making any perceptual distortion to the host signal by making use of the characteristics of the human auditory system.
The masker value determined from the temporal analysis is used as a reference point to determine the level of power of the watermark sequence in the analyzed block. With reference to the temporal masking curves and the length of the analyzed audio frames ,it can be concluded that the added information must be at least 24db below the power level of the audio maximum in the frame. The algorithm equally uses both the pre and post masking properties, therefore making the most significant error if the maximum of the host audio is situated at the end of the analyzed block. However the impact of
x(n) a(n) y(n)
WATERMARK EMBEDDING SCHEME
The sub maximums and the maskers from the contiguous blocks is not negligible and it helps the current masker to fulfill its role. As a result of the analysis the watermark samples are weighted by the coefficient a(n) in order to be adjusted to the psycho acoustic perceptual thresholds.
M SEQUENCE GENERATION
The m sequence is a pseudo random sequence. It is bipolar sequence. The m sequence is obtained from a shift register with feedback and is represented in the bipolar form. Prior to further processing the m-sequence is filtered in order to adjust it to the thresholds of the human auditory system.
SHAPING OF M-SEQUENCE
The main goal is to adapt the watermark to such a form that the energy of the watermark is maximized under the condition of keeping the auditory distortions to a minimum, although the SNR value is significantly decreased. Despite the simplicity of the shaping process the result is an inaudible watermark as the largest amounts of the shaped watermarks power are concentrated in the frequency sub bands with lower HAS sensitivity. In addition these frequency sub bands are an essential part of the watermarked audio and cannot be removed from the spectrum without making significant damage to perceptual quality.
CYCLIC SHIFTING: INCLUDING PAYLOAD INFORMATION
A cyclic shifted version of c(n) is used to achieve a multibit payload for one particular sequence s(n).Every possible shift may be associated with different information content. Therefore ,information payload is directly proportional to the length of the m sequence. The cyclical shifting of the shape filtered m sequence changes only the phase but not the amplitude of the spectrum. Therefore the desired spectral shaping is attained. There is always a possibility to make the tradeoff between the embedded data size and the robustness of the algorithm; as m sequence is decreasing the algorithm is able to add more bits into the host audio but the detection of the hidden bits and resistance to different attacks is decreased.
The watermark signal is embedded into the host signal using three time aligned process. In the first stage, the m sequence has been filtered with the shaping filter where a colored-noise sequence s(n) is the output. Samples of the s(n) sequence are then cyclically shifted ,where the shift value is dependent on the input information payload. At the output of the watermark embedding scheme, shifted version of s(n) ,sequence c(n) is being weighted and embedded to the original audio signal
y(n) = x(n) + . a(n).c(n)
Where x(n) denotes the input audio signal ,a(n)are coefficients from the temporal analysis block and alpha is a parameter that represents the tradeoff between the perceptual transparency and detection reliability. However addition of the c(n) sequence in the embedding process is done repeatedly in order to make the system resistant to time scaling attacks that tend to de-synchronize the extraction process. As increases, robustness of the embedded watermark is better but it is limited by the allowable perceptual distortion of the watermark audio.
The diagram of audio watermark detection scheme is as shown in the figure .The proposed extraction scheme does not require access to the original signal to detect the embedded watermark. The cornerstone of the detection process is the mean removed cross correlation between the watermarked audio signal and the equalized m-sequence. However before the watermarked signal is segmented into blocks in order to measure the cross correlation with the m sequence ,the detection algorithm filters it with the equalization filter.
Equalization is also performed on the incoming m-sequence in order to match the incoming watermarks as closely as possible, regarding known modification of watermark audio. Generally in correlation detection scheme it is often assumed that the communication channel is white Gaussian. Applying a whitening process would highly reduce any correlation in the audio and thus achieve optimum detection .The equalization filter suppresses the low frequency components with high energy and emphasizes the high-frequency part of the audio spectrum in order to obtain a noise like more flat spectrum. Frequency domain shaping process of the watermark is repeated in the extraction part as well, in order to optimize the matched filtering performance. Hence, input sequence of auto correlation process are m sequences with the same temporal and frequency domain characteristics., if there is no signal processing of the watermark and the detection is synchronized.
Before the start of the integration process, which determines the peak and the output value, the block power normalization part of the scheme makes uniform energies of the output blocks from the correlation calculations .Thereby every output block from the correlator has approximately equal impact on the integration process that lasts for approximately 84 consecutive frames. Otherwise a poor correlation result in block with high amplitudes of the host audio diminishes the result of many positive correlation detections and determines the peak and
WATERMARK EXTRACTION SCHEME
Its position. The peak value is related directly to the detection reliability, whereas its position corresponds to the cyclic shift-information payload. The detection reliability depends strongly on the number of the accumulated frames. In general tradeoff is made between the time of integration and the amount of hidden data.
The correlation method and the watermark extraction algorithm in general are reliable only if the correlation frames are aligned with those used in watermark embedding. Therefore one of the malicious attacks can be desynchronization of the cross correlation procedure by time scaling modifications. In that case the watermark detection scheme is not properly determining the shift value in the embedded watermark, resulting in high increase of BER. One of the methods against time scaling is to use redundancy in the watermark chip pattern. Thus there is a tradeoff between the robustness of the algorithm and computational complexity, which is significantly increased by performing multiple correlation tests.
APPLICATIONS OF WATERMARKING
Watermarking finds application in a wide variety of fields
2. AUTHENTICATION OF DATA
3. TRACING OF ILLEGAL COPIES
4. TO TRACK THE TAMPERING OF DATA
Advantage of watermarking over decoding is that while watermarked signals can be played with the original player decoded data need special players to play them.
One possible disadvantage that could be held against the watermarking scheme is that it reduces the quality of the audio. But with proper care the amount of perceptual distortion can be totally eliminated.
The watermarking technology has wide applications in the future of data communication. The existing technologies like decoding are not very user friendly and will soon be replaced by watermarking. Thus watermarking of audio is going to be the most wanted technology in the internet enabled age and many more new and advanced watermarking algorithms will be developed in the future which will take the watermarking technology to newer heights.
1. P.BASSIA, I.PITAS, Robust audio watermarking in the time domain IEEE Transactions on multimedia, June 2001
2. MITCHELL D.SWANSON, BIN ZHU, AHMEDH.TEWFIK, Robust audio watermarking using perceptual masking
3. N.CVEJIC, A.KESKINARKAUS, T.SEPPANEN Watermarking of audio using m sequences and temporal masking.
4. NADELJKO CVEJIC, TAPIO SEPPANEN Improving audio watermarking scheme using psychoacoustic watermark filtering.
3. REQUIREMENTS OF WATERMARKING SYSTEM
4. SPREAD SPECTRUM BASED WATERMARKING
6. WATERMARK EMBEDDING
7. WATERMARK EXTRACTOIN
I extend my sincere thanks to Prof. P.V.Abdul Hameed, Head of the Department for providing me with the guidance and facilities for the Seminar.
I express my sincere gratitude to Seminar coordinator
Mr. Manoj K, Staff in charge, for his cooperation and guidance for preparing and presenting this seminars.
I also extend my sincere thanks to all other faculty members of Electronics and Communication Department and my friends for their support and encouragement.