Double watermarks
FIELD OF THE INVENTION
The present invention relates to a method and apparatus for processing watermarks in a signal. The invention also relates to a signal.
BACKGROUND OF THE INVENTION
Watermarks are perceptually invisible messages embedded in the content of signals of audio and video. Watermarks can support a wide variety of applications such as monitoring and copy control.
A watermark can be used to embed copy protection information in a signal. In the choice for the watermark, a compromise needs to be made for instance between detection time, detection complexity, robustness and content quality deterioration.
For copy protection of audio signals, two types of watermarks can be employed, a robust watermark and a fragile watermark. A robust watermark is very resistant against attacks, but is able to carry only a small amount of payload. A fragile watermark on the other hand, is able to carry a higher payload than the robust watermark, but is only resistant to mild attacks. Therefore, a combination of a robust watermark and a fragile watermark is a common combination. The high robustness of the robust watermark is used along with the high payload that can be carried by the fragile watermark since the latter does not require redundancy. In the case the fragile watermark has been removed, there is still the robust watermark to detect that a fragile watermark should have been present.
This is described for instance in "If one watermark is good, are more better", Fred Mintzer and Gordon W. Brandaway, IBM T. J. Watson Research Center, in which the implementation of multiple watermarks is described, of which the watermarks are combinations of different types such as a combination of an ownership watermark that is very robust and a verification watermark that is quite fragile.
Since a robust watermark should be highly robust against all kinds of attacks, the detection of a robust watermark usually implies a high increase in complexity. If the application allows it, generally there can be a trade-off between this increase of complexity and detection time and/or robustness. However, in applications where this trade-off is not
possible, e. g. where high robustness and short detection time are required at the same time, the increase in complexity could be a problem.
Thus, there is a need of watermark processing for both video and audio signals that overcomes the problem with being sensitive for attacks, where a robust watermark can be employed in such a way that detection time is short and robustness high without the problem with increased complexity.
OBJECT AND SUMMARY OF THE INVENTION
An object of the invention is to provide a method for embedding watermarks in a signal, which method solves the problem stated above.
According to a principal aspect of the present invention, this is realized in a method for embedding watermarks in a signal, whereby multiple robust watermarks are provided.
According to another aspect of the invention, a first robust watermark is added to the signal, which watermark is robust against a limited set of signal processing that can be expected by a "friendly user" under normal conditions and a second higher complexity watermark is also added to the signal, which watermark is robust against all kinds of malicious attacks. Herein, the term "higher complexity" means that an additional watermark is more complex or has a longer timing algorithm than the first watermark that is searched for.
As a result of these limited robustness requirements, detection complexity and detection time for this first robust watermark can be very low, whereby in many situations, only the first watermark has to be detected.
In some situations where the first watermark was not considered to be present, the second watermark is searched for and found indicating that the first watermark should have been present.
There is provided, in accordance with a preferred embodiment of the invention a method for embedding watermarks in an information signal, wherein the method comprises the following steps:
- adding at least a first watermark being robust against a limited set of processing attacks that can be expected by friendly users under normal conditions,
- adding at least a second watermark being robust against all kinds of malicious attacks.
Preferably, all at least two robust watermarks are embedded simultaneously. Alternatively, one or more watermarks can be embedded time-multiplexed.
Preferably, adding said at least second watermark comprises watermarks being robust against operations such as mp3 encoding/decoding, D/A and A/D conversion, equalization and (modest) speed-control.
There is further provided, in accordance with a preferred embodiment of the invention, a method for detecting watermarks embedded in an information signal, wherein the method comprises the following steps:
- providing a watermarked information signal and evaluating said signal for embedded watermarks,
- in a first step detecting if a first watermark is present,
- in a second step detecting if a second watermark is present.
There is further provided an apparatus for detecting a watermark signal embedded in an information signal, comprising:
- at least a first robust watermark detector for detecting if a first watermark is present, and
- at least a second robust watermark detector for detecting if a second watermark is present, whereby the first detector searches for a first robust watermark during a limited detection time, or with a limited detection complexity, and if not found the second watermark detector searches for the second watermark during a detection time that is longer or with a detection complexity that is higher than in the first detector. Both detectors can also be combined in one detector.
According to an embodiment of the invention, there is provided a multimedia playing and/or recording apparatus, comprising an apparatus for detecting a watermark signal embedded in an information signal, comprising:
- at least a first robust watermark detector for detecting if a first watermark is present, and
- at least a second robust watermark detector for detecting if a second watermark is present, whereby the first detector searches for a first robust watermark during a limited detection time, or with a limited detection complexity, and if not found the second watermark detector searches for the second watermark during a detection time that is longer or with a detection complexity that is higher than in the first detector.
According to an embodiment of the invention, there is further provided a signal comprising at least a first watermark being robust against a limited set of processing attacks that can be expected under normal conditions by friendly users, and at least a second watermark being robust against all kinds of malicious attacks.
For a watermark to be effective in ownership applications, it is essential that it remains embedded and detectable in the content in which it is inserted. This is accomplished by the invention.
The above described aspects and other aspects of the invention will be apparent from and elucidated with reference to the embodiments(s) described hereinafter.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 shows a method according a preferred embodiment of the invention, whereby a first and a second watermark are detected in an audio signal.
FIG. 2 shows an apparatus according to a preferred embodiment of the invention.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS OF THE INVENTION According to an aspect of the invention, multiple robust watermarks are embedded, at least one being robust against a limited set of processing attacks that can be expected under normal conditions by friendly users, the other against all kinds of malicious attacks. The processing attacks could comprise operations such as mp3 encoding/decoding,
D/A and A/D conversion, equalization and speed-control.
Before describing preferred embodiments of the invention a brief introduction to embedding of watermarks in a signal will be given.
Embedding of watermarks in audio can generally be described by a formula
(I):
Y=X+αXW (I)
in which:
X is a frequency domain representation of the original audio.
Y is a frequency domain representation of the watermarkeid audio.
W is a frequency domain representation of a secret noise-like watermarked pattern. α is a global scaling parameter to control the strength of the watermark.
This embedding algorithm comprises adding a scaled version of W to an original audio X. In other words, a watermark as described above is simply additive noise. The samples of the watermark pattern can for instance be drawn from a normal distribution N(0, 1) with mean and standard deviation equal to 0 and 1, respectively, or they can also be
floating point values, or integer based. Samples of the watermark pattern can also be obtained in any other suitable way.
Other embedding techniques without scaling parameters for instance are also possible, but will not be further described since they are well known for a person within this technical field.
According to a preferred embodiment of the invention, a method for embedding watermarks in a signal is provided. The method comprises the following steps:
- adding a first watermark wl being robust against a limited set of processing attacks that can be expected by under normal conditions by friendly users,
- adding a second watermark w2 being robust against all kinds of malicious attacks.
Adding' each watermark, the first wl, and the second w2, could be done for instance using a method known per-se, such as the one described above.
More watermarks than two could of course be implemented if required. In the case where multiple watermarks of similar types are embedded, it can be advantageous to apply all watermarks simultaneously or alternatively one or more watermarks can be embedded time-multiplexed. If fragile watermarks are to be simultaneously embedded, it is preferred, or in some cases even necessary to embed them after embedding the robust watermarks to not destroy them.
In Fig. 1 a method according a preferred embodiment of the invention, whereby a first and a second watermark wl and w2 are detected in an incoming audio signal Qin is shown. In this example of an embodiment of the invention, two robust watermarks wl and w2 are embedded in an audio signal, for instance from a CD that is played. An incoming signal Qin comprising a first watermark wl and a second watermark w2 is searched for detection of the "simple" robust first watermark wl in a first step 1. If the first watermark wl is found, it is labelled as being present in a second step 2. In a third step 3, copyright management is applied.
If, on the contrary, in the first step 1, the first watermark wl is not found, a fourth step 4 is performed for detection of the second watermark w2. If this second watermark w2 is detected, it is labelled as "present" and transferred to step 2, whereby the third step 3 is performed as for the first watermark wl .
If, the second complex watermark w2 is not detected in the fourth step 4, the content copy is considered to be free from watermarks, step 5. Thereafter, copyright management is applied in step 3 as for the first watermark wl.
In Fig, 2 an apparatus 10 for performing the method according to the invention is shown. The apparatus 10 for detecting watermarks in a signal comprises a first low complex watermark detector 11 and a second higher complex watermark detector 12. Typically, both detectors are implemented in consumer electronic equipment such as CD- apparatus.
The first detector 11 performs initial detection for a robust low complex watermark wl and if this fails, i. e. no watermark is found, the signal Qin is transmitted to the second detector 12, which performs detection for a second robust higher complex watermark w2.
Preferably, the second detector 12 is designed not to be complex but having a long detection time > Td. The detection time Td is defined as the detection time in the first detector 11.
Now a general description of a detection algorithm will follow. The watermarks can be detected using any suitable type of conventional method such as the one described below.
Detection of watermarks in signals according to prior art:
{ Start of watermark detection }
■{ Start of phase 1: accumulation of data r
Input signal Z = 0
For a number n of partly overlapping frames the method steps are:
1 : load a frame ( e. g. 2048 samples) - y
2:decorrelate y (e. g. by convolving with < -1, 2, -1 -> Λy
3: apply a window (e. g. Hanning) and transform to a frequency domain - Y = F(y)
4 accumulate magnitudes only where Z = Z+ 1 Y |
"{ end of phase 1
{ start of phase 2 "correlation for detection" \
1 : the SPOMF-method is used
→ C = F1(P(F(Z)).P(F(W)*)) with proper windows applied, and where P is:
P(X) = x / |x| if x ≠ 0, and P(0) = 1
2: normalize the standard deviation of C and look for maximum p
- p = max (C/σc)
3: watermark is most likely to be present if p exceeds certain threshold T(e. g. T=5,5)
\ end of phase 2 \
\ end of watermark detection r
As used in the following claims, the words "comprise" or "include" or their conjunctions means "including, but not necessarily limited to.