Data processing: speech signal processing – linguistics – language – Speech signal processing – Psychoacoustic
Reexamination Certificate
1998-10-14
2001-04-17
Dorvil, Richemond (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Psychoacoustic
C704S205000, C704S500000
Reexamination Certificate
active
06219634
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to digital signal processing and, in particular, to a particularly robust watermark mechanism by which identifying data can be encoded into digital signals such as audio or video signals such that the identifying data are not perceptible to a human viewer of the substantive content of the digital signals yet are retrievable and are sufficiently robust to survive other digital signal processing.
BACKGROUND OF THE INVENTION
Video and audio data have traditionally been recorded and delivered as analog signals. However, digital signals are becoming the transmission medium of choice for video, audio, audiovisual, and multimedia information. Digital audio and video signals are currently delivered widely through digital satellites, digital cable, and computer networks such as local area networks and wide area networks, e.g., the Internet. In addition, digital audio and video signals are currently available in the form of digitally recorded material such as audio compact discs, digital audio tape (DAT), minidisc, and laserdisc and digital video disc (DVD) video media. As used herein, a digitized signal refers to a digital signal whose substantive content is generally analog in nature, i.e., can be represented by an analog signal. For example, digital video, and digital audio signals are digitized signals since video images and audio content can be represented by analog signals.
The current tremendous growth of digitally stored and delivered audio and video is that digital copies which have exactly the same quality of the original digitized signal can easily be made and distributed without authorization notwithstanding illegality of such copying. The substantive content of digitized signals can have significant proprietary value which is susceptible to considerable diminution as a result of unauthorized duplication.
It is therefore desirable to include identifying data in digitized signals having valuable content such that duplication of the digitized signals also duplicates the identifying data and the source of such duplication can be identified. The identifying data should not result in humanly perceptible changes to the substantive content of the digitized signal when the substantive content is presented to a human viewer as audio and/or video. Since substantial value is in the substantive content itself and in its quality, any humanly perceptible degradation of the substantive content substantially diminishes the value of the digitized signal. Such imperceptible identifying data included in a digitized signal is generally known as a watermark.
Such watermarks should be robust in that signal processing of a digitized signal which affects the substantive content of the digitized signal to a limited, generally imperceptible degree should not affect the watermark so as to make the watermark unreadable. For example, simple conversion of the digital signal to an analog signal and conversion of the analog signal to a new digital signal should not erode the watermark substantially or, at least, should not render the watermark irretrievable. Conventional watermarks which hide identifying data in unused bits of a digitized signal can be defeated in such a digital-analog-digital conversion. In addition, simple inversion of each digitized amplitude, which results in a different digitized signal of equivalent substantive content when the content is audio, should not render the watermark unreadable. Similarly, addition or removal of a number of samples at the beginning of a digitized signal should not render a watermark unreadable. For example, prefixing a digitized audio signal with a one-tenth-second period of silence should not substantially affect ability to recognize and/or retrieve the watermark. Similarly, addition of an extra scanline or an extra pixel or two at the beginning of each scanline of a graphical image should not render any watermark of the graphical image unrecognizable and/or irretrievable.
Digitized signals are often compressed for various reasons, including delivery through a communications or storage medium of limited bandwidth and archival. Such compression can be lossy in that some of the signal of the substantive content is lost during such compression. In general, the object of such lossy compression is to limit loss of signal to levels which are not perceptible to a human viewer or listener of the substantive content when the compressed signal is subsequently reconstructed and played for the viewer or listener. A watermark should survive such lossy compression as well as other types of lossy signal processing and should remain readable within in the reconstructed digitized signal.
In addition to being robust the watermark should be relatively difficult to detect without specific knowledge regarding the manner in which the watermark is added to the digitized signal. Consider, for example, an owner of a watermarked digitized signal e.g., a watermarked digitized music signal on a compact disc. If the owner can detect the watermark, the owner may be able to fashion a filter which can remove the watermark or render the watermark unreadable without introducing any perceptible effects to the substantive content of the digitized signal. Accordingly, the value of the substantive content would be preserved and the owner could make unauthorized copies of the digitized signal in a manner in which the watermark cannot identify the owner as the source of the copies Accordingly, watermarks should be secure and generally undetectable without special knowledge with respect to the specific encoding of such watermarks.
What is needed is a watermark system in which identifying data can be securely and robustly included in a digitized signal such that the source of such a digitized signal can be determined notwithstanding lossy and non-lossy signal processing of the digitized signal.
SUMMARY OF THE INVENTION
In accordance with the present invention, a watermark is added to a digitized signal as watermark data encoded in a basis signal which fits noise thresholds determined by constant-quality quantization approximation. Most of the processing resources required to determine a noise threshold spectrum for generation of the basis signal in some watermarking systems is devoted to quantization of the digitized signal to measure noise introduced by such quantization, generally by repeated quantization in an iterative search for a relatively optimum gain. By contrast, noise introduced by quantization is estimated in accordance with the present invention by determining a continuously differentiable function which approximates noise introduced by such quantization and using the fimction to solve for a relatively optimal gain to be applied during such quantization.
The noise thresholds specify a maximum amount of imperceptible noise power for respective ranges of frequencies over a set of offsets. The relatively optimal gain for a particular range of frequencies is determined by constraining a sum of gain-adjusted amplitudes within the range of frequencies to the maximum amount of imperceptible noise power and solving for a variable gain. Once a relatively optimal gain is determined for a range of frequencies, individual noise thresholds can be easily determined by adjusting individual amplitudes at each frequency and measuring the difference between amplitudes so adjusted and unadjusted amplitudes. If noise thresholds are stored as noise power, the measured differences are squared. Thus, noise thresholds for individual frequencies of a audio signal spectrum gain be very efficiently determined without requiring repeated quantization in an iterative search for a relatively optimum gain.
The reduction in processing resources is particularly apparent and beneficial when generating multiple basis signals in a search for a best offset when attempting to detect a watermark in a digitized signal which is suspected to include a watermark signal.
The continuously differentiable function includes a local quantization stepsize. A local quantization step
Dorvil Richemond
Ivey, Esq. James D.
Liquid Audio Inc.
LandOfFree
Efficient watermark method and apparatus for digital signals does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Efficient watermark method and apparatus for digital signals, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Efficient watermark method and apparatus for digital signals will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2491437