Abstract:
Audio watermarking is a technique providing a promising solution to copyrights protection
for digital audio and multimedia products. Using this technique, hidden information called a
watermark containing copyright information is imperceptibly embedded into the host media.
This watermark may be extracted from a suspected media to verify the authenticity. To
function as an effective tool to enforce ownership rights, the audio watermarking scheme must
satisfy the imperceptibility, robustness, security, data payload and computational complexity
requirements.
This thesis is focused on developing novel approaches of audio watermarking for copy
rights protection, as well as introducing new methods for objectively predicting the percep
tual quality of the watermarked audio signals.
We proposed a spread spectrum based audio watermarking algorithm which includes the
psychoacoustic model, multiple scrambling, adaptive synchronization, frequency alignment
and coded-image watermark. Essentially, a watermark modulated by pseudorandom num
bers (PRNs) is embedded in the time-frequency domain of the host signal. Without reference
to the host signal, the watermark is detected via performing a normalized correlation be
tween the watermarked signal and the corresponding PRNs. In watermark embedding, the
minimum masking threshold from the psychoacoustic model is employed to shape the am
plitude of the watermark signal, so that the noise caused by watermarking is kept inaudible.
Also, multiple scrambling is integrated into the embedding to prevent the watermarking
scheme from unauthorized detection. In watermark detection, adaptive synchronization and
frequency alignment are developed to help extract the watermarks from the attacked wa
termarked signals that suffer loss of synchronization. Additionally, we use the coded-image
watermark to represent the copyright information, so as to provide a semantic meaning for
reliable verification as well as extra security after encryption.