27-07-2012, 04:06 PM
Temporal Feature Modulation for Video Watermarking
Video Watermarking-base.pdf (Size: 328.96 KB / Downloads: 26)
INTRODUCTION
NOWADAYS, the problem of copyright violation of video
contents has become very serious. For example, a digital
recorder can acquire high-quality movies by capturing, which
is an easy and attractive tool for piracy [1]. Even though it
is very difficult to prevent piracy completely, it is useful to
track and punish the pirates by employing watermarking or
fingerprinting schemes [1]–[3].
Video watermarking techniques can be classified into three
categories according to the embedding domain: frame-based
modulation; three-dimensional (3-D) transform modulation;
and temporal feature modulation. Early algorithms extended
image watermarking techniques to video watermarking [4],
[5], which is called frame-based modulation. In 3-D transform
modulation, a group of video frames is represented using
3-D transforms, and then watermarks are embedded into 3-D
transform coefficients [6]. In temporal feature modulation,
a temporal feature is extracted from each video frame and
modulated to convey watermark information. In other words,
the temporal features from a series of frames form a onedimensional
(1-D) signal, which is used as the watermark
embedding domain. Haitsma and Kalker’s algorithm [7] can
be regarded as a temporal feature modulation scheme, which
computes the average pixel value for each frame and modulates
the average values.
ALGORITHM I: FRAME SKIPPING
We embed a temporal watermark by skipping a selected set
of frames from the original video. The frame skipping pattern
becomes the temporal watermark. The human visual system is
not sensitive to a slight change in the frame rate. For example,
29.97-Hz TV signals can be converted into 24-Hz motion
pictures without noticeable artifacts. Thus, the frame skipping
does not cause severe degradation if the original video has a
high frame rate and the skipping rate is relatively low.
CONCLUSIONS
We proposed two temporal feature modulation algorithms.
The first algorithm skips a selected set of frames from the
original video according to a watermark codeword, and the
decoder matches the watermarked video frames to the original
frames to estimate the codeword. The second algorithm
modulates NCGs of blocks to embed 1-bit information to each
frame, and its decoder can estimate the watermark codeword
in a blind manner. Experimental results demonstrated that both
algorithms are robust against H.264/AVC compression attacks
and temporal attacks, including frame removal, insertion, and
swapping.