Introduction - Advanced Video Coding Systems - page 3

is far beyond the capacity of the conventional disk storage device, e.g., CD-ROM
(700M), or DVD (4GB). Even with the state-of-the-art storage format, a blue-ray
disc with dual layers can store only about 5min of raw video data. In addition,
the total bitrate required would be approximately 1.5Gb/s. However, according to
ATSC 8-VSB digital television standard (McKinney and Hopkins 1995 ), only 6MHz
channel bandwidth can be allocated, which is far from the raw data rate. Therefore,
the raw video data must be compressed by a figure of approximately 100:1 200:1,

with relatively high quality with as few visible artifacts as possible. Along with the
explosive video data increase, especially high definition or even ultra-high definition
videos, the compression technique is more and more important to effectively store
and efficiently transmit them.
Although the raw video data usually have a huge volume, there are lots of redun-
dancies among them which provide room for compressing them significantly. There
are mainly five kinds of redundancies in video data.
(1) Spatial redundancy indicates that in an image (or a frame in video), high
correlation exists between neighboring pixels, especially for pixels in regular objects
or background area, where the distribution of brightness and color saturation are
more regular and not random. These correlations in image data appear when the
neighboring pixels or local structures are similar or even repeated. In other words,
the probability of dramatic variation among neighboring pixels is very small, while
the probability of neighboring pixels with slow-varying is very high.
(2) Temporal redundancy indicates that there are high correlations among neigh-
boring frames. This correlation appears since the physical characteristics (brightness
and color, etc.) are similar among neighboring frames. This is because the neighbor-
ing frames are captured within very short time intervals, e.g., about 0.17 s for videos
with the frame rate of 60Hz. The background usually does not change in such short
time intervals, and only some objects may have few changes in position.
(3) Set redundancy exists in a set of similar images or videos, which refers to the
common information found in more than one image or video in the set. Along with
the development of the cloud storage, more and more similar, or even near-duplicate
images and videos are uploaded to the cloud, and they might differ in quality, res-
olution, geometric variation, and photometric variations (color, light changes). The
common contents among them are repeatedly stored in multiple copies by traditional
data compression methods.
(4) Statistical redundancy means that in source coding process, the bits used for
different entropy coding methods are different. For real-world data with nonuniform
distribution, it will take fewer bits to compress them by representing frequently
occurring symbols with short binary codes on average. Lossless data compression
mainly removes the statistical redundancy.
(5) Visual redundancy means that there are lots of details in video images that are
not easily perceived by human visual system (HVS). In fact, HVS is only sensitive
to some regular changes in images, whereas it doesn't easily perceive the detailed
variations or random changes, e.g., the detailed changes in weeds. These features of
HVS make the lossy compression techniques capable of further improving the video
compression ratio.

∼