Video Compression - Signal Processing Systems - page 61

Fig. 17 Illustration of
inter-layer residual prediction

Current Residues

Final
Residues

−

Up-sample

Base Layer
Residues

Quality (SNR) scalability is to provide various decoded bit rate truncation
points so that users can choose different visual quality. It can be realized by
three main strategies in SVC—Coarse Grain Scalability (CGS), Medium Grain
Scalability (MGS), Fine Grain Scalability (FGS). CGS is to add a layer with a
smaller quantization parameter, like an additional spatial layer with better quality
and with the same frame resolution. But CGS can only provide few pre-defined
quality points. FGS focuses on providing arbitrary quality levels according to users'
channel capacity. Each FGS enhancement layer represents refinement of residual
signals. It is designed through multiple scans in the entire frame so that the quality
improvement can be distributed to the whole frame. MGS is recently proposed in
H.264/AVC Scalable Extension to provide quality scalability between CGS and
FGS. Please refer to [ 20 ] for the details.

3.3 3D Video Applications and Multiview Video Coding

The TV evolves from the monochrome one to the color one, and then to the
HDTV in the past 80 years. People still want to get a more vivid perceptual
experience from the TV. Stereo vision is the natural feeling for human beings.
Stereo parallax provides us depth perception for scene understanding. In the recent
years, three-dimensional TV (3DTV) becomes popular. It is because more and
more advanced 3D capturing devices (e.g. camera arrays) [ 26 , 27 ] and 3D displays
[ 9 ] are introduced. Multiview video as an extension of 2D video technology is
one of the potential 3D applications. At the same time, multiple 2D views are
displayed as shown in Fig. 18 . Viewers can perceive different stereo videos at
different viewpoints to reconstruct the 3D scene. In order to display multiple views
at the same time, the required transmitted data are much more than those in the
traditional single view video sequence. For example, an 8-view video requires 8
times of bandwidth compared to the single view video. Therefore, an efficient data
compression system for multiview video is required. The Multiview Video Coding
(MVC) system is shown in Fig. 19 .
MPEG-2 Multiview Profile (MVP) [ 16 ] is the first developed standard which
describes the bitstream conformance of stereoscopic video. Figure 20 illustrates
the two-layer coding structure. In the coding structure, the “temporal scalability”