Video Compression - Signal Processing Systems - page 59

SVC
Encoder

HDTV

Adaptation

Uniform
Bitstream

PC

Total 3. 27 Mbps HD720p (1280x720), 60fps

High Quality

Mobile

CIF (176x144), 15fps
Low Quality

D1 (720x576), 30fps
Medium Quality

Fig. 13 The concept and framework of Scalable Video Coding

can truncate the bitstream to reduce the resolution or visual quality of the videos in
the longer past under scalable video coding rather than just dump them under the
conventional video coding systems. Scalability of video coding also means more
truncation points are provided for the users. For these requirements, H.264/AVC
Scalable Extension (a.k.a Scalable Video Coding, SVC) [ 19 ] is established and
finalized in 2007. The new video coding framework can be depicted as Fig. 13 .
An unified scalable bitstream is only encoded once, but it can be adapted to support
various multimedia applications with different specifications from mobile phone,
personal computer to HDTV.
In general, there are three kinds of video scalability should be included—
temporal scalability, spatial scalability, and quality scalability. Temporal scalability
is to provide different frame rates (temporal resolution). In SVC, temporal scalabil-
ity is achieved by Hierarchical B-frame (HB) as shown in Fig. 14 . The decoding
order of frames in a GOP is from A , B 1 , B 2 ,to B 3 . More decoded frames lead
to higher frame rate. In fact, HB is first proposed in H.264/AVC High Profile as
an optional coding tools to improve coding efficiency because it can explore more
temporal redundancy in the whole GOP. This scheme will induce large encoding
and decoding delay and thus is more suitable for storage applications.
SVC provides spatial scalability (different frame resolutions) with a multi-
layer coding structure as shown in Fig. 15 . If the pictures of different layers
are independently coded, there is redundancy between different spatial layers.
Therefore, three techniques of inter-layer prediction included in SVC. Inter-layer
motion prediction utilizes the MB partition and MVs of the base layer to predict
the information in the enhancement layer as shown in Fig. 16 . Inter-layer residual
prediction upsamples the co-located residues of the base layer as the predictors for
the residues of the current MB as depicted in Fig. 17 . In this figure, the absolute
values of the residues are shown. Pixels with black color is close to zero. Inter-layer
intra prediction upsamples the co-located intra-coded signals as the prediction data
for the enhancement layer.