Cryptography Reference
In-Depth Information
Table 3.1.
Inter-layer prediction modes for an inter macroblock.
Mode0 Mode1 Mode2
Extension of Macroblock Partition
Yes
Yes
No
Refinement of Motion Vector
No
Yes
No
Prediction of Residue
Yes
Yes
No
In the prediction process, the residues and motion vectors of the subor-
dinate layers will be interpolated first if the subordinate layer has a lower
resolution. The macroblocks of different types are also processed differently.
For an intra macroblock, the inter-layer prediction is only permitted if the cor-
responding 88 block in the subordinate layer is further encoded within an
intra-coded macroblock. Prediction within an inter macroblock is always en-
abled. The current approach provides 3 different inter-layer prediction modes
for an inter macroblock. The partition of an inter macroblock can be derived
from the relevant sub-blocks at the subordinate layer as shown in Fig. 3.10.
The motion vectors can also be obtained by scaling those for the corresponding
sub-blocks. The scaled motion vectors can be further refined in the quarter-
pel range between 1 and +1. Depending on the macroblock partition and
the refinement of motion field, Table 3.1 is used to summarize the inter-layer
prediction modes for an inter macroblock.
3.2.3 SNR Scalability
To achieve SNR scalability, the residues after the inter-layer prediction are
successively quantized into multiple quality layers. At client side, the decoded
quality depends on how many quality layers are received. The video quality
is gradually improved as more quality layers are received.
Currently, the SVC algorithm supports two types of SNR scalability, which
are Fine Granularity Scalability (FGS) and Coarse Granularity Scalability
(CGS). The CGS offers a limited number of quality levels with distinct bit
rates while the FGS provides an infinite number of quality levels with contin-
uous bit rates. The quality layers are generated differently according to the
granularity of scaling.
Coarse Granularity Scalability
To achieve CGS, the stack structure shown in Fig. 3.9 is used. This is done
in a different manner to spatial scalability. The CGS is achieved by encod-
ing the video of the same spatial resolution in different spatial layers. Thus,
each spatial layer is used as a quality enhancement-layer. Fig. 3.11 illus-
trates the prediction structure of CGS, where one base-layer and two quality
enhancement-layers are generated where Qp0 > Qp1 > Qp2. Compared to
the stack structure in Fig. 3.9, the interpolation of prediction residues and
