Cryptography Reference
In-Depth Information
Mismatch error at frame j
Time
j-1
j + 1 j + 2
Error at decoded frames
Error at decoded frames
with leaky prediction
Time
j-1
j + 1
j+2
j+3
Fig. 3.14.
Illustration of mismatch error and drifting behavior.
Fine Granularity Scalability in Low-Delay Applications
Applications with Real-Time or Interactive features normally have the con-
straint of Low-Delay. In these applications, the extra delay for MCTF or Hi-
erarchical Prediction is not permissible. Thus, the traditional GOP structure
of IPPP is preferred for low-delay applications.
With the GOP structure of IPPP, it has been shown in [15, 16]
that the mismatch error will cause further drifting error. In FGS, the quality
enhancement-layer may not perform in the expected manner. Thus, a mis-
match error could occur. Due to drift, the mismatch error of the frame j will
propagate to later frames as illustrated by Fig. 3.14. Accordingly, mismatch
errors at different instances of time are accumulated in the temporal direction,
which significantly degrade the quality of the decoded video.
To reduce drifting errors, the concept of leaky prediction [15, 16] is adopted
by the current SVC algorithm. The concept is to introduce an in-loop gain
factor α at both encoder and decoder to cause the mismatch error to decay.
For better understanding, Fig. 3.15 illustrates the configuration of prediction
loop when the leaky prediction scheme is enabled. The transmission error is
modelled by adding an error term E(t) to the prediction residue of encoder,
R(t). The gain factor α is constrained in the interval of [0, 1). The transmission
error will decay in an exponential manner and the error is circulated in the
reconstruction loop of decoder. The dashed line in Fig. 3.14 characterizes the
effect of drifting behavior with leaky prediction.
3.2.4 Combined Scalability
In SVC [9], all of the scalability can be combined to simultaneously support
a wide range of spatial, temporal, and SNR scalability. Fig. 3.16 illustrates
the use of Combining Different Scalability. In this example, the Spatial Base
