Cryptography Reference
In-Depth Information
frames. The prediction and update steps are realized by using bidirectional
prediction illustrated in Fig. 3.5. HereL
n
stands for the low-pass original
frames at the level n.H
n
denotes the associated high-pass frames. To effec-
tively remove the temporal redundancy, motion compensation is conducted
before the prediction and update. To minimize the overhead, the motion vec-
tors for the update steps are derived from those used for the prediction. By
using n decomposition stages, up to n levels of temporal scalability can be
achieved. The video of lower frame rates can be obtained from the low-pass
frames derived from higher levels,
h
k
S
2k+1
−P
(Haar)
(S
2k
)=S
2k+1
−S
2k
,
(3.3)
1
2
h
k
.
l
k
S
2k
+ U
(Haar)
(h
k
)=S
2k
+
(3.4)
In addition to the (5, 3) wavelet, the Haar wavelet is also used in the
current SVC algorithm. This is used particularly, to improve coding e
ciency.
The terms in (5, 3) and the Haar wavelets are adaptively switched at the
macroblock level. Eq. (3.3) and (3.4) specify the corresponding operations of
the prediction and the update steps for the Haar wavelet. Similarly to the
(5, 3) wavelet, the filtering of Haar wavelet can be simplified to operate as an
unidirectional prediction.
Hierarchical B Pictures
In SVC [9], the base layer must be compatible with the main profile of
H.264/AVC [10]. The compatibility is achieved by using a representation of
the hierarchical B pictures for the lowest spatial layer. With hierarchical B
pictures, the temporal update step is omitted, and this leads to a purely pre-
dictive structure as shown in Fig. 3.6. Here A denotes the anchor frames and
B
n
represents the B pictures of the level n. In addition, the coding order in a
GOP is specified by a series of numbers.
Similarly to the layers encoded with MCTF, the first picture is indepen-
dently coded as an IDR picture, and all remaining pictures are coded in
BBPorBBI groups of pictures using the concept of hierarchical B
pictures. For the prediction of a B picture at the level l, the reference frames
come from the setA∪B
n
n<l. For instance, the B pictures of the
level 1, (i.e., the pictures labelled as B
1
in Fig. 3.6,) use only the surrounding
anchor framesAfor prediction. Similarly, the pictures B
l
use the surround-
ing anchor framesAas well as theB
n
n<lfor prediction. As shown,
the same dependency structure as the MCTF without the update steps is
used. Moreover, the structure of hierarchical B pictures can be supported by
the syntax of H.264/AVC [10].
