Game Development Reference
In-Depth Information
Fig. 4.4
Angular prediction
modes
Fig. 4.5
Reference samples
used for prediction
Figure
4.5
shows the neighboring samples used for prediction sample interpola-
tion. Assume the current prediction block is
Curr
, and its size is
M
N
. Then besides
the up-left corner sample, M left-down neighboring reconstructed samples and
M
left neighboring reconstructed samples are used for prediction, denoted with r[
i
]
(
i
×
2
M
, r[0] is the left-up corner sample), and
N
up neighboring reconstructed
samples with
N
up-right neighboring reconstructed samples are also used for predic-
tion, denoted with c[
j
](
j
=
0
∼
2
N
, c[0] is the left-up corner sample). If part of those
reference samples do not exist or haven't been reconstructed, they are inferred as
available or not available in an implicit way both in encoder and decoder. In detail,
if r[
i
](
i
=
0
∼
M
) is not available, all the left neighboring samples will be viewed as
not available. For r[
i
](
i
=
1
∼
=
M
+1
∼
2
M
), if they are not available but
r
[
M
]
is available,
