Game Development Reference
In-Depth Information
As a conclusion, we can achieve the coding gain when N is approaching to infinity
as,
N
k
2
PCM
1
ʵ
lim
N
ʻ
k
ₒ∞
lim
N
G
DPCM
=
ʻ
k
N
.
(4.23)
k
ₒ∞
2
DPCM
ʵ
lim
N
ₒ∞
It can be proved that the gains of the prediction coding and the transform coding
are the same when
N
is approaching to infinity from Eq.
4.23
(Jayant and Noll
1984
).
However, the prediction coding is more efficient than the transform coding when
N
is not infinity in theory (Jayant and Noll
1984
). The reason is that transform coding
can only explore the correlation within
N
samples. But prediction coding can utilize
infinite correlation by the closed-loop.
In practice, optimal predictor can hardly be designed. Moreover, the transform
coding has some advantages on robustness and error-tolerance. As a result, these two
coding approaches are often combined together to achieve the even better coding
performance.
The above theoretical analysis discloses that prediction coding can provide
promising coding gain. Nevertheless, theory cannot tell us how to develop sophis-
ticated algorithms to take the advantages of prediction coding efficiently as well as
simply. In the rest of this chapter, we will focus on the prediction techniques adopted
in actual video coding standards.
For the prediction coding in standard, intraprediction has been used in video cod-
ing standard sinceH.261, where differential DCcoefficient prediction between neigh-
boring transform block is used, which was also used in later MPEG-1 and MPEG-2.
In MPEG-4, besides DC coefficient prediction, AC coefficients can also be predicted
from the neighboring blocks. In the development of AVC/H.264, spatial directional
prediction was first proposed by Bjontegaard and Gisle (
1998
). Within decades of
study on intraprediction, people have gradually found that spatial directional pre-
diction is more efficient in most cases. As a result, spatial directional prediction
dominates in most state-of-the-art video coding standards, including HEVC/H.265,
AVS1 and AVS2. For interprediction, the earliest using in video coding standard is
P picture in H.261, where only block-based forward prediction from one picture is
used. Later, in MPEG-1, B picture coding was adopted, which support both forward
and backward prediction and two reference frames may be used. Since H.263++,
multiple reference frames prediction has been widely used, including AVC/H.264,
AVS1, and the latest HEVC/H.265 and AVS2 standards. Moreover, the prediction
structure also becomes more flexible, and the corresponding reference management
scheme becomes more andmore complex. FromMPEG-2, variable block sizemotion
compensation was used, where 16
×
8 block prediction was used for field coding,
and inMPEG-4 8
8 block-based motion prediction were added. Since AVC/H.264,
variable blocks motion prediction has been further extended and more kinds of par-
titions modes are developed.
×
