Game Development Reference
In-Depth Information
neighboring pixels in the same line as well as in the neighboring lines. This same
idea is applied to develop LOCO-I approach (Waggener
1995
) which is adopted into
the lossless image coding standard JPEG-LS (Pennebaker and Mitchell
1993
). The
following development for intraprediction coding includes AC/DC prediction in the
transform domain (CCITT
1987
;MPEG4
1999
) and spatial directional prediction
(ITU-T
2003
). Spatial directional prediction has become the dominant intrapredic-
tion method finally in recent video coding standards (ITU-T
2003
;AVSWorking
Group
2007
; Bross et al.
2013
;AVS2
2014
) due to its high efficiency as well as low
complexity.
In 1962, Seyler published a fundamental paper on interprediction (Seyler
1962
).
From then, researchers proposed many new methods for interprediction, such as
conditional replenishment (Mounts
1969
), 3D-DPCM (Candy et al.
1971
) and pixel-
based motion compensation (Netravali and Robbins
1979
). Finally, people attained
a consensus that block-based motion compensation (Koya et al.
1981
; Jain and Jain
1981
) is the optimal tradeoff between coding efficiency and practical tolerance on
complexity.
From the very beginning when prediction coding was introduced into video cod-
ing, it has attracted a very high research attention because of its completeness in
theory as well as its simplification in practice. There is no doubt that prediction cod-
ing plays an extremely important role, maybe the most important role, in the video
coding scheme. Girod (
1987
), Flierl and Girod (
2004
), Chen and Pang (
1992
), etc.
have done a thoroughly and sufficiently theoretical analysis on block-based motion
compensation. Here a brief theoretical analysis of prediction coding is discussed in
the following subsection.
4.1.1 Theoretical Basis for Prediction Coding
The earliest prediction coding solution was proposed by Cuter known as differen-
tial pulse code modulation (DPCM) (Cutler
1952
). After decades of development
on prediction coding, DPCM is still the basic form of prediction coding. In other
words, all the state-of-the-art prediction methods can be regarded as some kinds
of improvement, reform, or transformation of DPCM (Jayant and Noll
1984
; Wang
et al.
2002
). Therefore we can make a theory analysis on prediction coding by focus-
ing on DPCM as a basic model.
Figures
4.1
and
4.2
show DPCM encoder and decoder respectively. The kernel
part of DPCM is prediction. At encoder, the residual signal, which is obtained by
subtracting the prediction signal from the original signal, is quantized and coded,
instead of the original signal itself. The prediction signal is generated from recon-
structed signal, which has already been decoded previously, following some rules.
It can be observed that the DPCM encoder applies a closed-loop structure, which is
the ancestor of the closed-loop structure used in modern video coding encoders.
