Game Development Reference
In-Depth Information
Fig. 3.3
Illustration of directional prediction modes
avoid floating point operation, and a 4-tap linear interpolation filter is used to get the
subpixel.
For the chrominance component, the PU size is always N
N, and 5 prediction
modes are supported, including vertical prediction, horizontal prediction, bilinear
prediction, DC prediction, and the prediction mode derived from the corresponding
luminance prediction mode (Piao et el.
2013
).
×
3.2.2 Inter Prediction
Compared to the spatial intra prediction, inter prediction focuses on exploiting the
temporal correlation between the consecutive pictures to reduce the temporal redun-
dancy. Multi-reference prediction has been used since the AVC/H.264 standard,
including both short-term and long-term reference pictures. In AVS2, the using of
long-term reference picture is further extended, which can be constructed from a
sequence of long-term decoded pictures, e.g., background picture used in surveil-
lance coding, which will be discussed separately in a later section. For short-term
reference prediction in AVS2, F frames are defined as a special P frame (Ling and
Yu
2014
), in addition to the traditional P and B frames. More specifically, a P frame
is a forward predicted frame using a single reference picture, while a B frame is a
bi-predicted framewhich consists of forward, backward, bi-prediction, and symmetric
prediction, using two reference frames.
In a B frame, in addition to the conventional forward, backward, bi-directional,
and skip/direct prediction modes, symmetric prediction is defined as a special
bi-prediction mode, wherein only one forward motion vector is coded and the
