Game Development Reference
In-Depth Information
Fig. 4.14
Sub pixel position
The skip/direct mode in B frame is similar to that in F frame, where either temporal
or spatial derivation methods can be utilized.
4.3.4 Sub Pixel Interpolation
In AVS2, the motion vector is in quarter-pel precision for luma, and the sub-pel pixel
is interpolated with an 8-tap DCT-like interpolation filter (DCT-IF). For chrominance
component, the motion vector precision is in 1/8 pel and the 4-tap DCT-IF is used
(Lv et al.
2013
; Wang et al.
2014
).
For luma interpolation, Fig.
4.14
gives the position of the integer sample, 1/2
sample and 1/4 sample of the reference picture. The integer sample position is marked
with the capital letter while 1/2 and 1/4 samples position are marked with the small
case letter. For example, the prediction value of the sample position
a
0
,
0
,
b
0
,
0
, and
c
0
,
0
are obtained via filtering the nearest 8 integers of the interpolation point in
the horizontal direction. The prediction values are obtained as follows, where
Clip
function is defined as clipping the value into the sample bit-depth range: