Information Technology Reference
In-Depth Information
Interlace coding
Dirac supports interlace coding by coding sequences of fields, rather than frames.
5.3.2 Motion Estimation
Dirac uses hierarchical motion estimation that consists of three stages. In the first
stage, motion vectors are evaluated for every block of each frame to one pixel
accuracy using hierarchical motion estimation. In the second stage, these vectors
are refined to sub-pixel accuracy. In the third stage, mode decision is performed in
which motion vectors are aggregated by grouping blocks for similar motion.
Motion estimation is most accurate when all three components are involved, but
this is more expensive in terms of computation as well as more complicated
algorithmically. Dirac uses the luma (Y) component only for ME.
5.3.2.1 Finding Motion Vectors of One Pixel Accuracy Using Hierarchical
Motion Estimation
Hierarchical ME speeds things up by repeatedly down converting both the current
and the reference frame by a factor of two in both dimensions, and doing motion
estimation on smaller pictures. In hierarchical motion estimation, Dirac first
determines the number of down conversion levels that can be calculated using
equation 1 as follows:
⎛
width
height
⎞
⎛
⎞
⎛
⎞
level
=
min
⎜
⎜
⎝
log
,
log
⎟
⎟
⎠
⎜
⎝
⎟
⎠
⎜
⎝
⎟
⎠
(1)
2
2
12
12
In equation 1, the number of down conversion levels is 4 and 3 for the CIF
(352*288) and QCIF (176*144) frame format respectively. At each level, the
process of down conversion reduces the height and width of the current and
reference frame by a factor of 2 in each dimension. The size of the frame becomes
one quarter at each level. Hence CIF image of size 352*288 and QCIF image of
size 176*144 is reduced to 22*18 respectively in the last level.
The motion estimation is started from the lowest level resolution (level 4 in
CIF) frame and gradually moved to higher level resolutions and finally reaches the
original frame size. At each level of the hierarchy, except the smallest level,
vectors from lower levels are used as a guide for searching at higher levels. Each
block at the lower resolution level corresponds to four blocks at immediate higher
resolution level, so each block at the lower resolution level provides a guide
motion vector to at most 4 blocks at immediate higher resolution level. The block
sizes are variable. The middle blocks are of size 12*12. The other block sizes are
10*10, 10*12, 10*6, 10*4, 10*8, 8*10, 6*10 or 4*10 depending upon the location
of the block and the size of the frame and these are consistent at each level of the
motion estimation hierarchy.
Search WWH ::
Custom Search