Digital Signal Processing Reference
In-Depth Information
Figure 4.1
7.
Iterative
Viterbi
Surface
Optimization:
1.
Initial
Estimates
are
passed
in from the Bootstrap Stage.
2.
Each timeslice of the surface is optimized in isolation
and
form
a
surface
estimate.
3.
Search
parameters
are
re-estimated
based
upon
previous surface estimate.
4.
The system iterates until the surface estimate converges.
1.
Initialize
(
n
=
0)
surface
estimate
from bootstrap
stage:
S
(0
,
*
)
=
S
initial
sur face
,c
(
n,t
)
x
(
n,t
)
,
L
(
n,t
)
Estimate search parameters,
R
(
n,t
)
surface
S
(
n
,
*
)
and
→
,
from
2.
sur face
and
S
(
n
-l ,
*
)
.
3.
Individually optimize each contour of the surface at timeslice t of it-
eration n,
S
(
n,t
)
(4.30)
=DP-Path-Find
φ
V
(
S
(0,
t
)
,
x
) is
where
→
the
Voronoi
Order
from
the
t
th
timeslice
of
x
o
(
n,t
)
the initial surface estimate,
→
is an estimate of a starting point
x
(
n,t
)
of minimum value of
φ
V
,
y
is a point adjacent to
→
→
of maximum
V
such that it completes the closed contour,
R
(
n,t
)
value of
φ
is from
vobject
Eq. 4.29 and
c
(
n,t
)
is from Eq. 4.26.
sur face
Form surface
S
(
n
,
*
)
.If
S
(
n
,
*
)
S
(
n
-1,
*
)
≈
4.
/
go back to step 2, else return the surface
S
(
n
,
*
)
,
then increment n by 1 and
as our video object
estimate.
Search WWH ::
Custom Search