Digital Signal Processing Reference
In-Depth Information
DP-Path-Find (
f
order
,
x
o
,
x
END
),
R
(
x,y
),
c
(
x,y
))
→
→
→→
→
→
{
1
2
ARRAY
mscore,backtrack;
3
4
5
∪
{
x
END
}
)
{
for
x
→
(
β
(
f
order
,
x
o
,
x
END
)
→ →
→
mscore (
x
)
→
←
-
∞
;
}
6
7
mscore(
x
o
)
→
←
0;
→
8
9
∪
{
x
END
}
) in increasing
f
order
(
y)
{
→
→→
→
for
y
(
β
(
f
order
x
o
x
END
)
→
(
f
order
)
x
o
,
y
)
{
for
x
→
β
→ →
if(mscore (
y
)< (
R
(
x, y
)+
c
(
x, y
)
.
mscore(
x
))
{
10
11
12
→
→
→
→ →
→
(
R
(
x, y
)+
c
(
x,y
)
.
mscore(
x
)) ;
mscore (
y
)
→
←
→ →
→ →
→
backtrack (
y
)
→
←
x
;
→
}
13
}
14
}
15
16
return backtrack
;
17
}
Figure 4.13.
A
general
algorithm
for
dynamic
programming.
A
particular
problem
can
be
specified
the
parameters:
f
order
is
the
1-D
ordering
of subproblems,
x
0
is
→
the given starting point,
x
END
is the ending problem,
R
(
x,y
) is score contributions
→
→
→
to
maximize
w.r.t.
the
path,
c
(
x, y
)
is
the
relationship
between
subproblems, and
→→
= {
z
|
f
order
(
x
)
β
f
order
(
z
) <
f
order
(
y
)} returns the set of dependent
subproblems that occur after
x
before
y
in the order. The algorithm returns the path
in video space that maximizes the recurrence relation in Eq. 4.13.
(
f
order
,x,y
)
→→
→
→
≤
→
→
→
→
1. edges are part of a closed contour whose sum total edge intensities
are high
2. edges faces the object center X.
3. and edges are in a contour that circumnavigates the object center X.
For a given frame at time
t',
the first heuristic can be formulated as
follows:
arg
C
(
max
(
exp
(
||
C
|| -
L
)
2
))
∫
C
.
R
canny
(
I
(
x,y,t
)|
t=t'
,
s
)
d
→→
σ
C
(4.15)
Search WWH ::
Custom Search