Digital Signal Processing Reference
In-Depth Information
Figure 4.4.
Taxonomy of visual information,subdivided by class and by energy com-
ponent.
a surface down to a series of 1-D (contour) optimizations, we can ap-
ply dynamic programming. A volume-based approach does not have a
similar decomposition. We choose a surface-based representation over a
volume-based one because of the reduction of dimensionality.
NOTATION
Since this chapter will be referring to a surface and various decompo-
sitions of the surface, we define a series of notations to aid analysis:
1.
S
refers to a surface that represents a video object.
S
(n,t)
2.
when the surface is within an iterative computation,
refers to
the
t
th timeslice of the surface at iteration
n.
S
(n,
*
)
3.
when the surface is within an iterative computation,
refers to
S
(
n,
*
)
t
S
(n,t)
.
the surface at iteration
n,
i.e.,
=
S
(
*
,t)
4.
when
the
surface
is
NOT
within an
iterative
computation,
refers to the
t
th timeslice of the surface.
→
Vectors (
a
,
b
, . . .) are in ℜ
3
5.
→
on the video space.
ENERGY FUNCTION TAXONOMY
We organize our sources of segmentation information into a energy
function taxonomy (see Figure 4.4). We define an energy function as
Search WWH ::
Custom Search