A SYSTEM FOR IMAGE/VIDEO OBJECT QUERY BY SHAPE - Video Object Extraction and Representation: Theory and Applications - page 133

2. Support Structure Classification

We assume this support structure is an invariant characteristic over

the object class.

3. Support Structure Placement

We assume the placement of this support structure is consistently in
the middle of protrusions.

4. Support Structure Strength

We assume the support structure is valued along its points and its
value at a point is dependent upon the percentage of the shape that
the point must support, i.e., in physical terms, the load of the object
that the point must bear.

If an object follows this set of assumptions of support structure existence,
placement, classification and strength, we say the object has an implicit
support structure and this implicit support structure can be a represen-
tation of an object. For instance, since a skeleton does not change as
the human body moves, we can use this support structure to classify
unknown video objects as human. We can define a similarity measure
by comparison of their support structures. To refine our similarity mea-
sure, we can place other descriptions upon our support structure, e.g.,
people have the same skeleton, but some can be fat and others can be
thin.
Many objects do not have an implicit support structure. If an object
does not have a physical skeleton (e.g., a pool of water, a blob of gelatin)
or has an outline that does not give any hints about its internal structure
(e.g., a house, a car), then its implicit support structure may vary over
the class and may not be a good representation for the class.
For objects that follows these assumptions, the shape descriptor based
upon implicit support structure allows a novel functionality: the invari-
ance of similarity measure to bending at joints, i.e., where multiple pro-
trusions meet. Such a shape descriptor can both recognize the person
over a range of motion and become a compact representation of a person
waving his hand. By annotating such a shape descriptor that is based
upon this implicit support structure, we can easily describe a person
walking or a dog wagging its tail.

EXPRESSION OF IMPLICIT SUPPORT
STRUCTURE VIA VOS
From Chapter 3, Voronoi Order Skeleton (VOS) expresses our as-
sumptions of implicit support structure in a mathematical form. The
VOS corresponds to an object's physical skeleton, supporting our claims