A SYSTEM FOR IMAGE/VIDEO OBJECT QUERY BY SHAPE - Video Object Extraction and Representation: Theory and Applications - page 140

1. the DOT-major node,

A DOT-major node tree is empty or contains a DAG o (see Defini-

tion 5.1), composed of DOT-minor nodes and edges. Each edge value

on the DAG, contains a DOT-major node reference.

2. DOT-minor node,

A DOT-minor node is a node contained by a DOT-major node.

3. DOT-minor edge,

A DOT-minor edge is an directed edge from two DOT-minor nodes
in the same DOT-major node. It is contained within a DOT-major

node. It contains a DOT-major node reference and an edge value.

4. DOT-major node reference,

A DOT-major node reference is a pointer to a DOT-major node.
DOT-major node reference is restricted such that, if we induce graph
from the DOT-major node references where 1) its nodes correspond to
the DOT-major nodes and 2) its edges correspond to the DOT-major
node reference, leaving the containing DOT-major node and going to
the DOT-major node reference, then this graph must be a DAG.

5. and the edge value. The edge values are data structures on the

DOT-minor edges that represent the VOS segment and also contain
a DOT-major node reference.

Note that the ordered tree is a special case of the DOT. The DOT is an
extension of the ordered tree by generalizing not only the linear sibling
order to an order specified by a DAG, but also linear ancestor-descendant
order to one specified by a DAG. The DOT allows the complex parti-
tioning of the ordered tree and the efficient encoding of multiple ordered
tree representations into a single data structure (see Figure 6.8). This

comparison between DOTs is discussed in the Section 6.12.

5. SYSTEM DESIGN
The system implements a shape descriptor from the extraction of VOS
as discussed in Chapter 3, and a similarity measure from an efficient
comparison algorithm of DOT-represented support structures based on
DAG work in Chapter 5. The block diagram for our shape query system
is shown in Figure 6.9. In our tests, we are given a database of shapes.
We extract our shape descriptors from these shapes and link their shape
descriptors with their original contours. Given a query shape, we extract
a shape descriptor and then calculate its similarity measures against all
other shape descriptors within the database and return the top n -highest
scoring shapes from the database. The next section details the two