Digital Signal Processing Reference
In-Depth Information
Inductive Step for DAG-Based Recursive System Architecture: 1. A
Priori processing is defined as all processing of the information that can be done
without knowledge of recognition results. 2. Partitioning As mentioned in Section 4.,
the next step is to divide the problem into smaller sub-problems while still maintaining
the relationships between sub-problems. 3. Edge Value Extraction is when the DAG
structure of the information is created, the values of the edges must be computed. 4.
DAG-Comparison 5. Hypothesis testing is any post-processing information that uses
the recognizer hypothesis output within its computation.
Figure 5.11.
ation can be repeatedly applied to successive sub-problems within the
same system. A system can recursively apply complex partitioning to
divide the recognition problem into recognition of its macro-structures
and uses the DAG data structure to encode and maintain relationships
among macro-structures. Going into the recursion, the system then
recurses on the macro-structures and divides them into smaller macro-
structures until they are simple enough to be handled by a simple recog-
nizer without partitioning. Coming out of the recursive structure of the
system, the recognizer repeatedly applies the DAG-Compare operation
on the recognized macro-structures to integrate contextual information
and to recognize the composition of macro-structures. In Chapter 6, we
will combine this recursion with our DAG
o
data structure to form a new
data structure for 2-D shape representation.
INDUCTIVE STEP
As shown in Figure 5.11, each system level that uses DAG
o
and the
DAG-Compare is an inductive step in a recursion that has two major
purposes: 1) to decompose the problem into a set of macro-structure
recognition problems and 2) to recognize the datastream through a com-
position of macro-structures. The first purpose is accomplished through
partitioning and recursion; the second through the DAG-Compare op-
eration. Complex partitioning divides the problem in a robust manner,
tolerant to partitioning errors. The recursion on each edge effectively
encapsulates each call on the recognizer at the lower level, allowing in-
Search WWH ::
Custom Search