Databases Reference
In-Depth Information
TABLE 7.3: Pattern Samples of the Zernike Moment Feature Obtained Using
Discretization
Object ID
Feature Pattern
1
12343000132001234000030000400120033010033011400
2
01243000222001134001320022300220131012033011400
3
01243000121001234001200022400100130011034011400
4
12243000112001233001120023300120131001034002400
5
01143000122001234000120022300120132001033012400
7.3.2.2
Stage 2: Multi-Feature Recognition
In multi-feature recognition for multiple features of numeral character ob-
jects, each DHGN network performs recognition on a specific feature set. The
sizes of the networks are not uniform.
The recognition process begins when the coordinator node communicates
the feature patterns to the SI module node on each network according to
the specific feature assigned to the DHGN network. The communications of
patterns in this scheme follow the message-passing model described in Section
2.6.2.
The SI module node in each network divides and distributes the received
patterns to the available subnets in the network. Each DHGN subnet initiates
a recognition process at subpattern level. The results of each recognition pro-
cess are sent back to the SI module node where the maximum voting process
is used identify the best match pattern class for each pattern. After complet-
ing the voting process, the SI module determines the accuracy parameters
used in the scheme. These parameters can include commonly used recognition
accuracy parameters, such as precision rate, recall rate, accuracy level, and
error value. These values are communicated to the coordinator node for the
results evaluation stage.
7.3.2.3
Stage 3: Results Evaluation
The results evaluation stage determines the best or optimal feature to be
selected as the best representative for each pattern class in the recognition
scheme. This process occurs within the coordinator node. The values obtained
from the SI module nodes are compared to the accuracy parameter(s). This
evaluation stage of the DHGN multi-feature recognition applies a generic ap-
proach, and different sets of recognition accuracy parameters can be used in
the classification process. This approach allows for flexibility in the decisions
on classification, in that different accuracy factors can be observed and ana-
lyzed.
