Databases Reference
In-Depth Information
FIGURE 5.5: Structural reduction on binary character images into one-
dimensional bit-string representation.
conduct the recognition process using a simple one-dimensional DHGN subnet
structure. Therefore, it reduces the structural complexity of the DHGN sub-
nets in the network. An advantage of using this structural reduction approach
is that it reduces the structural complexity of patterns while maintaining the
integrity of the contents or elements of the patterns. Thus, the content infor-
mation in each pattern is preserved.
A limitation of this approach is the loss of structural information related
to the pattern. The structure of the pattern or data is unknown to the sys-
tem. Consider the images in Figure 5.5. The DHGN pattern recognizer does
not know that the image represents a character “E.” Rather, it acknowledges
the bit information and its association between neighboring pixels in a one-
dimensional formation.
5.2.2 Content Reduction
In content reduction, more generally known as the dimensionality reduc-
tion approach, features are selected or extracted from the data and are used
in the pattern recognition system. It also transforms the data from a high-
dimensional space to its equivalent low dimension format. Examples of dimen-
sionality reduction techniques include Principal Components Analysis (PCA),
the Linear Discriminant Analysis (LDA), Local Linear Embedding (LLE), and
Kohonen maps.
The dimensionality reduction approach allows the recognition system to
obtain the best and most cost-e
cient data representation of the original raw
