Databases Reference
In-Depth Information
data obtained from sensory devices or the surroundings. However, some di-
mensionality reduction techniques require expensive computations for feature
processing, selection, and extraction.
There are other techniques that have been proposed for dimensionality re-
duction that incur a low-level of computational complexity. For instance, in
content-based image retrieval (CBIR), histograms and signatures are com-
monly used for dimensionality reduction for images retrieved using color fea-
tures [72].
5.3 Remarks on DHGN DPR Scheme
This chapter has presented the Distributed Hierarchical Graph Neuron
(DHGN), an approach proposed for distributed pattern recognition. The
DHGN implements a divide-and-distribute technique for the HGN networks.
The single-cycle learning and in-network processing features of GN-based algo-
rithms in the DHGN lead to an e
cient recognition scheme that has high recall
accuracy [4]. Furthermore, because of its ability to distribute the recognition
process across a computational network, the recognition times of the DHGN's
distributed pattern recognition scheme are low and stable. The DHGN is able
to lower the storage capacity and communication complexities of the pattern
recognition process. In addition, the two-level recognition implemented in the
DHGN algorithm offers recognition at both the pattern and subpattern lev-
els, which contributes to the higher recall accuracy for simple and complex
data. Moreover, the use of dimensionality reduction schemes, such as binary
signature, implies low computational requirements for a DHGN deployment.
The extensive works on the DHGN DPR scheme for image and optical char-
acter (OCR) recognition can be reviewed in [4, 46, 73]. Note that the DHGN
scheme discussed in this chapter were being considered for complex data that
have multiple data values and dimensions. This consideration will be further
discussed in Part IV of this topic.
