In this section, the experimental study of the impact of
K-SOM’s popular learning rate generation functions, or
learning kernel, and codebook data representations on the
outcome image quality is presented. This then leads to a
conclusion of the appropriate specification of a learning
kernel and a codebook data representation. It is noted
again that the final objective of our research is to make it
feasible to realize a K-SOM quantizer in a hardware
platform. As a result, the complex operations, both in
terms of hardware utilization and processing time, which
require a floating point unit should then be avoided.