101Image Compression Using Vector Quantization
topological feature maps that are formed in the brain. In the KSOFM or
SOFM network, each neuron has an associated topological neighborhood of
other neurons. During the training process, both the winning neurons as
well as the neighborhood neurons of the winners are updated. The size of
the neighborhood is decreased as training progresses until each neighbor-
hood has only one unit.
The neighborhood of a unit in SOFM is a nothing set of connected neu-
rons. For example, the connections may form grids of rectangular, hexagonal
sizes. After training a winning unit is selected. The weight vector of each
unit in the winning unit’s neighborhood is then updated as follows:
+= +ε −wn wn nD xn wn
ii