We obtain the binary images form the original, and the binary images take pixelization later soon. That is pattern of size 16×16 binary version is divided horizontally and vertically to create to 16 subpatterns, each of size 4×4.
For each pattern detection neuron, there are 16 subpattern detection neurons labeled 1,2,3...16 connect to it. They are responsible for remembering the subpatterns of training pattern grids. Each shift subpattern neuron contain a toleration limit. Fig. 3 shows the complete network topology.