 |
| Fig. 4.1 - Network structure - Stages |
Structure of the neocognitron arises from a hierarchy of extracted features. One appropriate stage of the neocognitron is created for each stage of the hierarchy of extracted features. The network however contains one additional stage, labeled as stage 0, which is not used, in contrast to higher stages, for feature extraction.
All the stages of the neocognitron and a part of features extracted by them, which corresponds to hierarchy in figure 3.1, are shown in figure 4.1.
|
| Fig. 4.1 - Network structure - Stages |
Total number of stages of the neocognitron depends on the complexity of recognized patterns. The more complex recognized patterns are, the more stages of hierarchy of extracted features we need and the higher number of stages of the neocognitron is.