Generic contour features in neural network pattern recognition

In this paper it will be argued that there are two types of learning in human pattern classification. The first type is the evolutionary learning which has occurred over thousands of years, and has resulted in the set of pre-wired features for image analysis with which infants are born. The second type is the learning within the lifetime of the individual whereby the pre-wired features are used in pattern classification. It will be argued that these two types of learning should also be explicitly separated in neural networks. The approach taken in this research is to develop neural networks which learn to classify contour image patterns in terms of a set of predefined features.