Adaptive edge-only matching model for pattern recognition in humans and machines
Terry Caelli, Shyamala Nagendran · Annual Meeting Optical Society of America · 1986
A procedure for the classification of patterns in terms of their critical edge features is considered. The learning phase involves: (i) an initial multiscaled edge extractor for each sample within a class; (ii) developing multiscaled prototypes within each class; and (iii) emphasizing the unique features of each class.