Hyperspace organization for classification of non-stationary patterns

A method has been developed that maps quantized feature values directly to locations in a discrete hyperspace containing pattern class labels or lists of labels. Special efforts have been taken to deal with distortion and time delay uncertainty for signals with nonstationary feature statistics. In the latter case, pattern classes are modeled as trajectories through the hyperspace. During pattern recognition, the intersection of any point on the trajectory is sufficient to retrieve the class label. The procedures described are designed to access a discrete hyperspace for storage and retrieval of signatures in pattern recognition applications.<>