A correlation architecture for scale and rotation-tolerant pattern recognition.

摘要

A scale and rotation tolerant pattern recognition system was developed in this dissertation research. The architecture developed uses a scale and rotation tolerant correlation method coupled to a post-processing neural network. The architecture meets the requirements of a high-speed automatic target recognition (ATR) system and provides good pattern recognition capabilities. The architecture also has the ability to reject objects that were not included in the target set. In computer simulations, the architecture accurately identified members of the target set with 89.5% accuracy and rejected members of the non-target set with 93.8% accuracy. This accuracy was with a target set of scaled and rotated objects and a non-target set of scaled and rotated objects.