Abstract

Studies of the human visual system have led to neural network models for computing with a variety of architectures and algorithms that can exploit invariants in image processing, feature extraction,¹ and subsequently fault-tolerant associative memory for pattern recognition. Scale and rotation invariance is achieved by image data flowing through a nonuniform polar exponential sampling grid to a uniform output space.² Image domain bandpassing techniques are then used to extract invariant feature information from which fault-tolerant associative memory using the outer product formulism is applied to handwritten character recognition.

© 1986 Optical Society of America