A new approach for pattern recognition by neural networks with scramblers
Hosokawa, Omatu, Fukumi · 1989
An approach to pattern recognition that is based on a concept involving an invariance net and a trainable classifier is proposed. The invariance net plays an important role in producing a set of outputs that are invariant to translation, rotation, scale change, perspective change, etc., of the retinal input pattern. The trainable classifier is used to classify the scrambled data into the original patterns by using a backpropagation algorithm. The sigmoid functions are adopted as nonlinear elements in the neural networks, whereas B. Widrow et al.'s MRII (see IEEE Trans. Acoust. Speech Signal Proc., vol.ASSP-36, no.7, p.1109-18, 1988) are based on signum functions. Some numerical results are illustrated to show the effectiveness of the present algorithm for pattern recognition.>