OPTICAL DUAL-SCALE ARCHITECTURE FOR NEURAL IMAGE RECOGNITION

A novel neural-network architecture that combines image data reduction with focus of attention to achieve reduced training cost, improved no ise tolerance, and better generalization performance than comparable c onventional networks for image-recognition tasks is presented. The dua l-scale architecture is amenable to optical implementation, and an exa mple optical system is demonstrated. For one example problem, the best -case improvements of the dual-scale network over its conventional cou nterpart were found through simulation to he a factor of 6.7 in traini ng cost, 67.3% in noise tolerance, and 61.6% in generalization to dist ortions. The dual-scale network is also applied to one instance of a h uman face recognition problem.