Adaptive optical radial basis function neural network for handwritten digit recognition

Abstract

An adaptive optical radial basis function classifier for handwritten digit recognition is experimentally demonstrated. We describe a spatially-multiplexed system incorporating on-line adaptation of weights and basis function widths to provide robustness to optical system imperfections and system noise. The optical system computes the Euclidean distances between a 100-dimensional input and 198 stored reference patterns in parallel using dual vector-matrix multipliers. For this experiment software is used to perform the on-line learning of the weights and basis function widths. An experimental recognition rate of 86.7% correct out of 300 testing samples is achieved with the adaptive training versus 52.3% correct for non-adaptive training. The experimental results from the optical system are compared with data from a computer model of the system in order to identify noise sources and indicate possible improvements for system performance. O-8194-1544-X/94/$6.OO SPIE Vol. 2240 / 155 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

Optical Neural Network Classifier Architectures

“…Many classifier applications either use binary data directly or may be converted to a binary data format with acceptable degradation including: optical character recognition, fingerprint identification, (edge-enhanced) template matching, and certain disparity computations. More details of this binary-input system design and its application to handwritten digit recognition can be found in earlier publications 10 ' 30 and also in our report from a previous project. 31…”