Silicon retina: Image compression by associative neural network based on code and graph theories

Abstract

An associative neural network (NN) is constructed on the basis of code and graph theories to realize a silicon retina. Each neuron is an EXCLUSIVE-OR unit in the digital NN (DNN) based on finite field GF(2). Each neuron is an adder unit in the analog NN (ANN) based on real field Rb. The network has the following features: no multiplier, sparsity, cellular structure, high concurrency, and high speed. The DNN and the ANN can be applied to data compression for binary and analog images, respectively. The S/N rate in the reproduction image depends on the network structure. Secret image communication and image recognition are possible.