Gray image compression by analog silicon retina based on code and graph theories

A new associative neural network (NN) is described which is constructed based on code and graph theories. This NN is called a SANNET (Sophia associative neural network). Each neuron is an adder unit in the analog NN (ANN) based on real field R^b. The SANNET has many features: no multiplier, sparsity, cellular structure, high concurrency, high speed, and secret communication. The ANN can be applied to the data compression for gray images respectively. The rate of information compression is given by (n/l)^k where n, l and k represent the numbers of nodes, links and layers. The S/N rate in the reproduction image depends on the structure sparsity parameter δ=loop_m/cutset_a where cutset_a represents the average number of links incident to each node, and on the mapping from the original pixel to links in SANNET. Simulation results for gray image compression are given