SANNET: Image compression and regeneration by nonlinear associative silicon retina

Abstract

This paper describes the image compression and regeneration by a new nonlinear associative retina chip which is a sparse neural network. This retina chip is a dual network of Hopfield cellular network. The input information sequences is given to links as currents. The error correcting capacity (minimum basins of attraction) is decided by the minimum numbers of links of loop. The operation principle of the regeneration is based on current distribution of the neural field. The most important nonlinear operation is a dynamic quantization to decide the binary value of each neuron output from the neighbor value. The rates of compression used in the simulation is 2/3 × 1/8, where 2/3 and 1/8 means rates, he structural and the binarizational compression respectively.