Information recovery from rank-order encoded images

The work described in this paper is inspired by SpikeNET, a system developed to test the feasibility of using rank-order codes in modelling largescale networks of asynchronously spiking neurons. The rank-order code theory proposed by Thorpe concerns the encoding of information by a population of spiking neurons in the primate visual system. The theory proposes using the order of firing across a network of asynchronously firing spiking neurons as a neural code for information transmission. In this paper we aim to measure the perceptual similarity between the image input to a model retina, based on that originally designed and developed by VanRullen and Thorpe, and an image reconstructed from the rank-order encoding of the input image. We use an objective metric originally proposed by Petrovic to estimate perceptual edge preservation in image fusion which, after minor modifications, is very much suited to our purpose. The results show that typically 75% of the edge information of the input stimulus is retained in the reconstructed image, and we show how the available information increases with successive spikes in the rank-order code.