Showing papers by "Simon J. Thorpe published in 1997"

The Effects of Spatial Frequency on Binocular Fusion: From Elementary to Complex Images

Abstract: Factors limiting binocular fusion were studied using 2-dimensional difference of Gaussian (2D-DOG) stimuli. The proportion of fused stimuli and observer's response time were determined for stimuli that varied in spatial frequency composition between 0.22 and 4.8 cycles per degree. At small disparities, mean fusion response times were short and relatively stable but increased rapidly once the disparity reached a certain critical value. This 2-phase function implies the existence of 2 separate fusional mechanisms: a rapid neurally based fusional process, which operates at small disparities, and a second mechanism involving reflexive vergence movements operating at disparities 2 to 3 times larger. Both mechanisms are highly influenced by spatial frequency, being 4 to 5 times more effective at low spatial frequencies. Additional experiments demonstrated that with compound stimuli, the fusion limit is not determined by the highest spatial frequency components (as had been reported previously) but, rather, can take advantage of the additional fusional range associated with low spatial frequencies. Such cooperation may be obvious only in the case of 2-dimensional stimuli.

How can the visual system process a natural scene in under 150 ms? On the role of asynchronous spike propagation

Abstract: The human visual system can process previously unseen natural scenes in under 150 ms, even with extrafoveal viewing. Such data impose serious temporal constraints on the way in which information is processed. We argue that one of the keys to understanding this remarkable efficiency lies in the fact that biological vision makes use of asynchronous spike propagation, a feature absent in the vast majority of artificial neural networks. We describe our recent work that has explored the computational advantages of such asynchronous processing.

Implementing Hebbian Learning in a Rank-Based Neural Network

Abstract: Recent works have shown that biologically motivated net works of spiking neurons can potentially process information very quickly by encoding information in the latency at which different neurons fire, rather than by using frequency of firing as the code. In this paper, the relevant information is the rank vector of latency order of competing neurons. We propose here a Hebbian reinforcement, learning scheme to adjust the weights of a terminal layer of decision neurons in order to process this information. Then this learning rule is shown to be efficient in a simple pattern recognition task. We discuss in conclusion further extensions of that learning strategy for artificial vision.