The role of multi-area interactions for the computation of apparent motion

Abstract

Apparent motion (AM) is a robust visual illusion, in which fast displays of static objects in successively different positions elicit the perception of object motion. Neurons in higher order areas 21 and 19 compute object motion under such conditions and send feedback to early visual areas 18 and 17, which is instrumental in eliciting computation of motion in those very areas. To explore the computational dynamics of AM, we made a neural field model consisting of two one-dimensional rings of simple neurons expressing firing rates, one for areas 17/18 and one for areas 19/21. The model neurons, without any orientation or direction selectivity, computed apparent motion for the range of space-timings of stimuli associated with short- and long-range AM in humans. The computation of long-range AM in 17/18 required two model areas and the presence of feedback and conduction/computation delays between those areas. As in the in vivo experiments of long-range AM, the stationary stimuli were initially mapped as stationary in model area 17/18, but after the feedback also these lower areas computed AM. The dynamics of the two-area network produces short-range and long-range apparent motion for a large range of feedback strengths and a small range of lateral excitation near the bifurcation to an amplitude instability. The computation of AM in higher order areas was due to the neurons in these areas having large receptive fields as a consequence of divergent feed-forward connectivity. This implies that these areas compute long-range AM when early areas 17 and 18 do not, and therefore higher order areas must enslave lower order areas to compute the same, if the whole network is to arrive at a coherent perceptual solution.