Abstract
Convergent evidence supports a two-stage model of visual motion perception: local direction is computed in V1 and these signals are pooled in MT to derive the global motion of large objects. The influence of local and global motion processing can be teased apart using an equivalent noise (EN) analysis, in which direction discrimination thresholds are measured as a function of the directional variability of the stimulus. The ideal observer embodied by EN analysis computes the population vector average (PVA) of the directions present (an increasingly popular model of perceived direction in complex stimuli). If PVA were correct then performance would depend wholly on the degree of directional variability irrespective of the shape of the underlying directional probability density function (p.d.f.). However, we show that the extent of observers' global motion pooling increases as p.d.f.s are made increasingly leptokurtic/“peaky” (while local motion processing is unchanged). Subjectively, more platykurtic (“flatter”) distributions induce perceived transparency, and we propose that our estimate of the reduction in global pooling that results is the first objective behavioural measurement of motion transparency (since, unlike previous efforts, it is neither prone to criterion effects nor to subjects relying on directional variance). We also constructed “response classification histograms” by averaging all the directions presented at a single level of directional variability, and at a single directional offset (producing ∼75% correct discrimination), according to observers' responses. Results reveal the presence of substantial inhibition between directions differing by 30–45 degrees. This inhibition explains the perception of transparency in stimuli containing uniform distributions of motion directions (and the associated reduction in pooling), and may also contribute to other phenomena of global motion processing, such as direction repulsion.
Funded by the BBSRC (31/S17766)