Andrew Isaac Meso¹, Jonathan Vacher², Nikos Gekas³, Pascal Mamassian⁴, Guillaume S. Masson⁵; ¹King's College London, ²Université Paris Cité, ³Edinburgh Napier University, ⁴Ecole Normale Superieure, ⁵CNRS & Aix-Marseille Université

Human participants presented with composite stimuli containing two motion vectors, e.g. with upwards and rightwards directions, or very fast and very slow speeds, typically report transparent motion. The transition from coherent to transparent relies on visual processing mechanisms which disambiguate inputs to infer perceptual organisation. To study these mechanisms we use dynamic textures, Motion Clouds, which span a range of mean spatial frequency and bandwidth in seven test conditions, and for each case generate composites of paired stimuli in which the mean speed is the same, but the speed difference is discretely varied from very low to high. We use two experimental approaches. First, with Maximum Likelihood Distance Scaling (MLDS), on each trial we present four stimuli asking which two are more different from each other avoiding direct mention of transparency or coherence. Participants’ data from thousands of trials show differences between stimuli which generally increase for larger speed differences. The shape of the perceptual function and the sensitivity depend on spatial frequency and bandwidth. When participants do a second task, a Forced Choice detection in which they report whether they perceive one or two components in each presented trial, we find that they are only able to reliably report two components for stimuli with mid-spatial frequencies of 1cyc/deg, not 0.25 or 4.0 cyc/deg. Participants also show low sensitivity for broad bandwidths of 2.0 octaves. In a direct comparison of the Forced Choice and MLDS results, participants reliably perceive stimulus differences across speed differences even when they cannot report perceiving two components. The shapes of the MLDS curves vary across individuals being either convex or concave relative to a linear prediction. Our findings reveal that beyond perceiving one or two components, there are additional cues for example rigid/non-rigid on which participants reliably decompose multicomponent motion stimuli when distinguishing them.