Unified representation of motion and motion streak patterns in a model of cortical form-motion interaction

Abstract

Problem. Direction selective neurons in visual cortex (V1) encode spatio-temporal movements of visual patterns. It has been suggested that motion directions are also spatially encoded in the form channel as oriented motion streaks (Burr, Curr. Biol., 2000), while only fast motions lead to motion streak patterns (Apthorp et al., Proc. Roy. Soc. London B, 2013). Geisler (Nature, 1999) proposed that motion streaks aid determining visual motion direction estimation while their awareness is suppressed in normal vision conditions (Wallis & Arnold, Curr. Biol., 2009). The underlying neural mechanisms of such form-motion interaction are, however, still unknown. Method. We propose a neural model that acquires data from an event-based vision sensor that responds to temporal changes in the input intensity. Model area V1 uses spatio-temporal filters to detect visual motion and forwards activations to be integrated in model area MT. Orientation-selective contrast cells in model areas V1 and V2 spatially integrate recent visual events and respond to oriented structures parallel to movement direction when sufficiently fast motion is presented. Form cells' responses temporally cease already for slow motions. Results and Conclusion. We probed the model with dark/light random dot patterns moving at different directions and speeds, replicating experimental settings. For higher speeds oriented contrast-sensitive cells are co-activated along an orientation parallel to the motion direction, viz., signaling motion streaks or speedlines. For slow motions no such responses occur. Adaptation effects confirm experimental findings from psychophysics. The model suggests that motion streaks occur in the form channel as a direct consequence of fast coherent motions along single directions without the need to assume separate motion channel representations. The model makes predictions concerning the strength of the streak patterns and sheds new light upon mechanisms of computing motion from form. Meeting abstract presented at VSS 2014