Abstract
We constructed a novel stimulus that appears to rotate yet contains no local translational motion. The stimulus is constructed by applying a bandpass orientation filter to a noise image. On each consecutive frame, the filter is rotated about the Fourier domain origin and reapplied to the original image. The stimulus elicits a strong rotational motion percept within any arbitrary window, and is opaque to traditional computational models of motion processing. For example, when an extended gradient model is applied to the stimulus (Johnston, McOwan & Benton, 1999, Proc. R. Soc. London, Ser. B, 509–518), no coherent motion is detected. We examined whether our stimulus activates human MT/V5 using a blocked fMRI design. We compared the BOLD response evoked by the following four stimuli: (a) fractal rotation stimulus, (b) frame shuffled fractal rotation stimulus, (c) rigid rotation of a single frame from the fractal rotation stimulus, and (d) a frame shuffled version of our rigid rotation stimulus. For frame shuffled versions we permuted the order in which frames were presented. An examination of the response in human MT/V5 for our 10 subjects showed that both the fractal rotation stimulus and rigid rotation stimulus evoked similar (and significant) increases in response compared to their frame shuffled counterparts. Indeed we find no significant difference in MT/V5 activation between our fractal rotation and rigid rotation stimuli. Using standard psychophysical procedures, we then investigated whether our stimuli elicited motion aftereffects. As expected, the rigid rotation stimulus evokes both a static and dynamic motion aftereffect. We found no such aftereffects with our fractal rotation stimulus. We take our results to demonstrate the existence of a feature tracking mechanism that inputs to human MT/V5. The lack of a motion aftereffect may well imply this process to be fundamentally different to standard low-level motion analysis.