Abstract
Early motion processing is generally considered to be carried out by “first-order” (Fourier) and “second-order” (non-Fourier) mechanisms. The former extracts motion when pairwise spatiotemporal correlation of luminance signals is present. The latter extracts motion under other circumstances, and is often modeled as local nonlinear pre-processing, such as flicker detection, followed by spatiotemporal correlation.
To further investigate the computations underlying early motion processing, we created a novel set of spatiotemporal movie stimuli. Each frame of the movie is a black-and-white checkerboard. Colorings are determined by a recursive three-check “glider” rule: within the three checks of the glider, the total number of black checks must have a particular parity (even or odd). That is, each check of the movie is calculated from two previously determined checks, of which at least one is from an earlier time frame. This is similar to “isodipole” texture generation, except that here the recursion rule operates in space and time, rather than just space.
We determined the motion percepts elicited by 6 pairs of stimuli. Within each pair, the spatiotemporal configuration of the recursion rule was the same, but parities were opposite. The results were highly consistent across 5 subjects: 5 out of 6 pairs of stimuli were judged as moving in a definite direction (two-alternative forced choice). For 4 of the 5 pairs, changing parity reversed the apparent motion (reverse-phi motion).
These stimuli have no spatiotemporal correlation at second-order, and, moreover, there is no second-order correlation between the locations of checks that flicker, or the presence of edges. Thus, motion cannot be extracted by pairwise spatiotemporal correlation of image luminance, or of derived local features (flicker or edge). However, motion can be extracted by correlation of a derived feature at one position with luminance at another position, i.e., crosstalk between standard first-order and second-order mechanisms.
This project is supported by NIH EY7977.