Using electrical responses of directionally selective simple cells to simulate human motion perception

Two briefly flashed adjacent lines shown in rapid succession evoke a sensation of motion beyond that of seeing the line in two successive positions. Wertheimer (1912) called this “phi”. If a stroboscopically presented visual pattern is shifted in space and its contrast polarity is inverted, perceived motion is opposite to stimulus motion, hence “reversed phi” (Anstis 1970). This is predicted by the correlation model designed to describe turning responses of beetles to visual motion (Reichardt 1962). The model receives inputs from 2 successively activated points (1 and 2) and the first activation is delayed with respect to the second. Thus both arrive simultaneously at the correlator, a multiplicative unit. This computes an output whose sign indicates the direction of motion. Experiments testing the correlation model have traditionally used patterns periodic in space and/or time. But the model should correctly predict perceived motion of single objects, too: for stimuli with equal contrast polarity, the sign of the output should be positive. For opposite contrast polarities, the output sign should be negative, indicating reversed phi. Here we test this prediction using two vertical lines flashed in succession, varying spatial and temporal offset, and contrast polarity. Line pairs presented on a grey background were either both bright or dark (equal contrast polarity), or one line was bright and the other dark (opposite contrast polarity). Observers indicated the direction of motion by pressing a button. Perceived direction is veridical for equal contrast polarity line pairs, but is reversed for opposite contrast polarity line pairs with spatial offsets of 0 to 12 min of arc, and temporal offsets of 8 to 25 ms. Further increase of either offset yields veridical perception. These results are consistent with the correlation model and the energy model (Adelson Bergen 1985) for short-range motion of single objects. The mechanisms proposed by the latter model are close to properties of nerve cells. We simulate human perception of apparent motion using responses of directionally selective simple cell as inputs (Priebe Ferster 2005). The cells responses to lines in space and time are added and the probability of firing a spike is computed. As predicted by the energy model the results of our simulations correlate well with responses of human observers perceiving phi and reversed phi