Purpose:

To provide evidence of visual system's configural processing in biological motion detection, by comparing human vs. ideal observers.

Method:

A three-frame sequence of a 12-point human walker, as signal, was embedded in dynamic random-dot noise. In each trial subjects detected the walker in a yes-no task, with feedback. There were 10 counter-balanced blocks, 120 trials each. One of five levels of signal strength was used per block: 4, 6, 8, 10, or 12 points were randomly sampled from the 12 walker points per trial. Detection sensitivity per signal strength level was defined by the number of noise dots at 75% correct detection, in a staircase procedure (Neri, Morrone, & Burr, 1998).

An ideal observer was designed using knowledge of how the random noise dots and walker dots were generated. It assumes knowledge of the dynamic form of the walker but does not know which walker dots are present. The ideal observer made decisions based on the number of noise or walker dots located at appropriate positions, which follows binomial distributions.

Results:

When the walker was upright, subjects' sensitivity was a quadratic function of the number of displayed walker dots, a result predicted by the ideal observer. This suggests that subjects used the dynamic configural representation of biological motion. In contrast, when the walker was upside-down, subjects' sensitivity was a linear function of the number of displayed walker dots, suggesting that subjects were unable to efficiently use the dynamic walker representation, but instead detected the signal walker based on a low-level process. The latter is consistent with the prediction by Barlow & Tripathy (1997) based on smooth motion correspondence of individual dots.