Abstract
Local motion sensors are selectively sensitive to the spatiotemporal Fourier energy generated by retinal image movement. We have developed a Fourier-domain signal-to-noise paradigm to investigate how the outputs of motion sensors are combined during integration. In the space domain, stimuli were two-frame 1-D random-line kinematograms. In the Fourier domain, frequency components were divided into noise components and signal components. Noise components shifted in phase by a random amount in either direction. Signal components either (i) shifted in phase in the same direction by a constant phase angle; or (ii) shifted in phase in the same direction by a constant velocity. Signal direction varied randomly from trial to trial, and the observer was required to report the direction of motion seen in each trial. We measured the minimum ratio between signal and noise components required for reliable direction discrimination. Data showed a performance advantage for the constant-phase stimulus over the constant-velocity stimulus. This result, and data from subsequent experiments, is discussed in the context of current motion integration models.