Non-Fourier information in bandpass noise patterns

Random dot patterns and white-noise luminance textures are widely used in psychophysical experiments to study low-level visual processes. Because these noise patterns are broadband, bandpass filtered versions are employed to limit their frequency content. It is not recognized, however, that bandpass noise patterns have amplitude-modulation (AM) components. The AM signal is not present in the Fourier spectrum, nonetheless, it is a valid signal for second-order mechanisms. We characterize the properties of the AM signal in bandpass noise textures: the relevant periodicities of the AM signal are always much lower than the actual passband; the upper frequency limit of the AM signal increases with the linear bandwidth. We present psychophysical data to demonstrate the perceptual significance of the AM signal in bandpass noise. We provide a method for obtaining AM-free bandpass patterns, and compare psychophysical performance in experiments employing AM-present and AM-free bandpass noise patterns as stereoscopic stimuli. The results show that the AM component contributes to stereoscopic discrimination performance at large disparities. We suggest that the low-frequency AM signal is a possible confounding factor in experiments employing bandpass noise textures, and that linear filtering can isolate spatial scales effectively only for linear systems.