Frequency tuning of perceptual templates changes with noise magnitude

Classification-image analysis has proven to be a valuable tool for revealing features used to perform visual tasks in noise. We use this methodology to investigate how the magnitude of noise influences detection mechanisms, and more specifically, to examine whether observers use a consistent perceptual template across noise magnitude as is often assumed in models. The experiments consist of 2AFC detection of a Gaussian target profile in white noise with RMS contrast levels ranging from 1.25% to 20%. Target contrast was manipulated to maintain a performance level of approximately 80% correct at each noise level. The estimated classification images are presented along with a spatial frequency analysis that consists of radial averages of the frequency domain. The resulting frequency weights show significant within-subject differences across noise levels, as do sampling efficiencies derived from these frequency weights. At low levels of external noise, the classification images are attenuated at low spatial frequencies, giving rise to a more bandpass appearance. At high noise levels, the spatial frequency weights have much less low-frequency attenuation, making them closer to an ideal matched filter. Our results provide direct evidence against the notion of a single consistent perceptual template mediating detection across different levels of noise.