Role of knowledge in human visual temporal integration in spatiotemporal noise

Previous studies have shown how human observers' knowledge about the signal's spatial frequency, spatial phase, and spatial locations affects human performance in detecting and identifying signals in spatial noise. These results have led to the idea that human observers can be modeled as suboptimal Bayesian observers that use a priori information to generate probabilities or likelihoods for hypothesis. This approach has also been applied more recently to object recognition. We investigate whether human observers have the ability to use information about the temporal profile of a temporally modulated signal in temporal information processing. We measure human performance in detecting a time-varying signal embedded in spatiotemporal (dynamic) noise with and without a cue that contains information about the temporal phase of the signal. Results show improvement in performance in the phase-cued condition, suggesting that human observers act as if they have the ability to use knowledge about the temporal shape of the signal when performing temporal information processing. Human performance is consistent with a suboptimal Bayesian observer and a newly proposed Max-Min observer. The results also suggest that models based solely on the integration of the early temporal filters in the human visual system and/or any further integration (e.g., probability summation), which do not make use of knowledge about the signals' temporal profile, are incomplete models of human visual detection in spatiotemporal noise.