Abstract
Observers can detect animals in natural scenes rapidly and accurately. Rousselet et al. (Nat. Neurosci. 2002) briefly flashed (26 ms) 2 natural scenes, both in the periphery (centered at 3.6 deg). In these conditions, animal detection was slightly less accurate compared to when observers viewed a single scene. However, probability summation successfully predicted the decrement in d-prime for the double-scene condition compared to the single-scene condition. We retested the double- and single-scene conditions, and employed a novel condition where observers viewed 2 animal scenes. Three observers performed a rapid go/no-go animal/non-animal categorization task in all 3 experimental conditions: single-scene, target-plus-distractor, and double-target, with a total of 8,960 trials per observer. Probability summation was successful in accounting for the pattern of d-prime across the experimental conditions.
Consistent with previous modeling work, we found that an amplitude-only classifier could perform our task. If observers used amplitude spectra to perform our task, then resetting amplitude spectra to be the same across all images should lower d-prime. This is what we found. However, altering amplitude spectra introduces distortions that could even affect a phase-only observer. To test this hypothesis, amplitude spectra were swapped across images, but only within an image category. In these conditions, an amplitude-only observer should perform just as well as with the original images. Contrary to the amplitude-only hypothesis observers had lower d-primes for these amplitude-swapped images compared to normal images. Furthermore, this experiment led to the same low d-primes that were obtained in the amplitude-equalized experiment. This is consistent with the idea that the amplitude manipulations in both these experiments introduce distortions affecting our ability to extract phase information.