Abstract
The physical features used when human subjects are categorizing natural images on the basis of their context are still poorly known. It is usually believed that amplitude informations are sufficient to categorize scenes.
In the present study, we first equalized amplitude and furthemore modified phase informations in a rapid categorization task in which 1440 achromatic naturals images were categorized as “Man-made” or “Natural”. In a first experiment, phase information was preserved but amplitude information was equalized by applying the amplitude spectrum averaged across the two categories on each stimulus. Twelve subjects were tested. In a second experiment using the amplitude equalized scenes, new stimuli were built by mixing a variable percentage of random noise phase with the original scene phase, creating 10 new sets of stimuli from 0–99% random noise phase by 11% steps. Ten human subjects were used to counterbalance all conditions. Each subject saw every scene at only one noise level. In both experiments, subjects were performing a go/no-go rapid visual categorization task; scenes were flashed for only 26 ms, subjects were responding with a finger lift to targets and kept pressing a button with distractors. They performed successively two blocks of context categorization, using either “Man-made scene” or “Natural scene” as targets.
When amplitude spectrum is equalized, a very little impairment (greater with manmade targets) is seen on performance. When noise was gradually added to phase components, performance was not gradually affected. It remained stable up to 66% of normal phase where a sharp performance decrease was observed with 55% added noise and up to 99% added noise. With high noise conditions subjects were inhibiting their go responses, a biais that was even stronger in the “Man-made” task.
Categorization can be done with enough phase information when no amplitude information can be used.