Rapid Categorisation of Natural Images with Extrafoveal Presentations

Abstract

Last year, we showed that most of the visual processing for categorising centrally presented images could be done in 150 ms. The human subjects were given a go — no go categorisation task in which they had to release a button when they could identify an animal. All the pictures were natural scenes taken from a vast CD-ROM data bank allowing access to several thousands of stimuli so that each image was only seen once. Targets (animals) included fish, birds, mammals, reptiles, insects in their natural environment; distractors included landscapes, trees, flowers, fruits. They were presented for a very short duration (20 ms) in order to avoid eye movement. In the present study we wanted to evaluate how the performance was affected with extrafoveal stimulations. We used the same task, but the visual stimuli could be presented at random, either centrally where subjects had to maintain fixation or laterally (left or right hemifield). Behavioural data and concomitant evoked potentials were recorded. The subject's performance was extremely good with a mean score of 90% correct and a mean reaction time of 500 ms. Comparison of performance between central and lateral stimulations showed that most subjects could process lateral and central stimuli with the same speed (reaction time distributions were not statistically different), and that, although extrafoveal presentations induced a drop in the percentage of correct responses (mean decrease 5%), it was not significant in all subjects. Moreover, when compared with last year's study the increase of the number of spatial locations where the visual stimulus could appear did not induce a significant behavioural cost, in terms of speed or accuracy of the response. We also looked at a possible hemispheric specialisation with no clear results. The results obtained here show that even with shared spatial attention and extrafoveal vision, the human visual system is remarkably efficient in an identification task.