Visual processing is generaly understood as resulting from a complex and iterative processing. After discussing six models of visual object recognition, we bring out that the time required to perform complex visual processing is a crucial criterion for understanding vision. We first propose an experimental categorization task in order to determine the time required by subjects to decide wether an animal is present in photographs of natural scenes. Results show that the minimal visual processing time takes less than 150 milliseconds. This speed of visual processing is particularly robust : this latency does not change when presentations are extrafoveal, when the attention is not focused on the stimulus locus, and when the colors of the images are removed. This speed of visual processing does not differ significantly when stimuli are simple shapes, or when subjects have to decide whether the images are in color. Moreover, electrophysiological signs of an implicit categorization of « Animal » scenes appear significantly when new subjects are performing another task. These results show that this task can be achieved by preattentive, parallel, feedforward visual processes. Focused attention, color and foveal acuity are not required to perform this task rapidly.The cerebral bases of this task have been found with dipolar modelisation of ERPs, and with a new event-related fMRI protocol. The task differentialy activates mainly the areas 19 and 31, the fusiform and cingulate gyri. Less activity was found following target trials than distractor ones, which we interpret as inhibition processes, compatible with the competitive attentional models. Visual object recognition could be then considered as simple feature parallel detections. The visual decision (adequation between the actual stimulus and the task performed) could be achieved via inhibition of alternative representations, that could be acts as attentional selection.
