In the human visual system, latency of visual information is known to be related to luminance. In other words, for an image containing a foreground brighter than a background, the arrival of visual information onto the first stage of processing is asynchronous as stimuli corresponding to brighter signals are treated before. In this paper, this hypothesis is taken into account for designing an architecture aimed at extracting edges. An account of its performance is given through a formalization of the figure-ground separation problem and comparisons between synchronous versus asynchronous processing are presented for two different architectures, feedforward and recursive with diffusion. Conceptually, asynchrony is also shown to justify the iterative process required by the diffusion and leads to the notion of active segmentation.