Purchase this article with an account.
Frouke Hermens, Gediminas Luksys, Wulfram Gerstner, Michael H. Herzog; Visual backward masking: Feed-forward or recurrent?. Journal of Vision 2005;5(8):764. doi: 10.1167/5.8.764.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In the past years there has been a renewed debate on the mechanisms of backward masking. Some of the existing models assumed that masking occurs by the interactions between two feed-forward processing streams (e.g. Breitmeyer, 1984). The recently introduced object substitution model assumes that masking occurs in recurrent architectures in which the mask replaces the target representation (Di Lollo et al., 2000). Here we show that a 2-dimensional, one-layer Wilson-Cowan type model, which relies on lateral connections only, is able to describe a broad range of masking effects. These effects include masking by light, noise, pattern, and shine-through masks. The model mimics processing in the visual area V1 and comprises an excitatory and an inhibitory layer only. Neural activities corresponding to the edges of an object are dynamically enhanced while inner elements are suppressed. In the model, masking occurs when edge and target activity interfere with each other. When homogeneous structures of the mask are filtered out before corresponding neural activities can interfere with the target activity, masking is diminished. On the basis of our model simulations we propose that masking effects occurs during basic early visual information processing such as object contour detection. At least some masking phenomena can be described by a single layer network without the need for interactions between feed-forward processing stages or between higher and lower-level areas.
This PDF is available to Subscribers Only