Application of stochastic visual textures to the analysis of central visual processing

The human visual system readily identifies a region of visual space which differs from its milieu. This identification process proceeds rapidly and in parallel, without focused attention. The authors use a novel class of stochastic visual textures to define the spatial extent of neural interactions involved in this process. Because the interactions are phase-sensitive and appear to detect local texture features, these feature-detection computations are generally thought of as local processes. However, given the highly nonlinear nature of the cortical interactions, separation of local from longer-range interactions is not a simple matter. The authors approached this question by investigating the dependence of the visual evoked potential to check size and correlation structure. These manipulations lead to hypotheses for the spatial organization of the neural computations.