PERCEPTUAL-LEARNING IN PREATTENTIVE VISION

Preattentive vision is a stage of visual processing which operates in parallel across the visual field. This stage can be modelled by assumi ng two filtering stages, a standard linear spatial filter (simple cell ) followed by a coarser spatial filter (second stage filters). A quant itative analysis of this model shows that human performance on some ba sic visual tasks such as texture segmentation, can be accounted for by these local, attention-free and bottom-up processes. However, while t exture segmentation seems to involve only simple processes, observers show a slow improvement over time when performing segmentation tasks. Performance on texture segmentation tasks improves during the first 3- 4 days of practice, then saturates and stays at about the same level f or as much as 3 years without further practice. This learning is speci fic for target location, background orientation and eye of origin. The specificity of the learning process implies plastic changes at a leve l of processing where the specific features are separately encoded; th us monocular learning would imply the existence of modifiable connecti on in cortical visual area V1. These changes may involve strengthening of long range connections between orientation-selective cells (second stage of filtering), thus reducing sensitivity to the homogeneous and redundant background. Experiments involving detection of a low contra st target flanked by high contrast masks show an increase of interacti on range with practice. Longer range interactions seem to be generated via chains of local interactions by some associative (Hebbian) proces s.