Visual grouping and contour perception.

The response to a sensory stimulus depends heavily on its spatial and temporal context. A wide number of studies have investigated contextual effects of perceptual and neurophysiological responses. These studies primarily focused on the primary visual cortex (V1), reporting that responses of V1 classical receptive fields are influenced by the surrounding context. One type of observed contextual modulation in V1 is collinear facilitation, which entails the enhancement of neural responses to an element surrounded by collinear elements. Such facilitatory interactions are thought to form the neural basis of contour perception. The dependence of these interactions on the relative orientation and position of elements has been described in terms of an association field.
A number of computational models have used the association field concept to predict human performance in detecting a contour embedded in a background of randomly oriented elements. These models typically construct a saliency map of the image indicating which elements are likely to be part of a contour. Their validity can be assessed by evaluating how well they can predict the image regions that will attract eye movements during contour integration. In the first study, observers' eye movements were recorded during a contour integration task. An association field model was able to predict saccade targets. In addition, we showed that fixation duration and saccade size followed a time course which depended on the saliency and percept of a contour.
The results of the first study showed that the presence of potential contours in the image influenced observers' eye movements during contour integration. This suggests that an initial, but still incomplete, saliency map actively guided eye movements. However, it remained unclear which brain processes were involved in the presaccadic selection of a saccade target. In a second study, we aimed to examine whether presaccadic EEG activity is modulated by the presence of a contour or a region with high association strength in peripheral vision. We found that presaccadic EEG activity, mainly over parietal and occipital brain areas, was predictive of the distance between the saccade landing position and the contour. In addition, when a contour was absent, presaccadic activity predicted the association strength at the saccade landing position. Our results suggest that the presaccadic amplitude reflects the degree to which top-down processes can override bottom-up saliency.
Other spatio-temporal contextual effects on low-level visual processing have been studied in the context of apparent motion (AM). AM refers to the percept of motion occurring when two stationary stimuli are alternately presented at two different locations. It has been found that the detectability of stimuli is reduced in the presence of AM. Previous studies have attributed such masking to interference caused by AM-induced excitation, claiming that V1 neurons respond as if a stimulus is physically moving. In a third study, we investigated this claim by modelling grating detectability during AM using a physiologically inspired population code model. The model predicted only a small amount of V1 activation, which could not account for the observed masking nor for any perceptual completion of the motion path. Our model revealed that AM masking is instead due to strong suppression of V1 responses, which is consistent with the theoretical framework of predictive coding.
A fourth study provided further evidence in favor of AM-induced suppression by measuring and modelling perceived grating contrast during AM. Multiple studies have found that the perceived contrast of a grating is reduced when V1 responses to that grating are suppressed. If AM indeed suppresses neural responses to gratings at the level of V1, the perceived contrast of the grating should be reduced in the presence of AM. A population code model similar to the model presented in the third study provided a full account of performance in a contrast discrimination task. The model indeed revealed a reduction in perceived contrast caused by strong AM-induced suppression. A model only incorporating AM-induced excitation could not account for the data.