Abstract
Humans can discriminate the mean orientation of an array of Gabors. Judging mean orientation may reflect the activity of texture processing mechanisms or pooling of orientation signals. We investigated whether humans can discriminate the mean orientation of second-order Gabors.
Methods: Observers discriminated whether the mean orientation of arrays of 8 or 16 Gabors was to the left or right of vertical. The Gabors were either first-order (luminance modulation of a noise mask) or second-order (contrast modulation of a noise carrier). The arrays could contain, either one type of Gabor, or a combination of half of each Gabor type. The distribution of orientations in an array could be, very narrow (i.e. aligned), very broad (high orientation noise) or half drawn from an orientated distribution and half randomly orientated. We estimated the amount of internal noise affecting the task and the number of samples used. The visibility of the Gabors was matched by measuring the minimum discriminable orientation difference for single Gabors.
Results: Mean orientation thresholds increased with the width of the orientation distribution. Observers' performance was similar for first-order, second-order and combined arrays when the orientation of all the Gabors was drawn from one distribution. Observers' performance with 8 first-order Gabors did not change when 8 randomly orientated second-order Gabors or low contrast first-order Gabors were added but adding high contrast first-order Gabors increased orientation thresholds. Observers performance with 8 second-order Gabors was slightly reduced by adding randomly orientated first-order Gabors and greatly impaired if the observer did not know which type of Gabor held the orientation information.
Conclusion: Observers are able to integrate first and second-order orientation across space. This integration process can, under some conditions, combine first and second-order information.
Support: NSERC RGPIN 46528-01