Parallel detection of orientation differences in the presence of orientation gradients

We measured thresholds and reaction times for detecting a target, defined by orientation contrast, as a function of the number of elements displayed simultaneously and of the linear orientation gradient present in the display. This test served to evaluate how well the human visual system is able to ignore smooth gradients in orientation - similar to what it does with gradients of luminance or wavelength. Smooth orientation gradients are common in natural environments as opposed to the usual laboratory (search) experiments. It turns out that targets defined by a discontinuity in the transition between line orientations can be processed in parallel, i. e. that 'search' times increase by between 0.5 and 6 ms, on average, per additional element displayed, irrespective of the number of elements. But thresholds of orientation difference for the detection of the target increase linearly with the orientation gradient present in the display, and tend to increase more strongly for small gradients, indicating a special bonus for (near) collinearity. The averaged data follow a Weber-law type while this is not true for the individual observers' data. These results show that the visual system is indeed able to detect targets based on orientation contrast, rather than on absolute orientation [cf. Nothdurft (1985). Vision Research, 25, 551-560], but that the orientation gradient cannot be ignored.