Abstract
Purpose: Human visual sensitivity for high spatial frequency stimuli, varies as a function of orientation, with lowest sensitivities at oblique orientations (i.e. the oblique effect). However, we have recently shown (ARVO, 2001) that with stimuli of broadband spatial content (either noise or natural scenes), performance is worst at horizontal, best at obliques, and intermediate at vertical orientations. Here, we investigate this “horizontal effect” with images of outdoor scenes that were dominated by scene content of one of four orientations (0°, 45°, 90°, or 135°) to determine whether the orientation of the predominant content (i.e. semantic content and/or phase relations) altered the ability to detect oriented structure. Methods: Sets of natural scenes were selected on the basis of having predominant content (Fourier amplitude) at one of the four orientations, and a fifth set of non-oriented content. The orientation bias in amplitude was removed by filtering the images to be isotropic. Ability to detect an added increment of amplitude (within an orientation band) at one of the four orientations was measured for the five scene types. Stimulus presentation was 400 msec followed by a noise mask and d' was calculated. Results: Regardless of the orientation of the content bias depicted in the original scene, the ability to detect an oriented increment was poor for the horizontal orientation, as reported previously. However, performance was worst for the orientation of the test increment that matched the natural scenes' bias, even though the amplitude bias had been removed prior to testing. Conclusions: We attribute the horizontal effect to an orientation bias inherent in the visual system's processing of broadband stimuli that may have evolved to discount the horizon thereby emphasizing other content. The present results suggest an additional orientation bias that is related to the phase spectrum (or semantic meaning) of the scene.
Office of Naval Research grant# 000149910516