Abstract
Two different biases have been reported for 2D orientation. For estimation and bisection tasks, deviations from horizontal tend to be exaggerated (Durgin & Li, 2011). For noisy orientation comparison tasks, symmetrical biases toward both cardinal orientations have been reported (Girschick et al., 2011; Tomassini et al., 2010). One source of additional bias in a comparison task might be memory encoding. To test for memory-dependent bias effects we developed a forced choice staircase procedure in which observers (N=52) judged, for each presented texture of oriented Gabor patches, whether the mean orientation was clockwise or counterclockwise from a reference line present on the screen. Three different reference orientations were used (15°, 45° and 75°); the standard deviations of orientation for the test textures were 0°, 9°, or 18°. For half the participants, measurements relative to the three reference orientations were blocked so that textures were being compared to the same orientation within each block (low memory load). For the remaining participants the different reference orientations were interleaved in random order so that line orientation had to also be encoded on each trial. If noise-dependent orientation biases are perceptual, these two experimental conditions should produce the same patterns of increased bias toward the cardinal orientations as orientation noise increases. In fact, a reliable Blocking x Noise interaction (p <.05) indicated that cardinal bias was only reliably present when the different references were interleaved (p <.05). When blocked so that a memory representation of the reference stimulus could be used from trial to trial, no reliable cardinal biases were found (p > .10). A follow-up study in which observers lay rotated 45° counterclockwise relative to the display supported the interpretation that the cardinal bias in the interleaved condition was a memory bias relative to the range of reference stimuli presented rather than a cortical cardinal orientation bias.
Meeting abstract presented at VSS 2014