Abstract
Two very different bias functions may be observed in the study of perceived 2D orientation of visual stimuli. One bias function is symmetrical between vertical and horizontal. This “cardinal” bias (e.g., Gischick et al. 2011) has sometimes been argued to be a front-end coding bias due to the overrepresentation of vertical and horizontal in natural scenes and in visual cortex. The second, “categorical” bias function (Dick & Hochstein, 1989) is asymmetrical; it exaggerates angular deviations from horizontal while underestimating deviations from vertical with the consequence that orientations that are only 37-40° from horizontal are judged to bisect the angular distance between horizontal and vertical (Durgin & Li, 2011). Here we report that both of these biases appear to be yoked to the perceived gravitational reference frame: When observers are positioned at an attitude of 45° so that retinal and gravitational reference frames are dissociated, it is the perceived gravitational reference frame that matters most. In the case of the categorical bias, the gravitational horizontal seems to provide a particularly strong reference orientation, consistent with the importance of the gravitational ground plane. In the case of the cardinal bias, which is measured by comparison between two orientations, the gravitational reference-frame dominance is clearly inconsistent with an account based on cortical over-representation of cardinal orientations. More likely, for purposes of comparison, orientations are coded relative to a perceptually-given reference frame that is divided into quadrants. Cardinal bias might therefore be understood as resulting from representational range compression within each perceptual quadrant. In a series of five experiments we show that noisy orientation textures composed of variably-oriented gabor patches can show both kinds of cognitive orientation bias, and that both biases are yoked to perceptual rather than retinal reference frames.
Meeting abstract presented at VSS 2015