Stochastic Properties of Segmentation-Rivalry Alternations

Abstract

Segmentation rivalry is the alternation of perceived depth that occurs when depth information from, eg, occlusion and stereo disparity cues are inconsistent (Arrington and Held, 1995 ARVO Abstracts). It was conjectured that the rivalry is based on reciprocal inhibition between end-stopped cell populations that code the same spatial location but opposite direction of surface segmentation in depth. Here the stochastic properties of these depth alternations are analysed and compared to prior analysis [Blake, Fox, and McIntyre, 1971 Journal of Experimental Psychology 88 (3)] of binocular rivalry alternations, which suggested that successive phase durations are approximated by a theoretical gamma distribution: f(x)=b ^a /E^{log \Gamma(a)} x^a−1exp(−bx). Phenomenally in-front (F) versus phenomenally behind (B) phase durations during segmentation rivalry were collected at 2 min arc stereo disparity that produces a maximal alternation rate and ∼50% duty cycle. Least-square error fit to the two duration data sets was very good (F: a=3.69, b=0.514; B: a=6.54, b=0.780) and the chi-square goodness-of-fit showed insignificant (p<0.01) departures from the theoretical gamma distribution when the histogram was smoothed and in some cases when it was not smoothed.

(Mathematical expressions may not appear as intended)