Fig. 1.

Illustrations of cue incongruence and the performance of previous models relative to psychophysical data. a The space of possible slants depicted by texture and disparity cues. To illustrate cue integration we provide a cartoon of the slant of 3D surfaces defined by congruent (top) and incongruent (bottom) cues for viewing through red–cyan anaglyph glasses. Sensory estimates are modelled as Gaussian distributions. For congruent cues (top row), fusion (purple curve) produces a more reliable estimate (i.e., lower variance) than for either component (orange, blue curves). For incongruent cues (bottom row), accuracy is maintained by down-weighting the influence of the less reliable cue. b Illustration of the bias (offset in the mean) and reliability (inverse of the variance) parameters used to describe the data. c, d Behavioural measures extracted from Girshick and Banks⁷ showing c perceived slant (quantified by bias away from the more reliable cue) and d reliability as a function of cue conflict. Pink and green curves indicate the predictions made by the maximum likelihood model and the Normalization model²⁵, respectively