How are texture and stereo used in slant discrimination?

When surface slant is specified by disparity and texture gradients, both sources of information are, in principle, useful for estimating slant. Several investigators have argued that combined stereo-texture estimation can be modeled as weighted combination: Sc=WtSt+WsSs (St = texture slant estimate, Ss = stereo estimate, Wt = texture weight, Ws = stereo weight, Ws+Wt=1). Do observers use the combined estimate (Sc) when discriminating surface slants? Three stimuli were presented sequentially in a haploscope. Textures were Voronoi patterns. Two "estandard stimuli" had the same texture- and stereo- defined slants (St=Ss=S0) and one "odd stimulus" had different texture and/or stereo slants (St=S0+c*Dt, Ss=S0+c* Ds). Observers indicated the interval containing the odd stimulus. A staircase varied "c" to determine threshold for various directions (Dt,Ds) in the space of possible stimuli (St, Ss). If observers used St or Ss (whichever provided better information) to do the task, stimuli would be discriminable as soon as either St or Ss reached its independent threshold (ignoring probability summation). If observers used only the combined estimate Sc, then stimuli for which Sc=S0 (i.e., Dt = -(Ws/Wt)Ds) should be indiscriminable for any value of c. Furthermore, performance should be better for cue-consistent stimuli (Ds=Dt) than for single-cue-varied stimuli (Ds=0 or Dt=0). We found that in some cue-conflict directions, thresholds were significantly higher than in the cue-consistent direction (although never infinitely higher). This result indicates that the combined slant estimate is used together with the individual cue estimates in slant discrimination. When Dt and Ds have opposite sign, observers' use of Sc resulted in poorer performance than that predicted by use of St and Ss alone. Thus, observers use all three estimates even when more efficient strategies are available.