Evidence of non-linear combination of stereo and motion information

Current theories of depth-cue combination postulate independent modules in the visual system for processing separate cues to depth. Further, the outputs of these modules are associated with weights based on the reliabilities of the depth estimates, as described by the modified weak fusion (MWF) model (Landy et al., 1995). In this study, the MWF model was tested by asking observers to provide depth judgments from (i) stereo-alone, (ii) motion-alone, and (iii) stereo and motion combined. The observers' task was to judge whether the apparent elongation of a horizontally oriented cylinder was more or less deep than an apparently circular cylinder (ACC) — see Johnston et al. (1994). The MWF model predicts that the mean estimate of observers' settings in the combined-cue condition will be a linear combination of the depth magnitudes from the stereo-alone and motion-alone conditions, each weighted by a reliability measure computed from the variances of observers' judgments in the two conditions. Our results do not conform to the predictions of the MWF model: perceived depth judgements were overestimated in the combined-cue and motion- alone conditions whereas stereo-alone judgements were veridical, despite the greater variance (less reliability) of the motion-alone condition. On the other hand, the MWF model predicts that perceived depth in the combined-cue condition should be closer to the depth estimate based on the more reliable cue (stereo, in our investigation). The present results are predicted by a new theory (Domini, Vuong, and Caudek; Di Luca, Domini, and Caudek ) which postulates that the visual system takes into account natural covariation of depth cues.