Combination of Cues and Priors in Depth Perception

Various visual cues provide information about depth and shape in a scene. In this paper , research on depth cue combination is reviewed. Using normative considerations, we have developed the \Modiied Weak Fusion" (MWF) model of cue combination. This model includes a simple, linear cue combination rule (weighted averaging of depth estimates). But, it also raises three key issues for depth cue combination in human vision that lead to apparent nonlinearities and interactions between cues: (1) cue promotion (most cues must be scaled and disambiguated before averaging is a meaningful operation), (2) dynamic weighting of cues (a cue's weight depends on scene content, including the weight of other cues), and (3) robustness of cue combination (a cue's weight may change with the degree of discrepancy of the estimate of depth derived using that cue relative to the estimates from other cues present in the scene). We have perceptual evidence for all of these processes including the linear depth combination rule, cue reweighting with changes in cue reliability, cue reweighting with changes in depth discrepancy between cues, and cue interactions contributing to cue scaling (that is, to an estimate of the viewing distance used to scale velocities or disparities to depth values). Finally, we summarize more recent work concerned with the way in which diierent cues' a priori biases interact to disambiguate simple 3-D stimuli.