Abstract
Introduction. A key issue in the encoding of the 3D structure of the world is the neural computation of the multiple depth cues and their integration to a unitary depth structure. This cue integration is often conceptualized as a reliability-weighted summation of all the depth cues to each other, scaled to the unitary world metric for effective action. Analysis. A simple linear weighted summation rule required for weak cues to combine to a strong depth percept would massively overestimate the perceived depth in cases where each cue alone provides a strong depth estimate. For example, if each depth cue alone provided, on average, 0.3 of the veridical depth in the scene, the linear summation of ten depth cues would overestimate the total depth by a factor of three. On the other hand, a Bayesian averaging, or ‘modified weak fusion’, model of depth cue combination would not provide for the enhancement of perceived depth from weak depth cues that is observed empirically. Results. These problems can be solved with an asymptotic, or hyperbolic Minkowski, weighted combination rule, providing for stronger summation for the first few depth cues with progressively weaker summation as the number of cues increases towards the veridical level. However, this rule does asymptote far too slowly to account for classic depth cue summation data. An accelerated asymptotic rule is therefore proposed to fit the empirical strength of perceived depth as the number of depth cues is increased. Conclusion. Despite a substantial literature on the topic, the theoretical analysis of depth cue combination remains drastically incomplete, and some form of accelerated asymptotic cue combination will be required to capture the quantitative phenomenology of human depth perception.