Miscalibration of depth cues in the developing visual system

摘要

Perception of depth from both monocular and binocular cues develops in the first year of life. However, reduction of discrimination thresholds via weighted averaging of depth cues does not develop until much later in childhood (Nardini, Bedford & Mareschal, PNAS 2010). Cue combination models often assume that cues are well calibrated (unbiased). In that case the optimal strategy is to take a reliability-weighted average. If, however, depth cues are not well calibrated in young observers, then averaging them may not be the optimal strategy. To investigate this possibility we measured the integration, weighting, and calibration of two depth cues in adults (N=3) and 5- to 8-year-olds (N=20). Observers judged which plane, a 45?° standard or a variable comparison, was more slanted. Slant was defined by a texture gradient, a binocular disparity gradient, or both together. We measured (1) slant discrimination thresholds when given single vs combined cues, (2) the relative weighting for texture vs disparity, measured by comparing consistent with ?±5?° conflicting planes, and (3) the relative bias in slant perceived via texture vs disparity, measured by comparing texture-only with disparity-only planes. Adults reduced their discrimination thresholds when given combined cues vs the best single cue, and gave more weight to the more reliable cue; children did not reduce their discrimination thresholds or give more weight to the more reliable cue. While adults showed small (median 3?°) biases between cues, children showed much larger biases: median 14?°, including biases greater than 22.5?° in 1/3 of observers. These results indicate that in mid-childhood, the visual system is still learning to calibrate depth cues against each other. This raises the possibility that the developing visual system does not combine depth cues because of calibration issues, in line with ideal observer strategies for dealing with biased cues (Ernst & di Luca, 2012).