Efficiency of stereopsis in random-dot stereograms.

The statistical efficiency of stereopsis was investigated by measurement of the discriminability d' of a step edge in depth in a random-dot stereogram and its comparison with d' calculated for an ideal observer model. Efficiencies of approximately 20% were found for stimuli with very few (less than 30) dots. As the number of dots was increased, efficiency fell markedly to 2% and below. Over the ranges explored in this study, there was little variation in efficiency with changes in dot density (with dot number constant) or stimulus size. Efficiency loss was also found to be almost entirely due to incomplete use of the available information rather than to intrinsic observer variance. Results were compared with two simple models that implemented possible detection strategies.