Human visual search follows a suboptimal Bayesian strategy revealed by a spatiotemporal computational model and experiment

摘要

a The stimulus image was a 1/f2 noise image (15° diameter, root-mean-squared (RMS) contrast 0.2) with an embedded target. The target was a 6 cycle/degree Gabor grating with a diameter of 0.3°, oriented 45° counterclockwise from the vertical and windowed by a symmetrical raised cosine (half-height width of one cycle). The area outside the image was set to the mean gray value of the image (0.5, not shown here). b, c Detection rate (b) and foveal target visibility (c) as a function of the target RMS contrast. Dots are raw data (one color per subject). In b, the curves are the Weibull function fit to the dots. In c, the curves are calculated from Weibull functions fit to hit rates and the correct rejection rate of the raw data (Eq. (1)). The intersection between the dashed line (d′ = 3.0) and the curves is the target RMS contrast chosen for each subject. d–i Target visibility map at different stimulus exposure times in the detection task. The dots are raw data merged from four subjects. The surface is the fitted visibility map function (Eqs. (10)–(12)). Not all locations were tested in all exposure times so the number of dots may differ in each subplot. Hori/Vert: horizontal/vertical dimension of the search field. Fixation location was at the center of search field.