Abstract
We investigated the mechanisms by which perceptual learning improves performance with practice using high- and low-pass filtered external noise. Observers were trained in a 4AFC orientation identification task with Gabor stimuli embedded in external noise at two spatial locations. The spatial frequency characteristics of external noise were systematically manipulated through high-pass and low-pass filtering and added to the signal stimulus. Contrast thresholds were measured as a function of the cutoff spatial frequency of the high-pass and low-pass filters and practice. Adaptive staircases adjusted contrasts to track 79.4 and 70.7 percent correct with 1620 trials per session for a total of 18 sessions. Observers identified the orientation of the Gabor in the pre-cued location in each trial. Different spatial frequency conditions and locations were intermixed randomly during training. This is the first study to use filtered external noises during training. Observers showed systematic reductions in contrast thresholds over the course of training. Although the amplitude of the threshold versus cutoff frequency functions decreased, the relationship between the low-pass and high-pass threshold versus cutoff spatial frequency functions remained substantially equivalent through the course of perceptual learning. The performance returned essentially to initial levels after a switch to new retinal locations of testing, indicating specificity of the learning. A perceptual template model elaborated to estimate spatial frequency sensitivity of the template (Lu & Dosher, 2001) was used to account for the threshold versus cutoff functions through the course of training. We found that perceptual learning improved performance in external noise without changing the spatial frequency tuning of external noise sensitivity, in parallel with previous findings in spatially cued attention (Lu & Dosher, 2004). The improved tuning of the perceptual template with practice may occur in orientation rather than spatial frequency tuning.
Meeting abstract presented at VSS 2016