Abstract: PURPOSE: Although previous studies consistently found that perceptual learning of stereopsis significantly improved stereoacuity, the underlying mechanism of stereo learning remains hot debate. Here, we practiced subjects to judge stereopsis that defined by three types of binocular disparity (i.e. zero-, first-, and second-order), aiming to better understand the characteristic of stereo learning within a distributed and cascade plasticity framework.
METHODS: Thirty subjects were randomly assigned into three training groups (each for zero-, first- or second-order disparity, separately). Disparity thresholds were measured before and after 8-10-days of training. Ten additional control subjects only took disparity threshold assessments. Stereoscopic images were delivered through shutter goggle synchronized to the computer. Stereoacuity was assessed with the Titmus Fly Test before experiment to ensure all subjects had normal stereoscopic percepts.
RESULTS: Training significantly improved disparity threshold by 66.82%, 42.86% and 35.38% in the zero-order training condition, 55.08, 80.06% and 44.02% in the first-order training condition, and 50.67%, 52.67% and 86.86% in the second-order training condition for zero-, first- or second-order disparity, respectively. There was no significant improvement for the control group. Interestingly, the greatest improvement in first-order and second-order disparity threshold was found in corresponding disparity training group; on the contrary, improvement in zero-disparity threshold was comparable in all three disparity training groups.
CONCLUSIONS: We found both general (related to zero-disparity) and specific improvements (related to first- and second- order disparity) in stereo learning. We concluded that stereo training occurs at different visual processing stages and the effects of training might depend on the specific training sites.

Key words: perceptual learning, stereopsis, binocular disparity, transfer