Sharpening coarse-to-fine stereo vision by perceptual learning: asymmetric transfer across the spatial frequency spectrum

Abstract

Neurons in the early visual cortex are finely tuned to different low-level visual features, forming a multi-channel system analysing the visual image formed on the retina in a parallel manner. However, little is known about the potential 'cross-talk' among these channels. Here, we systematically investigated whether stereoacuity, over a large range of target spatial frequencies, can be enhanced by perceptual learning. Using narrow-band visual stimuli, we found that practice with coarse (low spatial frequency) targets substantially improves performance, and that the improvement spreads from coarse to fine (high spatial frequency) three-dimensional perception, generalizing broadly across untrained spatial frequencies and orientations. Notably, we observed an asymmetric transfer of learning across the spatial frequency spectrum. The bandwidth of transfer was broader when training was at a high spatial frequency than at a low spatial frequency. Stereoacuity training is most beneficial when trained with fine targets. This broad transfer of stereoacuity learning contrasts with the highly specific learning reported for other basic visual functions. We also revealed strategies to boost learning outcomes 'beyond-the-plateau'. Our investigations contribute to understanding the functional properties of the network subserving stereovision. The ability to generalize may provide a key principle for restoring impaired binocular vision in clinical situations.

Keywords: generalization; specificity; stereopsis; vision enhancement; visual plasticity.

Figures

Figure 1.

Stereo stimulus. (a) The stereogram consisted of two slightly different pictures—one to each eye. At the centre each square was a target Gabor patch surrounded by four reference Gabor patches. To eliminate any possible monocular cues, the vertical and horizontal coordinates of each Gabor patch and also the patch features, the carrier phase, were randomly jittered according to a uniform distribution. A custom-built mirror stereoscope was used to view the stereo pairs, so that the left eye would see the left square and the right eye would see the right one. Binocular disparity was generated by shifting the two target Gabor patches, one on each side, horizontally in opposite directions (uncrossed disparity, both shifted temporally; crossed disparity, both shifted nasally). (b) Binocular fusion of the two monocular images creates a cyclopean image. The visual task was to determine the stereoscopic depth of the target Gabor (in front/behind) relative to the four adjacent references. This schematic diagram illustrates crossed disparity—the target Gabor patch appeared in front of the reference patches.

Figure 2.

Experiment 1. Perceptual learning of stereoacuity: specificity for carrier orientation and spatial frequency. (a) The training protocol consisted of three training stages: stage 1, V5 (vertical carrier: 5 cpd); stage 2, H5 (horizontal carrier: 5 cpd); stage 3, V10 (vertical carrier: 10 cpd). A 3-parameter exponential function was used to quantify the learning profile. Mean thresholds (n=10) for each of the three stimulus configurations were measured before and after each training stage. Error bars indicate the standard error of the mean unless stated otherwise. (b) The pre- and post-training threshold data of individual observers (n=10) are illustrated in the nine figure panels: 1st row, s15 versus s1; 2nd row, s29 versus s15; 3rd row, s43 versus s29. Arrows indicate the sequence of direct training from stage 1 to 3. White panel area denotes statistically significant. Grey panel area, not statistically significant. Note that the abscissa label for each row is displayed at the right-hand end of all the panels highlighted in dark blue.

Figure 3.

Experiment 2. Bandwidth of generalization across the spatial frequency spectrum. (a) Mean stereoacuity as a function of session. In stage 1, group LH (first row, n=10) was trained with V1 stimuli and group HL (second row, n=11) was trained with V20 stimuli. In stage 2, observers crossed over and trained at the untrained spatial frequency: group LH, V20; group HL, V1. A 3-parameter exponential function was used to quantify the learning profile. (b) Mean stereoacuity across spatial frequencies. (c) Per cent improvement in mean stereoacuity (I) as a function of spatial frequency (f). A 3-parameter Gaussian function, I=It×e−(1/2)(f−ft/σ)2, was used to quantify the generalization of stereoacuity learning across spatial frequencies, where It is the per cent improvement occurring at the trained spatial frequency (ft) and σ denotes standard deviation.

Figure 4.

Enhancing coarse-to-fine stereoacuity with perceptual learning. Mean per cent improvement in stereoacuity resulting from direct training as a function of spatial frequency (stage 1 of the two experiments; V5 from the first experiment; V1 and V20 from the second experiment). There was no significant difference in per cent improvement among the three frequency groups (ANOVA: F=0.795, p=0.461).