Abstract
Shading patterns on a corrugated, textured surface contain two signals: first-order modulations of luminance (LM) and correlated second-order modulations of the local luminance amplitude (AM). Human vision is sensitive to both of these signals, and their alignment is beneficial as a cue to shape perception. Experienced observers see LM and AM gratings aligned in-phase (LM+AM) as shaded corrugations, and anti-phase (LM-AM) as flat reflectance changes: when the two mixtures are presented together in a plaid. First, we trained naïve observers with strong, trial-by-trial feedback. LM/AM mixes were presented in separate spatial locations and feedback consisted of a disparity defined corrugated surface superimposed on the LM+AM cue; LM-AM cues were paired with a flat surface. Performance improved to a maximum over the first hour of training, after which there was no further improvement. Even though this suggests that rapid perceptual learning occurred, when we flipped the feedback to reinforce the LM-AM cue as corrugated, observers flipped their responses without any deterioration in performance. This sudden reversal suggests that the performance benefit was at a categorical rather than perceptual level. Subsequently, we trained new observers with intermittent feedback to reinforce the perception of LM+AM as corrugated. This time, the LM/AM mixes were presented in a plaid for a brief duration (250 ms). Observers showed gradual increase in performance, over a period of 5 days, as they learned to discriminate the mixtures at a perceptual level; post-training AM thresholds were lower than pre-training thresholds. This benefit transferred to 45° rotations of the plaids, and also to plaid combination angles close to 90°, but failed to transfer to higher spatial frequencies or to non-orthogonal plaids. We conclude that the alignment of LM and AM as a shading cue can be learned perceptually, for briefly presented stimuli, given reinforcement by indirect feedback.