Generative Phenomenology of form perception: Perceptograms and cortical models for amblyopic form distortions
We introduce Generative Phenomenology: making viewable images of perceptions (Perceptograms) and generating the images from neural models, as a powerful technique for understanding the neural bases of perception. Amblyopia, a disorder of spatial vision, provides a perfect case because signals from the two eyes go through partly different cortical neurons, and many amblyopes report form distortions when viewing sinusoidal gratings through their amblyopic eye (AE) but not through the fellow eye (FE). Using a dichoptic display, we acquired high-fidelity perceptograms for 24 gratings shown to AE while sums-of-gratings plaids were shown to FE with contrast, frequency, phase, and orientation of the plaid gratings adjusted to match the two percepts exactly. Plaids provided exact matches to 92.6% of distortions. A formal equation that the signals generated in visual cortex by the test gratings seen through AE match the signals generated by their matched perceptograms seen through FE for each observer, was used to analytically derive cortical filters processing AE signals as linear transforms of standard steerable filters modeling normal V1 neurons for FE. Passing gratings through AE filters accurately generated the measured perceptograms. The filter transformations reflected complex changes in V1 receptive fields and possibly in cross-correlations. The AE filters also explained amblyopic deficits in perceiving sinusoidally modulated circular contours and were consistent with orientation perceptive fields estimated from reverse-correlation experiments. Changes in neuronal receptive fields thus have profound effects on perception, to the extent that observers can see more features than are present in the viewed stimulus.