Purchase this article with an account.
Rosemary Le, David Alex Mely, Thomas Serre; Computational Mechanisms Responsible for the Hermann Grid Illusion. Journal of Vision 2014;14(10):58. doi: https://doi.org/10.1167/14.10.58.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The Hermann grid is a well-known illusion. In its classical form, one perceives non-existent gray spots at the intersections of a white grid over a black background. Textbooks attribute the phenomenon to the center-surround organization of retinal ganglion cells. But in recent years, variations of the illusion have been created demonstrating that the center-surround organization cannot be the sole mechanism. While many qualitative theories have been proposed, no computational model has yet been shown to account for all variations. Here we consider several computational models of early vision including a baseline model of retinal ganglion cells, as well as increasingly more sophisticated models of the primary visual cortex (V1) that include divisive normalization, cardinal bias, and orientation-dependent lateral connections. We conducted a psychophysics experiment where participants (n=20) ranked multiple variations of the illusion according to their relative strength. Many of the variations used were created by researchers who previously studied this illusion. Together, the illusions ranged from non-existent to extremely strong. The average of the participants' rankings produced a ground truth against which model output rankings were compared. Spearman's correlation measured the consistency of the model's ranking to the ground truth. Model parameters were constrained by neurophysiologically data and optimized to best fit subjective illusion strength data. We find that the most complete model of V1 (which includes normalization, cardinal bias, and lateral connections) is the best predictor of human illusion perception when compared against simpler models. Our results thus confirm that the origin of the Hermann grid illusion is cortical in nature and that the relative strength of its variations appear to stem from the complex interaction of several well-established cortical processes.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only