Purchase this article with an account.
Rashmi Sundareswara, Paul R. Schrater; A Perceptual inference model for bistability. Journal of Vision 2007;7(9):803. doi: 10.1167/7.9.803.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In previous theoretical and empirical work (Schrater and Sundareswara, NIPS, 2006; Sundareswara and Schrater, VSS 2006), we showed how Bayesian models of perceptual inference could give rise to bistability when implemented via a sampling process. We also showed that flanking a bistable figure (Necker cube) with fields of similarly oriented background objects increase the time spent in the Necker percept matching the background. The dynamics of this context effect was in quantitative agreement with the predictions of the Bayesian model assuming that the statistics of the background act as a contextual prior that changes the relative heights of a bi-modal posterior distribution on the shape and orientation of the bistable figure. The goal of the present study is to test predictions of the proposed linkage between bistability and Bayesian inference. In particular, we tested for effects of background objects on Necker Cube switching dynamics for three context manipulations: 1) changing the orientation similarity between Necker percepts and background cubes in the context; 2) changing the spread of the orientation distribution of background objects 3) changing the shape of the background objects to minimize feature similarity to the Necker Cube, while maintaining orientation information. Perceptual states were recorded across time using a method similar to (Mamassian and Goutcher, 2005) and data was analyzed using a Markov Renewal Process (MRP) framework. The analysis generates empirical distribution functions for state-contingent first transition and survival probabilities. Our results show that increasing the match between Necker percept and background orientations and the spread of the background orientations parametrically decrease the context effect. These effects are weaker but similar when the background shapes are non-cuboidal. The results are in quantitative agreement with the predicted effects of our Bayesian model of bistability.
This PDF is available to Subscribers Only