Bayesian Modelling
 


line drawing bayes -- pascal mamassian

Visual perception is fundamentally ambiguous because any single retinal image is consistent with an infinite number of three-dimensional scenes.  With the help of Bayesian modelling, we are interested in unravelling the prior constraints used by the human visual system to disambiguate the retinal images.  Examples of prior constraints under study include the assumptions that the light source is located above the observer and that the observer's viewpoint is located above the attended object.

Collaborators:

Mike Landy (NYU), Larry Maloney (NYU) and Dan Kersten (Univ. Minnesota).


Further references:

Kersten, D., Mamassian, P. & Yuille, A. (2004). Object perception as Bayesian inference.  Annual Review of Psychology, 55, 271-304.

Mamassian, P. & Landy, M. S. (1998).  Observer biases in the 3D interpretation of line drawings.  Vision Research, 38, 2817-2832.

Mamassian, P. & Landy, M. S. (2001).  Interaction of visual prior constraints.  Vision Research, 41, 2653-2668.

Mamassian, P., Landy, M. S. & Maloney, L. T. (2002).  Bayesian Modelling of Visual Perception. In R. Rao, B. Olshausen and M. Lewicki (Eds.) Probabilistic Models of the Brain: Perception and Neural Function (pp. 13-36). Cambridge, MA: MIT Press.


Notes on the figure:

The figure illustrates the 3D interpretation of curved contours in an image. Such a mapping between the image and the surface of an object is valid only if the observer is looking at the object from above. This assumption was revealed thanks to a combination of visual psychophysics and Bayesian modelling (Mamassian and Landy, 1998). Permission is hereby granted for non-profit education purposes as long as proper credit to Pascal Mamassian is provided.

Back to Pascal's homepage