Recovering a 3D shape from a single 2D image: computational model and psychophysics

CIS Colloquium, Sep 30, 2009, 11:00AM – 12:00PM, Wachman 447

Recovering a 3D shape from a single 2D image: computational model and psychophysics

Tadamasa Sawada, Purdue University

I will present a new computational model that recovers a unique 3D shape from a single 2D image by applying a priori constraints: symmetry, planarity, maximum compactness and minimum surface area. The recovered 3D shape is symmetric or approximately symmetric. The model does not use information about the surfaces of the 3D shape. Instead, the model uses information about contours in the 2D image and it recovers their 3D interpretation. The model’s performance is strongly correlated with the performance of human subjects. I will conclude by discussing the problem of identification whether a 2D image is an image of a 3D symmetric shape. I will present results of psychophysical experiments and a computational model explaining the underlying perceptual mechanisms.