--------------------------------------------------------------------------------------------------------------------

                          Surfaces with Occlusions from Layered Stereo

                                            Michael H. Lin
                                         Stanford University

--------------------------------------------------------------------------------------------------------------------

Stereo is one of the fundamental problems of computer vision. Although
steady progress has been made in recent algorithms, producing accurate
results in the neighborhood of depth discontinuities remains a challenge.
Moreover, among the techniques which best localize depth discontinuities,
it is common to work only with a discrete set of disparity values,
hindering the modeling of smooth, non-fronto-parallel surfaces.

We propose a new method which estimates scene structure as a collection of
smooth surface patches. The disparities within each patch are modeled by
a spline, while the extent of each patch is represented via a labelled,
pixelwise segmentation of the source images. Disparities and extents are
alternately estimated by surface fitting and graph cuts, respectively, in
an expectation-maximization-style iterative framework. Input images are
treated symmetrically, and occlusions are addressed explicitly. Boundary
localization is aided by image gradients. Promising experimental results
are presented.

This is joint work with Carlo Tomasi.