| | |
Summary: Bayesian Video Matting Using Learnt Image Priors
Nicholas Apostoloff and Andrew Fitzgibbon
Robotics Research Group, University of Oxford, Oxford, OX1 4AJ, UK
Video matting, or layer extraction, is a classic inverse problem
in computer vision that involves the extraction of foreground
objects, and the alpha mattes that describe their opacity, from
a set of images. This paper presents a method inspired by nat-
ural image statistics where a second order prior is learnt that
models the relationship between the spatiotemporal gradients
in the image sequence and those in the alpha mattes. This is
used in combination with a learnt foreground color model and
a prior on the alpha distribution to help regularize the solution
and greatly improve the automatic performance of the system.
Video matting is similar to the computer vision problem
of layer extraction [1, 2, 7, 12, 13, 14], but has a stronger
emphasis on deriving object boundaries which accurately
represent the sub-pixel blending of foreground and back-
ground layers. This problem is interesting because it is a
difficult inverse problem: the number of unknowns exceeds
the number of measurements, so regularization is crucial to
|
