We are interested in automatic methods that use scene specific
assumptions, as manual approaches are not scalable for converting
large video libraries or broadcast content. Two automatic techniques
that use priors are Make3D [3] and Photo Pop-up [4]. Such methods
operate by; building statistical models of 3D scenes, segmenting images,
and applying a classifier to assign planar relationships between
segments. These approaches are limited in the types of scenes that
are supported, and provide no temporal stability for video input.