Kisilev, Pavel; Freedman, Daniel
Keyword(s): color transforms, recoloring, relighting, shadow removal
Abstract: In this paper, we present a unified approach for the problem of learning color transforms, applications of which include shadow removal, object recoloring, and scene relighting. The detection of source and target regions is performed using a Bayesian classifier. Given these regions, the learned transform alters the color properties of the target region so as to closely resemble those of the source region. The proposed probabilistic formulation leads to a linear program (similar to the classic Transportation Problem), which computes the desired transformation between the target and source distributions. This formulation allows the target region to acquire the properties of the source region, while at the same time retaining its own look and feel. Promising results are shown for a variety of applications.
External Posting Date: September 6, 2009 [Fulltext]. Approved for External Publication
Internal Posting Date: September 6, 2009 [Fulltext]