Alpha-matte-based depth map enhancement for hairy objects

Abstract

Most existing depth estimation methods generate an erroneous depth map when a foreground object contains hairy boundaries and this error is very critical in many applications. To solve this problem, a new depth map enhancement method is proposed. The given depth map is enhanced by propagating depth values of the neighbourhood based on an alpha matte. The experiment results show that the boundary of a hairy object is significantly refined by the proposed method.