Silhouette Extraction from Multiple Images of an Unknown Background

Silhouette information is important in many applications such as virtual reality and visual hull construction. However, silhouette extraction from images relies on a known background. In this paper, we propose a method for silhouette extraction from an arbitrarily unknown background. To do this, we use multiple calibrated images. Each input image is first segmented in regions. Then, the consistency constraints in individual image and multiple views as projection and geometric intersection are enforced to carve out the silhouette in each image. As an extension, we also propose a space-carving-like method to produce the visual hull of the object without the silhouettes. The methods are intensively demonstrated on real examples.