Multiresolution Textures from Image Sequences

Eyal Ofek, Erez Shilat, Ari Rappoport and Michael Werman

Rendering is one of the most important tasks in computer graphics and animation. It is widely recognized that texture maps are essential for adding to the visual content of the rendered image. Extraction of textures from a single photograph poses severe difficulties and is sometimes impossible, while artificial texture synthesis does not address the full range of desired textures.

In this paper we present a method for computing high quality, multiresolution textures from an image sequence. The method has the following features: (1) it can be used with images in which the textures are present in different resolutions and different perspective distortions; (2) it can extract textures from objects with any known 3-D geometric structure; specifically, we are not restricted to planar textures; (3) removal of directional illumination artifacts such as highlights and reflections; (4) efficient storage of the resulting texture in a multiresolution data structure; and (5) no restrictions are imposed on the computed texture, which can be a constant color texture or a richly colored one.

We present an attractive application of our technique, in which an existing real object participates in an animation sequence and is endowed with synthetic behaviour. For some image examples, click here.

IEEE Computer Graphics and Applications, 17(2):18-29, March 1997. Also Technical Report, Institute of Computer Science, The Hebrew University, 1994.