Charles R. Dyer

Professor

Computer Sciences Department
University of Wisconsin
1210 W. Dayton St.
Madison, WI 53706-1685

telephone: (608) 262-1204
fax: (608) 262-9777
email: dyer@cs.wisc.edu
www.cs.wisc.edu/~dyer

Ph.D., University of Maryland, 1979
Interests: Computer vision, image-based rendering, shape representation, motion analysis, visual exploration

Research Summary

As important steps towards making computers see, my research group studies methods for analyzing and synthesizing digital images, focusing on image-based rendering, motion analysis, three-dimensional shape representation, and visual exploration of unknown scenes.

The goal of our image-based rendering work is to generate new views of a real three-dimensional scene by combining a set of images or video of that scene that were taken from other viewpoints. The challenge is obtain photorealistic images and videos without having to first recover a complete three-dimensional scene reconstruction.

One recent result is a technique for interpolating between two views of a dynamic scene. That is, given two input views that were captured at different times and from different viewpoints of a scene that changed in the interim, the problem is to synthesize virtual views that interpolate both viewpoint and scene motion. We have developed algorithms that solve this problem given two widely-separated input views, sparse point correspondences between these two views, and no information about the cameras' positions and orientations. Under certain conditions the moving objects in the scene are guaranteed to move along straight-line, constant-velocity trajectories. Applications of this work include filling in gaps in movies, performing smooth `hand-offs' between cameras at different locations, and creating movies from still images. Current work is investigating extensions that can be used in augmented reality applications. This work is based on our earlier work on view morphing.

We have also developed another method, called voxel coloring, which takes any number of input images, corresponding to arbitrary views around a scene, and reconstructs a photometrically-consistent volumetric (voxel-based) representation of the scene. Using this reconstruction, the scene can be rendered from any view by reprojecting the voxels.

Sample Recent Publications

Interpolating view and scene motion by dynamic view morphing (with R. Manning), to appear in Proceedings of the Image Understanding Workshop, 1998.

Photorealistic scene reconstruction by voxel coloring (with S. Seitz), Proceedings of the Computer Vision and Pattern Recognition Conference, pp. 1067-1073, 1997.

View morphing (with S. Seitz), Proceedings of the SIGGRAPH Conference, pp. 21-30, 1996.

This page was automatically created December 30, 1998.
Email pubs@cs.wisc.edu to report errors.