Stereo Vision for View Synthesis

Daniel Scharstein
Cornell University
Ithaca, NY 14853, USA


We propose a new method for view synthesis from real images using stereo vision. The method does not explicitly model scene geometry, and enables fast and exact generation of synthetic views. We also re-evaluate the requirements on stereo algorithms for the application of view synthesis and discuss ways of dealing with partially occluded regions of unknown depth and with completely occluded regions of unknown texture. We present experiments indicating that it is possible to efficiently synthesize realistic new views even from inaccurate and incomplete depth information. We argue that image-based scene representations are ideal for view synthesis applications.

In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'96), pages 852-858, San Francisco, CA, June 1996.