HP Labs Technical Reports

Automatically Synthesising Virtual Viewpoints by Trinocular Image Interpolation - Detailed Report

Pollard, Stephen; Hayes, Sean; Pilu, Maurizio; Lorusso, Adele

HPL-97-166

Keyword(s): image processing; image based rendering; computer graphics; stereo matching

Abstract: We present a computationally simple and fully automatic method for creating virtual views of a scene by interpolating between sets of images obtained from different viewpoints. The technique is intended to give an immersive experience and a sense of viewing a real environment. The method uses correspondence points identified in the primary images. The matching of corresponding locations in the images has two stages. The first is based on 'corner points' and is used to extract the epipolar geometry that relates the images, then a second stage of edge matching recovers a more complete set of correspondences. These edge correspondences are then interpolated and used in an efficient morphing algorithm that operates one scan- line at a time to perform image synthesis.

40 Pages

Back to Index