Title :
Ordinal and metric structure of smooth surfaces from parallax
Author :
Baratoff, Gregory
Author_Institution :
Comput. Vision Lab., Maryland Univ., College Park, MD, USA
Abstract :
The work presented in this paper is a first attempt at evaluating an ordinal representation for the local shape of smooth surface patches. To this end we investigate the reliability of estimating the tilt and slant components of the local surface orientation from binocular parallax. This is of interest because parallax is a commonly used cue, but has not been evaluated as to its susceptibility to image measurement errors. Our theoretical sensitivity analysis shows that the tilt is less sensitive than the slant to errors in the estimation of the viewing geometry, but that image measurement errors affect both slant and tilt seriously. In view of these results, it can not be claimed that the tilt is more reliably estimated from parallax than the slant. Despite the discovered sensitivity to image measurement errors, the tilt remains a very useful piece of information, in that it still specifies partial ordinal relations even when substantially degraded. We present a simple and efficient method to keep track of such partial information during all ongoing computation
Keywords :
computational geometry; error analysis; feature extraction; image reconstruction; image representation; sensitivity analysis; binocular parallax; image measurement errors; local shape; metric structure; ordinal representation; ordinal structure; partial information; sensitivity analysis; slant component; smooth surface patches; tilt component; viewing geometry; Automation; Computer vision; Contracts; Degradation; Educational institutions; Geometry; Image analysis; Laboratories; Shape; Surface topography;
Conference_Titel :
Pattern Recognition, 1996., Proceedings of the 13th International Conference on
Conference_Location :
Vienna
Print_ISBN :
0-8186-7282-X
DOI :
10.1109/ICPR.1996.546033