DMCA
Shape Representation and (1992)
BibTeX
@MISC{Using92shaperepresentation,
author = {Image Segmentation Using and H Delingette and M Hebert and K Lkeuchi},
title = {Shape Representation and},
year = {1992}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
this paper we propose an approach that attempts to solve this conflict by using the duality in a way similar to the approach. Several psychophysical experiments have proved that the hu- man eye is able to capture the main shape of an object by seeing only a few characteristic elements or features. These features can be either geometric (distance discontinuities, surface orientation discontinuities, cor- ners, minimum of curvature, etc.), or higher level such as reflectance properties. While these features capture most of the shape information, it is difficult without a priori knowledge to build a full reconstruction of the object. Several solutions have been proposed. Besl and Jain' built curvature-based object representations by clas- sifying surfaces according to the sign of their principal curvatures. Pentland and presented a physically-based algorithm to recover a model in a unique manner from a set of features and vibration modes. In this approach, the key is to find the correct features, thus restraining the algorithm to either smooth or structured shapes. Another approach is to use both sets, according to geometric properties (sym- metry, connexity, etc.), and in a second stage models are fit to the segmented parts. This idea of hierarchical representation was initiated by Marr and and pushed further by the seminal ACRONYM vision system by using generalized cylinders. Pentland's `representation by parts' using deformed superquadrics'. proved to have some successful results, but encounters some limitations. While the feature grouping requires some accurate feature extrac- tion and hi h level reasoning, the fitting of to the range data has some unstable behaviour, due to its non-linear nature, and is suitable for only smooth and simple shapes. Those techn...
Keyphrases
shape representation range data full reconstruction correct feature physically-based algorithm seminal acronym vision system level reasoning generalized cylinder characteristic element non-linear nature simple shape deformed superquadrics several psychophysical experiment surface orientation discontinuity cor ners successful result shape information feature grouping vibration mode distance discontinuity reflectance property geometric property principal curvature unique manner several solution hu man eye hierarchical representation curvature-based object representation priori knowledge second stage model unstable behaviour main shape structured shape sym metry accurate feature extrac tion
