Recovering the 3D shape of an object from its 2D image contour is an important problem in computer vision. This paper begins with a survey and critique of some of the previous shape from contour methods that have been proposed. The author then develops two object-based heuristics and uses them to recover shape constraints from contour. The basic assumption guiding these object-based heuristics is that objects tend to be highly structured and can be parametrized with respect to an object-centered, orthogonal coordinate system. The structured nature of objects is the motivation for the non-accidental alignment criterion: parallel lines within the object's bounding contour are significant and are related to the 3D object-centered coordinate system. The regularity and symmetry inherent in many man-made objects, coupled with a human preference for an orthogonal perceptual space, is the motivation for the orthogonal basis constraint: an oblique set of coordinate axes in the image is presumed to be the projection of an orthogonal set of coordinate axes in the scene. These object-based heuristic methods are demonstrated on both real and synthetic image contours.

Date Published: 9 April 1993
PDF: 21 pages
Proc. SPIE 1832, Vision Geometry, (9 April 1993); doi: 10.1117/12.142170

Published in SPIE Proceedings Vol. 1832:
Vision Geometry
Robert A. Melter; Angela Y. Wu, Editor(s)