Abstract: In this work, we present effective methods for detecting and
representing invariant features of three dimensional (3-D) objects from single
images. Our methods are independent of the viewpoint position, and can be used
in the recognition process of 3-D objects. The invariant features are; shapes
of the object surfaces, organization of the surfaces within the object, and the
new feature which is shapes of privileged surfaces (PS) of the object. Also we
explain the way these features are represented and matched against stored model
of the object. Our work is built around the knowledge based approach and depends on
the fact that while it is true that the appearance of a 3-D object may change
completely as it is viewed from different viewpoints, it is also true that many
aspects of the object projection remain invariant over large ranges of
viewpoints. This approach is different from the other approach that derives or
calculates depth information and orientations of surfaces using more than one
image of the object, or the approach that statistically detects and matchs the
object features. The main contribution of our work is the introduction of privileged
surface of 3-D object which plays an important role in the recognition process.
Also we introduce heuristic methods for detecting shapes of surfaces, object
organization, object & model representations and matching. The idea of
matching privileged surfaces reduces the ambiguity that may arise due to
limited visibility of the object surfaces (existence of hidden surfaces) when
matching features of both the image and the model. Also matching privileged surfaces reduces the need of detecting and
matching orientations of surfaces. This approach has been applied to 3-D
objects which are bounded by planar surfaces. However, the work can be extended
in a straightforward manner to support 3-D objects that are bounded by curved
surfaces.
