A Perceptual Grouping Hierarchy for Appearance-Based 3D Object Recognition

Andrea Selinger and Randal C. Nelson
Department of Computer Science
University of Rochester
Rochester, NY 14627

Abstract: In this presentation we consider the problem of 3D object recognition, and the role that perceptual grouping processes must play. In particular, we argue that a single level of perceptual grouping is inadequate, and that reliance on a single level of grouping is responsible for the specific weaknesses of several well-known recognition techniques. Instead, we argue that recognition must utilize a hierarchy of perceptual grouping processes, and describe an appearance-based system that uses four distinct levels of perceptual grouping to represent 3-D objects in a form that not only allows recognition, but reasoning about 3D manipulation of a sort that has been supported in the past only by 3D geometric models.

The use of a hierarchy of grouping processes is not new, of course. What distinguishes our system is the grouping processes we use at the higher levels, which give certain exceptional capabilities. Our first two levels use good continuation and image proximity, in common with many approaches. Our third level of grouping however, is closest to what the Gestalt psychologists termed ``familiarity'', in that it uses experience memory to drive the grouping process. This category has not been much explored in the compuational perceptual grouping literature, but it turns out to be the essence of our recognition process, and to bear a certain similarity to the Cubist artistic renditions developed by Picasso and Brache at the beginning of the 20th century.

Our fourth level of grouping uses proximity, not in image space, but rather in viewing/manipulation space (which amounts to out-of-plane rotation space for rigid objects). This induces a topology on view-based object appearances, that when coupled with the lower-level representations can permit complex 3D manipulations to be planned, specified, and carried out on a purely visual basis, without the need for 3D geometric models. Again, this is a means of grouping that has received little attention in the grouping literature, but which produces representations that are practical to obtain, and of considerable potential value.