The viewing hemisphere of a 3-dimensional object can be partitioned into areas of similar views, termed view bubbles. We compare two different procedures of generating view bubbles. On the one hand, local object features are tracked from one view to another by utilizing the continuity of successive views while the object rotates. On the other hand, the features are matched in different views which are assumed to be independent. We had a quantitative and a qualitative criterion of comparison (the size of the view bubbles and view similarities inside the view bubbles, respectively). To compare the sizes of view bubbles for both procedures we performed statistical analyses. For the qualitative comparison we assessed the correspondences provided by both procedures. The simulations were done on natural images of two objects. Both procedures, tracking as well as matching, proved to be appropriate to generate a distribution of view similarities on the viewing hemisphere. Canonical views arise. We could not find a difference between both procedures concerning the quantitative size criterion, but tracking outperforms matching concerning the qualitative condition. Tracking provides much more precise correspondences than matching. The continuous information seems to be neccessary to build the correspondences. Accordingly, tracking is the more appropriate method for recognizing the changes of features, whereas matching is more suitable if features of the same appearance are to be found. Our results are supported by related insights from psychophysical research.
[1]
Joachim M. Buhmann,et al.
Distortion Invariant Object Recognition in the Dynamic Link Architecture
,
1993,
IEEE Trans. Computers.
[2]
Allen M. Waxman,et al.
Adaptive 3-D Object Recognition from Multiple Views
,
1992,
IEEE Trans. Pattern Anal. Mach. Intell..
[3]
Christoph von der Malsburg,et al.
Tracking and learning graphs and pose on image sequences of faces
,
1996,
Proceedings of the Second International Conference on Automatic Face and Gesture Recognition.
[4]
M. F.,et al.
Bibliography
,
1985,
Experimental Gerontology.
[5]
Jochen Triesch,et al.
GripSee: A Robot for Visually-Guided Grasping
,
1998
.
[6]
Laurenz Wiskott,et al.
Labeled graphs and dynamic link matching for face recognition and scene analysis
,
1995
.
[7]
K. Hoffmann,et al.
Visual Inspection of Three-Dimensional Objects by Human Observers
,
1996,
Perception.
[8]
Hanspeter A. Mallot,et al.
Phase-based binocular vergence control and depth reconstruction using active vision
,
1994
.
[9]
M. Wertheimer.
Untersuchungen zur Lehre von der Gestalt. II
,
1923
.
[10]
P. Kellman.
Perception of three-dimensional form by human infants
,
1984,
Perception & psychophysics.
[11]
Jan C. Vorbrüggen.
Zwei Modelle zur datengetriebenen Segmentierung visueller Daten
,
1995
.