Two Methods for Comparing Different Views of the Same Object

The viewing hemisphere of a 3-dimensional object can be partitioned into areas of similar views, termed view bubbles. We compare two procedures of generating view bubbles: tracking of object features, i.e., Gabor wavelet responses, by utilizing the continuity of successive views and matching of features in different views, which are assumed to be independent. Both procedures proved to be appropriate to detect canonical views. We found no difference concerning the size of the view bubbles, but tracking provides more precise correspondences than matching. Tracking is more appropriate for recognizing changes of features, whereas matching is more suitable if features of the same appearance are to be found. 1 Subject of Investigation For pose-invariant object recognition and pose estimation of objects it is necessary to utilize an appropriate object representation. A naive representation might consist of densly spaced views of an object’s viewing sphere. Our aim is to reduce such a “full” representation to only some representative views and the relations between them. Such a sparse representation belongs to the aspect graph approaches proposed by, e.g., [1], [5]. To choose representative views (aspects) for a final representation our plan is first to determine for each view of a “full” representation a surrounding area of similar views, termed view bubble (see figure 1). Later the aspects for the final representation of the object can be derived from the overlaps of the view bubbles. In this paper we describe the generation of the view bubbles. We restrict our investigation to the upper hemisphere of an object’s viewing sphere. We compare two procedures of determining the similarity of two views: matching a representing graph of one view to another view, and tracking object features, i.e., Gabor wavelet responses, from one view to another view. During the matching procedure each view is treated independently, whereas the tracking procedure utilizes the continuity of neighbouring views. Our investigations were guided by the question which procedure matching or tracking is more appropriate to find for each view of the hemisphere view bubbles of maximal size containing views of maximal similarity? ∗Institut für Neuroinformatik, Systembiophysik, Ruhr-Universität Bochum, Universitätsstr. 150, D-44780 Bochum, Germany. E-mail: [email protected] †Department of Image Processing, Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic, Pod vodárenskou vĕz̆ı́ 4, 182 08 Praha 8, Czech Republic. E-mail: [email protected] British Machine Vision Conference 2 view (79, 14) view (6, 6)