Boundary descriptors may be used effectively in various image based recognition problems. However, when images are acquired in uncontrolled conditions, changes in pose and illumination complicate the alignment process that must occur prior to comparison. The alignment is further complicated when the objects of interest are characterized by open contours and when distinct feature points are absent. We present an application that employs an iterative approach to the alignment of open contours and suggest a method for selecting corresponding points along a pair of contours for the purpose of determining similarity. We demonstrate the success of this approach in recognizing individual bottlenose dolphins from the profiles of their dorsal fins.

Algorithms for computing the aspect graph representation are generalized to include a larger, more realistic domain of objects known as articulated assemblies those objects composed of rigid parts with articulated connections allowed between parts. The generalization suggests two slightly different representations: one that directly summarizes the possible general views of the object and another (hierarchical) form summarizing the possible general configurations and their respective views. Algorithms are outlined for computing both representations. The generalized aspect graphs of assemblies formed using translational connections are examined.>

Consideration is given to a generalization of the aspect graph concept which is applicable to objects which may have articulated connection between their (rigid) parts. This generalization is a major step forward in terms of applying the aspect graph concept to a larger and more realistic domain of objects. An algorithm for computing the generalized aspect graph is described, and an example is analyzed in detail.>

An algorithm is presented for constructing the perspective projection aspect graph of polyhedra. The first phase determines the set of surfaces involved in the parcellation of viewing space around the object. Three types of surfaces are involved: object planes, in which the faces of the object lie, auxiliary planes, defined by the visual interaction of edge-vertex pairs, and auxiliary quadric surfaces, defined by the visual interaction of edge triplets. Auxiliary planes and surfaces in a sense enumerate the potential self-occlusions of the object. A list is constructed for each type of surface. Next the algorithm constructs the geometric incidence lattice representing the arrangement of object planes, and distinguishes each 3-face (node) as representing either object or viewing space. The auxiliary planes and surfaces are then added to the lattice. Each 3-face (node) in the lattice has visibility attributes which identify the potentially visible object features from viewpoints within it. These attributes are updated throughout the process of building the lattice. Finally, the lattice is traversed to make a final update of the visibility attributes and merge neighboring 3-faces which have the same visible features. The resulting structure represents the perspective projection aspect graph.