View-independent shape representation and matching

A planar shape representation method is described which uses a binary tree to represent a shape. The nodes of the tree are triangles obtained from the shape through iteratively merging the neighboring boundary segments of the shape. Every two neighboring boundary segments define a triangle, and the selection of the neighboring boundary segments to merge is based on the relative sizes of such triangles along the shape boundary. Since these relative sizes are view-independent, the partitioning of a shape is also view-independent. This leads to the result that the tree representation is view-independent. In shape matching, the projection transformation is estimated to exist between the two shapes from the top-level nodes of the two tree representations. One tree (shape) is adjusted according to the found estimations to remove shape difference caused by projection. The two trees are compared node by node to find the shape dissimilarity between the two shapes. Since one shape is adjusted according to the estimated projection difference, the shape matching method is view-independent.<>