Feature based hierarchical shape matching

A technique for matching 2D regions in digital images using shape signatures is described. Successive derived signatures at varying resolutions are produced first, in which major shape feature information for the region being represented is preserved. Correspondences between the shape features for two different regions are established using a pair of derived signatures of similar resolution. A featural error sum is then computed to provide a measure of dissimilarity between the regions. Coarse resolution versions of candidate regions are compared first to allow unlikely cases to be rejected rapidly and then refined versions of the remaining cases are considered. Some experimental results for leaf outline data are presented and compared with those obtained when only the original signatures are considered