A metamorphosis-based shape recognition method

We present a framework for matching and recognition of planar shapes based on a method from computer graphics based animation, called "shape metamorphosis". In our approach, the "degree of morphing" between two shapes is employed as a dissimilarity measure. A physics-based energy minimization approach is used for optimally computing the "degree of morphing". This measure is shown to have metric properties and invariance to translation, rotation, scaling and mirror symmetry. Experimentations in the recognition of planar shapes, hand-drawn figures, and online cursive words indicate the robustness of the recognition paradigm.