Discrete Optimization for Shape Matching

We propose a novel discrete solver for optimizing functional map-based energies, including descriptor preservation and promoting structural properties such as area-preservation, bijectivity Laplacian commutativity among others. Unlike the commonly-used continuous optimization methods, our approach enforces map to be associated with pointwise correspondence hard constraint, which provides stronger link between optimized of point-to-point maps. Under this obtains maps lower energy values compared standard strategies. Perhaps more importantly, recovered from preserve are thus higher overall quality. demonstrate advantages on range energies shape categories, existing techniques within framework. Finally, in hand, we introduce Effective Functional Map Refinement (EFMR) method achieves state-of-the-art accuracy SHREC'19 benchmark.