“…For the latter, graph spectral representations [10], which are invariant to isometry (preserving geodesic distances), can capture large and complex deformations. Spectral methods [10,11,12,13,14] are popular for general graph partitioning and are applied in computer vision for mesh matching [15,16,17,18] and shape retrieval [19]. Pioneered in the late 80s, [12,13,14], spectral correspondence methods match points in shapes by comparing eigenvectors of a proximity matrix derived from the mesh structure.…”