Shape Matching based on Non-rigid spectral transformation

This paper provides an efficient method for shape matching by using non-rigid spectral transformation. Our method builds a joint weighted graph with geometric features of volumetric distance, geodesic distance and normal variation. And we find the meaningful correspondence based on TPS (thin-plate spline) approach in the spectral embedded space which can be used for classifying shapes into different categories. Finally the experimental results demonstrate that our method can accurately and efficiently classify a wide variety of shapes, especially which may have different poses and significant topology changes.