A popular approach for finding the correspondence between two nonrigid shapes is to embed their two-dimensional surfaces into some common Euclidean space, defining the comparison task as a problem of rigid matching in that space. We propose to extend this line of thought and introduce a novel spectral embedding, which exploits gradient fields for point to point matching. With this new embedding, a fully automatic system for finding the correspondence between shapes is introduced. The method is demonstrated to accurately recover the natural maps between nearly isometric surfaces and shown to achieve state-of-the-art results on known shape matching benchmarks.
All Science Journal Classification (ASJC) codes
- !!Signal Processing
- !!Computer Vision and Pattern Recognition