Bayesian spherical wavelet shrinkage: applications to shape analysis
Xavier Le Faucheur, Brani Vidaković, Allen Tannenbaum · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2007
Multiscale analysis has become indispensable in image processing and computer vision. Our work is motivated by the need to efficiently represent 3D shapes that exhibit a spherical topology. This note presents a wavelet based model for shape denoising and data compression. The 3D shape signal is first encoded using biorthogonal spherical wavelet functions defined on a 3D triangulated mesh. We propose a Bayesian shrinkage model for this type of second generation wavelets in order to eliminate wavelet coefficients that likely correspond to noise. This way, we are able to reduce dimension without losing significant information by estimating a noiseless version of our shape.