(The Eurographics Association, 2018) Sarkar, Kripasindhu; Bernard, Florian; Varanasi, Kiran; Theobalt, Christian; Stricker, Didier; Ju, Tao and Vaxman, Amir
We formulate the problem of point-cloud denoising in terms of a dictionary learning framework over square surface patches. Assuming that many of the local patches (in the unknown noise-free point-cloud) contain redundancies due to surface smoothness and repetition, we estimate a low-dimensional affine subspace that (approximately) explains the extracted noisy patches. This is achieved via a structured low-rank matrix factorization that imposes smoothness on the patch dictionary and sparsity on the coefficients. We show experimentally that our method outperforms existing denoising approaches in various noise scenarios.