We propose a new family of algorithms for denoising data assumed to lie on a low-dimensional manifold. The algorithms are based on the blurring mean-shift update, which moves each data point toward
