Exemplar-Based EM-like image denoising via manifold reconstruction

Discovering local geometry of low-dimensional manifold embedded into a high-dimensional space has been widely studied in the literature of machine learning. Counter-intuitively, we will show for the class of signal-independent additive noise, noisy observation does not destroy the manifold structure thanks to the bless of dimensionality. Based on this observation, we propose to reconstruct image manifold for a collection of exemplars by iterative Wiener filtering and neighborhood search. The byproduct of such manifold reconstruction from noisy data is an exemplarBased EM-like (EBEM) denoising algorithm with minimal number of control parameters. Despite its conceptual simplicity, EBEM can achieve comparable performance to other leading algorithms in the literature. When combined with bootstrap resampling strategy, EBEM could significantly outperform the current state-of-the-art BM3D denoising for images satisfying certain global symmetry property.

• Type: Journal
• Year: 2010
• Where: ICIP
• Authors: Xin Li
• Keywords: Bootstrap Resampling Strategy, ICIP 2010, Image Processing, Manifold, Signal-independent Additive Noise