Several imaging systems provide large amount of images with complex degradation models and low signal to noise ratio. Specific adapted restoration methods should be developed. With the computing power currently available, new paradigms emerge, such as the “non-local” one (Buades et al., 2005), with very good performance (see, e.g., Lebrun et al. 2012; Milanfar, 2013). However, extensions of this methods to specific image modality might not be trivial. We focus on extensions of such methods for the restoration of non-conventional image modalities (low-light imagery, coherent imagery, tomography, …): presence of complex degradation models (blur, missing data, non-Gaussian, non-stationary and correlated noise) and requirement of fast computation large volume of data.