Benchmark comparing the wavelet shrinkage algorithm to the dictionary learning denoising algorithm when dealing with various noise levels, using the dictionary from Fig. 5. Each experiment is repeated 100 times and the results are averaged. We use the maximum value for the patch-overlaping parameter. The sparse coding uses OMP and is set to reach an error margin where σ is the noise standard deviation and C is a gain factor set to 1.15. The wavelet algorithm uses five scales of undecimated bi-orthogonal wavelets, with three bands per scale. The red and blue lines correspond to wavelet and learned dictionary denoising. The horizontal axis is the peak S/N between the noised and the source images, and the horizontal axe is the peak S/N between the denoised and the source images.