Frequency penalization directly targets the high frequency noise these methods suffer from. It may explicitly penalize variance between neighboring pixels (total variation) , or implicitly penalize high-frequency noise by blurring the image each optimization step . If we think about blurring in Fourier space, it is equivalent to adding a scaled L2 penalty to the objective, penalizing each Fourier-component based on its frequency. Unfortunately, these approaches also discourage legitimate high-frequency features like edges along with noise. This can be slightly improved by using a bilateral ﬁlter, which preserves edges, instead of blurring .