“…Hence, in order to obtain a high quality recovery from the ill-posed linear inverse problem (1.1), a proper regularization on the images to be recovered is needed. Successful regularization based methods include the Rudin-Osher-Fatemi model [54] and its nonlocal variants [38,63], the inf-convolution model [17], the total generalized variation (TGV) model [7,8], the combined first and second order total variation model [6,47,52], and the applied harmonic analysis approach such as curvelets [14], Gabor frames [22,40,44,48], shearlets [46], complex tight framelets [41], wavelet frames [4,9,10,13,19,24,30,32,35,36,58,64], etc. The common concept of these methods is to find sparse approximation of images using a properly designed linear transformation together with a sparsity promoting regularization term (such as the widely used 1 norm).…”