MULTI-RESOLUTION BASED SPATIALLY ADAPTIVE MULTI-ORDER TOTAL VARIATION FOR IMAGE RESTORATION

We propose a novel total variation based image restoration method that combines first and second order total variation functionals in a structurally adaptive manner. Specifically, as opposed to recent practical combined order methods, which choose the relative weight between the functionals of different derivative orders as a global scalar parameter, we construct a functional that uses a spatially varying relative weight between functionals of different orders so as to adapt to the local image structure. Further, as opposed to known combined order methods which leave the relative weight as a user parameter, the proposed method automatically computes the relative weight using a novel multi-resolution based formulation. The proposed spatially adaptive combined order TV (SAMTV) performs better than the state of art total generalized variation (TGV) method in experiments.