1: \begin{abstract}

2: In this paper we formulate and study a variational problem in the space of functions of bounded Hessian $BH(\Omega)$. This forms a higher order extension of the well known ROF functional (total variation minimisation) which is widely used for image reconstruction. Our functional involves convex functions of the total variation and the total variation of the first derivatives. We prove existence and uniqueness of minimisers via the method of relaxation. We use the split Bregman method in order to solve numerically the corresponding discretised problem and we prove some convergence results. We apply our model to image denoising (Gaussian and impulse noise) and deblurring as well as image inpainting. The numerical results show that our model avoids the creation of undesirable artifacts and blocky-like structures in the reconstructed images which is a disadvantage of the ROF model.

3: \keywords{Functions of bounded Hessian \and Split Bregman \and Denoising \and Deblurring \and Inpainting \and Staircasing}

4: \subclass{\\49J52 \and 65K10 \and 65N21 \and 90C25 \and 52A41 \and 68U10}

5: \end{abstract}

6: