1: \begin{abstract}

2: 	\noindent\small

3: 	We introduce a new non-smooth variational model for the restoration of

4: 	manifold-valued data which includes second order differences in the

5: 	regularization term. 

6: 	While such models were successfully applied for real-valued

7: 	images, we introduce the second order difference and the corresponding variational models for manifold data, which up to now only existed for cyclic data.

8: 	The approach requires a combination of techniques from numerical analysis, 

9: 	convex optimization and differential geometry.

10: 	First, we establish a suitable definition of absolute second order

11: 	differences for signals and images with values in a manifold.

12: 	Employing this definition, we introduce a variational denoising model

13: 	based on first and second order differences in the manifold setup.

14: 	In order to minimize the corresponding functional,

15: 	we develop an algorithm using an inexact cyclic proximal point algorithm.

16: 	We propose an efficient strategy for the computation of the corresponding

17: 	proximal mappings 

18: 	in symmetric spaces

19: 	utilizing the machinery of Jacobi fields.

20: 	For the $n$-sphere

21: 	and the manifold of symmetric positive definite matrices,

22: 	we demonstrate the performance of our algorithm in practice.

23: 	We prove the convergence of the proposed exact and inexact variant of the

24: 	cyclic proximal point algorithm

25: 	in Hadamard spaces. These results which are of interest on its own include,

26: 	e.g., the manifold of symmetric positive definite matrices.

27: \end{abstract}

28: