1: \begin{abstract}

2: 	Total Variation (TV) and related extensions have been popular in image 

3: 	restoration due to their robust performance and wide applicability. While

4: 	the original formulation is still relevant after two decades of extensive 

5: 	research, its extensions that combine derivatives of first- and 

6: 	second-order are now being explored for better performance, with examples

7: 	being Combined Order TV (COTV) and Total Generalized Variation 

8: 	(TGV).  As an improvement over such multi-order convex formulations, 

9: 	we propose a novel non-convex regularization functional which 

10: 	adaptively combines Hessian-Schatten (HS) norm and first order TV (TV1) 

11: 	functionals with spatially varying weight. This  adaptive weight   

12: 	itself is controlled by another regularization term;  the total

13: 	cost becomes the sum of this adaptively weighted HS-TV1 term, 

14: 	the regularization term for the adaptive weight, and the data-fitting

15: 	term.  The reconstruction is obtained by jointly minimizing w.r.t. 

16: 	the required image and the adaptive weight.  We construct a block 

17: 	coordinate descent method for this minimization with proof of convergence,

18: 	which alternates between minimization w.r.t. the required image and 

19: 	the adaptive weights.  We derive exact computational formula for minimization

20: 	w.r.t.  the adaptive weight,  and construct an ADMM algorithm for minimization

21: 	w.r.t. to the required image.   We compare the proposed   

22: 	method using   image recovery   examples including   MRI   reconstruction and 

23: 	microscopy   deconvolution.

24: 	%   show that the proposed COROSA approach performs better than

25: 	%    related TV based methods.

26: \end{abstract}

27: