1: \begin{abstract}

2: In plug-and-play (PnP) regularization, the knowledge of the forward model is combined with a powerful denoiser to obtain state-of-the-art image reconstructions. 

3: This is typically done by taking a proximal algorithm such as FISTA or ADMM, and formally replacing the proximal map associated with a regularizer by nonlocal means, BM3D or a CNN denoiser. 

4: Each iterate of the resulting PnP algorithm involves some kind of inversion of the forward model followed by denoiser-induced regularization. 

5: A natural question in this regard is that of \textit{optimality}, namely, do the PnP iterations minimize some $f+g$, where $f$ is a loss function associated with the forward model and $g$ is a regularizer? 

6: This has a straightforward solution if the denoiser can be expressed as a proximal map, as was shown to be the case for a class of linear symmetric denoisers. 

7: However, this result excludes kernel denoisers such as nonlocal means that are inherently non-symmetric. 

8: In this paper, we prove that a broader class of linear denoisers (including symmetric denoisers and kernel denoisers) can be expressed as a proximal map of some convex regularizer $g$. 

9: An algorithmic implication of this result  for non-symmetric denoisers is that it necessitates appropriate modifications in the PnP updates to ensure convergence to a minimum of $f+g$. 

10: Apart from the convergence guarantee, the modified PnP algorithms are shown to produce good restorations.

11: \end{abstract}

12: