Plug-and-play (P&P) priors is a concept developed by Prof. Charles A. Bouman, his students, and collaborators. It is a distributed optimization framework that allows the use of modern denoising operators as prior models in regularized Bayesian inversion.

The P&P framework is based on the alternating direction method of multipliers (ADMM), and decouples the forward model and the prior terms in the maximum-a-posteriori cost function. This results in an algorithm that involves repeated application of two steps: an inversion step only dependent on the forward model, and a denoising step only dependent on the image prior model. The P&P takes ADMM one step further by replacing the prior model optimization by a denoising operator of choice.