Plug-and-Play with POA based Maximum a Posteriori Denoisers for Image

Restoring damaged images has been an issue in the fields of image processing and basic computer vision for decades. Due to their improved performance, techniques based on discriminative convolutional neural networks (CNNs) have recently garnered a lot of interest. However, most of these frameworks only operate well on one type of picture restoration assignment, hence their performance on other types of image restoration is usually subpar. Research techniques leverage the fact that the Maximum a Posteriori (MAP) optimisation may be broken down into smaller sub-problems, such as a MAP denoising optimisation, to resolve this issue. We introduce the first full-stack method for MAP estimation in deep neural network image denoising. We prove that our approach will always result in a smaller MAP denoising goal, which can subsequently be employed in the general picture restoration optimisation technique known as the Puzzle Optimisation technique (POA). Our technique is theoretically analysed, and its quantitative efficacy is demonstrated through a number of tests. Our experiments validate the theoretical foundations of MAP and demonstrate that the suggested approach can reach 70x quicker presentation than the state-of-the-art.