A CONVERGENT IMAGE FUSION ALGORITHM USING SCENE-ADAPTED GAUSSIAN-MIXTURE-BASED DENOISING

Abstract

We propose a new approach to image fusion, inspired by the recent plug-and-play (PnP) framework. In PnP, a denoiser is treated as a black-box and plugged into an iterative algorithm, taking the place of the proximity operator of some convex regularizer, which is formally equivalent to a denoising operation. This approach offers flexibility and excellent performance, but convergence may be hard to analyze, as most state-of-the-art denoisers lack an explicit underlying objective function. Here, we propose using a scene-adapted denoiser (i.e., targeted to the specific scene being imaged) plugged into the iterations of the alternating direction method of multipliers (ADMM). This approach, which is a natural choice for image fusion problems, not only yields state-of-the-art results, but it also allows proving convergence of the resulting algorithm. The proposed method is tested on two different problems: hyperspectral fusion/sharpening and fusion of blurred-noisy image pairs.

Existing System

Alternating direction method of multipliers (ADMM), proximity operator

Proposed System

We propose plugging a GMM-based denoiser into an ADMM algorithm; this GMM-based denoiser is a modified version of a patch-based MMSE (minimum mean squared error) denoiser; the modification consists in fixing the posterior weights of the MMSE denoiser. We prove that the resulting ADMM converges to a global minimum of a cost function that we explicitly identify. We apply the proposed framework to the two fusion problems mentioned above, showing that it yields results competitive with the state-of-the-art.

Architecture

BLOCK DIAGRAM