Existing datasets of infrared and visible images only contain few extreme scenes, we construct a dataset of images with haze based on the M3FD dataset. We pick 450 aligned image pairs from M3FD dataset and synthesize hazy visible images using the ASM. Due to the unique imaging principle of infrared images, rarely affected by haze, there is no need to do additional process for infrared images. Finally, a dataset named MHS has been released, which contains 450 pairs of images in hazy conditions.

Multi-modality image fusion aims to combine diverse images into a single image that captures important targets, intricate texture details, and enables advanced visual tasks. Existing fusion methods often overlook the complementarity of difffferent modalities by treating source images uniformly and extracting similar features. This study introduces a distributed optimization model that leverages a collection of images and their signifificant features stored in distributed storage. To solve this model, we employ the distributed Alternating Direction Method of Multipliers (ADMM) algorithm.