Reversible Efficient Diffusion for Image Fusion

TL;DR

The paper introduces the Reversible Efficient Diffusion (RED) model, a novel supervised diffusion framework that enhances multi-modal image fusion by preserving details and improving efficiency, overcoming traditional diffusion limitations.

Contribution

The paper proposes RED, a new supervised diffusion approach that maintains high-quality image fusion without distribution estimation, improving detail preservation and computational efficiency.

Findings

RED achieves superior detail preservation in fused images.

The method reduces noise accumulation compared to traditional diffusion models.

Enhanced efficiency in training and inference processes.

Abstract

Multi-modal image fusion aims to consolidate complementary information from diverse source images into a unified representation. The fused image is expected to preserve fine details and maintain high visual fidelity. While diffusion models have demonstrated impressive generative capabilities in image generation, they often suffer from detail loss when applied to image fusion tasks. This issue arises from the accumulation of noise errors inherent in the Markov process, leading to inconsistency and degradation in the fused results. However, incorporating explicit supervision into end-to-end training of diffusion-based image fusion introduces challenges related to computational efficiency. To address these limitations, we propose the Reversible Efficient Diffusion (RED) model - an explicitly supervised training framework that inherits the powerful generative capability of diffusion models while avoiding the distribution estimation.