HRIF: haze removal image fusion for visible and infrared images

Abstract

Abstract Infrared and visible image fusion is an advanced processing procedure in the measurement of visual sensors. Existing methods usually assume that the input multimodal images are clear. However, complex scenes such as smoke and haze cause significant interference to traditional fusion methods. Therefore, this paper proposes a novel unified haze removal image fusion (HRIF) framework , which achieves joint optimization of haze removal and multimodal fusion through an end-to-end network architecture. By designing a dual-channel attention mechanism, the feature extraction and noise suppression capabilities of HRIF are effectively enhanced. The model can dynamically monitor smoke distribution and adaptively adjust the fusion process. Additionally, a large multimodal fusion dataset HazeScene for smoke or haze scenarios was constructed, and an average smoke concentration (ASC) evaluation metric was proposed. Experiment results demonstrate that HRIF significantly outperforms existing state-of-the-art image fusion methods in metrics such as PSNR, MSE, CC, SSIM, and ASC, especially excelling in high-concentration haze scenario. This research provides a new solution for multimodal image fusion and visual measurement in haze environments, and has broad application prospects in multiple fields such as autonomous driving and disaster monitoring. The code of the HRIF algorithm is available at https://github.com/windrunners/HRIF .