MERFusion: A multiscale edge-preserving filter combined with Retinex enhancement for infrared and visible image fusion

The limitations of a single sensor stem from its equipment's optical capabilities, which prevent it from capturing diverse data on targets across multiple dimensions, thus enabling sophisticated vision tasks through image fusion. Currently, mainstream fusion methods suffer from significant loss of intricate details and the introduction of artifacts during low-light image fusion, resulting in unsatisfactory visual effects. To address these issues, this paper proposes a multiscale edge-preserving filter combined with Retinex enhancement for infrared and visible image fusion. The method incorporates low-light enhancement technology into the image fusion process, effectively tackling the challenges associated with low-light image fusion. Initially, we propose a weighted Retinex model for visible image enhancement, which is designed to efficiently incorporate details, texture, and brightness information in the darker regions of the source image. Furthermore, the proposed filter, benefiting from multiscale segmentation and edge preservation, decomposes the image into three distinct layers: base, infrared, and visible. Additionally, the designed gradient saliency fusion rule is adept at preserving the salient characteristics of infrared targets. Finally, by refining and integrating the detail layer with the base layer, we achieve the final fused image. Experimental findings indicate the superiority of this paper's method over current state-of-the-art methods.