A multi-weight fusion framework for infrared and visible image fusion

Yiqiao Zhou, Kangjian He, Dan Xu, Hongzhen Shi, Hao Zhang

Abstract: Infrared and visible image fusion (IVF) aims to generate a fused image with important thermal target and texture information from infrared and visible images. However, the existing advanced fusion methods have the problem of insufficient extraction of visible image details, and the fused image is not natural and does not conform to human visual perception. To solve this problem, we propose an effective infrared and visible image fusion framework inspired by the idea of multi-exposure fusion. First, we design an adaptive visible light exposure adjustment module to enhance the low-brightness pixel area information in the visible image to obtain an adaptive exposure image. Secondly, three feature weight maps of the input infrared, visible light and adaptive exposure images are extracted through the multi-weight feature extraction module: DSIFT map, saliency map and saturation map, and then the feature weight maps are optimized through the Mutually Guided Image Filtering (MuGIF). Then, we use the Gaussian and Laplacian pyramids to decompose and reconstruct the feature weight map and input image to obtain the pre-fused image. Finally, to further enhance the contrast of the pre-fused image, we use a Fast Guided Filter to enhance the pre-fused image to obtain the final fusion result. Qualitative and quantitative experiments show that the proposed method exhibits better fusion performance on public datasets compared with 11 state-of-the-art methods. In addition, this method can retain more visible image details, and the fusion result is more natural. Our code is publicly available at https://github.com/VCMHE/MWF_VIF.