Improvement of infrared and visible images' fusion algorithms in night-vision context

To address the issues of insufficient contrast and weak detail representation in the fusion of infrared and visible light images under low-light nighttime conditions, this study improves the Gaussian filter-based context enhancement (GFCE) fusion algorithm. First, an adaptive enhancement of the infrared image is performed using a guided filter. Then, a mixed multi-scale decomposition method based on the guided filter is employed to sharpen and denoise the detail layers obtained from the decomposition, thereby enhancing the image's clarity. Additionally, this paper introduces a perceptual-based method for selecting regularization parameters, which determines the relative amount of infrared spectral features to be injected into the visible light image by comparing the perceptual saliency of information in both infrared and visible light images. Experimental results show that the optimized IGFCE algorithm significantly outperforms the GFCE algorithm across key evaluation metrics, including average gradient (AG), edge intensity (EI), and spatial frequency (SF), enhancing the visual performance of nighttime scenes. The AG improves 22.9% and 16.7% in the two datasets (LLVIP and TNO); the EI improves 24.0% and 15.2% in the two datasets; and the SF improves 27.7% and 17.7% in the two datasets, respectively.