Infrared and Visible Image Fusion Method based on Residual Network

Yansheng Gao, Zhiyuan Cheng, Hua Su, Zhou Ji, Jiahao Hu, Ziqian Peng

Abstract: In the same scene, infrared images and visible light images contain significantly different information, and image fusion techniques can combine them into a single image to enhance image quality. Over the past few years, the field of image fusion has experienced a significant transformation with the advent of deep learning techniques. However, existing fusion strategies for combining infrared and visible light images based on deep learning suffer from the lack of well-designed loss functions, leading to suboptimal fusion results. Addressing the aforementioned issue, a fusion method for infrared and visible light images based on residual network is presented in this study. This method leverages the deep characteristics of residual networks to extract features and combine shallow and deep features to quantify the information content within the original images, guiding the training of the fusion network to achieve the final fused image. The experimental outcomes on TNO and RoadScene datasets validate the efficacy of the proposed method in attaining fusion results of superior quality.