A review of infrared and visible light image fusion algorithms based on deep learning

In the digital era, the preservation and dissemination of Intangible Cultural Heritage (ICH) increasingly rely on advanced artificial intelligence technologies. Infrared and Visible Image Fusion (IVIF), a pivotal technique in image processing, has seen broad application across computer vision, remote sensing, medical imaging, and defense. Recently, it has shown promising potential in the digital preservation and restoration of traditional art forms. This paper presents a comprehensive review of the theoretical foundations and research progress of IVIF, with a focus on mainstream fusion methods. It critically compares traditional techniques with deep learning-based approaches in terms of image enhancement, detail preservation, and multi-source information alignment. Special attention is given to the characteristics and application scenarios of three representative architectures: autoencoders, generative adversarial networks (GANs), and end-to-end neural networks. Task-oriented fusion strategies are also discussed. Extensive experiments on multiple public datasets, evaluated using diverse metrics, reveal the strengths and limitations of different methods. The paper further summarizes existing datasets and evaluation criteria, identifies key challenges in real-world IVIF applications for ICH, and outlines future research directions. This work aims to advance the integration of AIdriven image fusion technologies into the digital safeguarding of ICH, providing both theoretical insights and technical guidance for interdisciplinary innovation.