A review of thermal infrared-to-visible image translation algorithms

Abstract Thermal infrared-to-visible image translation (I2V) captures objects’ emitted radiation. It aims to map the thermal infrared radiation information into visible color images that conform to human visual cognition. This enables visual systems to acquire richer semantic and structural information in low-illumination environments. I2V holds significant application value in fields such as nighttime surveillance, intelligent driving, medical imaging, and military perception. It can notably improve target recognition accuracy and visual readability, and promote the unified processing of cross-modal visual systems. This paper systematically reviews the research progress of thermal infrared-to-visible image translation. It introduces traditional methods including pseudo-color encoding, reference-image-based translation, and multispectral fusion, as well as deep learning methods based on convolutional neural networks, generative adversarial networks, and transformer architectures. Combined with existing experimental results, the paper summarizes the development trends of different methods in terms of translation quality. It also discusses potential future breakthrough directions, such as cross-domain semantic consistency modeling, model generalization and robustness improvement, and optimization for lightweight and real-time applications.