XTM-GAN: Generative Adversarial Networks for Tone Mapping of Dual-Energy X-ray Security Images
Da Cai, Tong Jia, Hao Wang, Dongyue Chen · 2024
X-ray imaging technology is vital in national public security. However, compressing 16-bit high dynamic range X-ray images into 8-bit displayable images often leads to loss of crucial details, hampering effective security inspection. To address this, we propose XTM-GAN, a dual-energy X-ray tone mapping generative adversarial network, to enhance X-ray imaging quality. By leveraging the properties of dual-energy X-rays, we construct a dedicated dataset called DEXray. We design a dual-branch generation network and a discrimination network that combines global and local image information. Through end-to-end generative adversarial training using pairs of dual-energy HDR and LDR X-ray images, which can generate high-quality X-ray security images. We train and compare our algorithm using the DEXray dataset against other state-of-the-art tone mapping algorithms. The results demonstrate the superior performance of our approach. The proposed method has significant practical applications and holds research significance in the fields of security and transportation.