부산대학교 도서관

The image fusion of the optical and infrared images is widely studied in this year to provide better samples for target detection and classification. In this paper, we propose a generative framework for image fusion of the optical and infrared images. This framework is composed by a hierarchical generative adversarial network (GAN) including one generator and two discriminators. Different to the existing fusion method, we design a two-channel encoder for better fusion performance. Specifically, the intensity of the optical image is extracted and fed into an encoder, which is then integrated with the infrared image. In this way, the proposed framework is capable of learning the mutual knowledge between the optical and infrared images in terms of the intensity distribution. Extensive simulations on two datasets evaluate that the proposed method achieves significantly better performance than the existing approaches in terms of PSNR, RMSE and SSIM score, as well as the visualization results.