Spatially High Resolution Visible and Near-Infrared Separation using Conditional Generative Adversarial Network and Color Brightness Transfer Method

Since near-infrared (NIR) image information is useful in improving visible range (VIS) image, acquisition of both images in a more simple and economic way has drawn much research interest. Deep-learning based approach is found to be effective in the separation from a mixed NIR and VIS image captured by a conventional camera, however, it has a problem of high computational complexity, especially for an image of high spatial resolution. In this paper, we propose a method for separating high-resolution VIS and NIR images using a deep-learning based on a conditional generative adversarial network. Experimental results show that the proposed method can reduce the computational complexity by 97 times as compared with the previous work without loss in image quality.