GAN-based Thermal Infrared Image Colorization for Enhancing Object Identification

Colorizing a thermal infrared image is considered as a challenging task in computer vision and particularly in object recognition. The solution for thermal infrared colorization is quite problematic that not only transforming a grayscale to RGB image where the chrominance is predicted from the given luminance, but also the luminance should be estimated as well. In this paper, we propose a GAN-based method to encourage the similarity and well preserve details of images. Also, we deploy a multi-term objective function including content, perceptual, total variation, relativistic GAN losses and DSSIM to enhance the generated RGB images. Quantitative and qualitative experiments on a large available dataset show that our method obtains impressive results and outperforms compared to others existing colorized methods.