36.4: A Comprehensive Quality Assessment of Video Super‐Resolution for Ultra High‐Resolution Application

As the performance of video super‐resolution (SR) algorithms continues to improve, traditional video quality assessment such as peak signal‐to‐noise ratio (PSNR) and structural similarity (SSIM), which are widely used for natural images, are gradually failing to meet the evaluation requirements as they do not effectively take all the characteristics of super‐resolved images into account. In addition, the newly emerged SR algorithms based on generative adversarial networks (GAN) bring a whole new challenge to quality assessment while improving SR performance. Therefore, this paper will introduce the shortcomings of existing video quality assessment methods and propose a comprehensive SR video quality assessment method including visual saliency, structural fidelity, texture information and temporal metrics.