Artifacts Reduction GAN For Enhancing Quality Of Compressed Panoramic Video

Panoramic video has the characteristics of high resolution, massive information and high sense of immersion, bringing unprecedented visual sensory enjoyment to us. However, considering the limited capacity of current cellular network, videos transmitted to users are often encoded, e.g. by High Efficiency Video Coding (HEVC), resulting in block artifact problems that significantly affects the visual quality. Thus, it is necessary to enhance the quality of compressed panoramic videos. Inspired by the convolutional networks (CNN) and generative adversarial networks (GAN), the paper proposes a deep GAN model -Artifacts Reduction GAN (ARGAN) which is able to enhance the quality of compressed panoramic videos. ARGAN has the ability of reducing artifacts caused by HEVC. Meanwhile, it can increase the visual realistic of the enhanced videos. We tested the performance of our model under PSNR, SSIM and Perception Index. Qualitative results are also provided to display the visual effects of ARGAN. Experimental results show that our method is superior to other quality enhancement methods in both qualitative and quantitative aspects.