Abstract
With the prevalence of 4G/5G infrastructure and mobile devices, mobile video streaming has become an ubiquitous element of daily life. Nevertheless, the online delivery of high-resolution videos, such as 2K and 4K formats, encounters significant challenges due to bandwidth limitations and network fluctuations. Existing neural-enhanced video streaming systems primarily struggle with two issues: the difficulty of recovering intra-frame high-frequency content and reusing the inter-frame content correlation. Addressing these challenges, this paper introduces a novel approach, designated as DoMo, which reconsiders the potential of mobile-side video super-resolution (SR) from a cloud perspective. We implement DoMo for the VP9 codec and test on real on-demand streaming media videos. Empirical results indicate that DoMo not only surpasses current state-of-the-art neural-enhanced solutions by achieving a 3.32 - 4.54 dB improvement in the peak signal-to-noise ratio (PSNR), but also outperforms traditional non-SR decoding methods by 6.80 - 8.89 dB.
Fan DANG
Assistant Professor
My research interests include industrial intelligence and edge computing.