How AI Technology is Dramatically Improving Video Compression for Broadcast and OTT Content Delivery

Video compression for broadcast TV services started more than 20 years ago. Since then, there has been a major codec standard created about every 10 years, with MPEG-2 released around 1995, AVC in 2005 and HEVC in 2015. Over time, several key improvements such as dual-pass encoding, statistical multiplexing and software migration were made to compression technology in order to boost performance. Tuning the algorithms and the algorithmic tools also allowed significant improvements to be made. — Now a new and disruptive technology — Artificial Intelligence (AI) — is driving the next frontier of video compression enhancements with the promise of faster advancements. AI is being used to improve several key areas: to achieve better video quality (VQ) at a given bit rate, or lower bit rate at the same VQ. Advances are also being made in other directions, like higher density to where the same VQ/bitrate efficiency will use less computing resources, and in providing better quality of experiences (QoE). — This paper will present three examples of AI applied to video encoding to optimize broadcast and OTT content delivery: Dynamic Encoding Style (DES) for a better VQ /bitrate trade-off, Dynamic Resolution Encoding (DRE) for better QoE and density, and Dynamic Frame rate Encoding (DFE) also for improved density and QoE. — In addition, the paper will explore the operational and end-user benefits enabled by AI and machine learning. Additionally, it will provide measurement for the applications that are presented and address the possible future evolutions of AI for video compression.

Published

2019-10

Content type

Original Research

Keywords

AI, Artificial Intelligence, ML, Machine Learning, Video, Compression, Density, CPU, Latency, Television, Live

DOI

10.5594/M001868