V-Nova, a leading provider of video compression solutions, has conducted a survey together with Intel and Meta to evaluate the impact of MPEG-5 Low Complexity Enhancement Video Coding (LCEVC) in enhancing the performance of existing codec implementations.
The study, published on 3rd October 2022 by SPIE, the International Society for Optics and Photonics, quantifies the impact of LCEVC in improving the compression efficiency vs complexity trade-offs of native SVT-AV1, x264 and x265 for VOD applications.
Over 650,000 encodes were tested, comprising of 140 self-similar clips, across 8 resolutions, multiple bitrates and 25 different pre-sets, using four different visual quality metrics: VMAF, VMAF_NEG, PSNR and SSIM.
Results showed improvements in:
● compression performance for all tested codecs across all operating points. Using dynamically optimized convex-hull methodology in conditions typically used for VOD encoding by streaming and social network services, the use of LCEVC showed quality improvements when used in conjunction with the tested codecs.
● computational performance for all tested codecs. The LCEVC-enhanced videos, for example, are shown to achieve the same quality levels at ~40% reduction in computations when used in conjunction with SVT-AV1. The combination of LCEVC + SVT-AV1 builds on already notable improvements of SVT-AV1 and showcases the concrete possibility for widespread and energy efficient deployments.
● decodability of AV1. Enhancing AV1 with LCEVC was shown to broaden the set of mobile devices capable of playing back full HD AV1 content. Mobile battery life was also extended by up to 50% compared with non-enhanced streams using state-of-the-art AV1 software decoders Dav1d and GAV1.
“The results published by SPIE have been the results of an amazing effort between all parties, leveraging on the knowledge, experience and methodology of Intel and Meta, to finally provide a set of quantitative results of the benefits of MPEG-5 LCEVC in a very relevant use case,” said Guido Meardi, CEO, V-Nova.
The joint paper is available free at the SPIE conference proceedings page, here.