VVC-MV-CM : a complexity-managed multiview extension for VVC with adaptive inter-view prediction

Tools

Gallena Watthage, Reka Sandaruwan and Fernando, Anil (2026) VVC-MV-CM : a complexity-managed multiview extension for VVC with adaptive inter-view prediction. Applied Sciences, 16 (7). 3254. ISSN 2076-3417 (https://doi.org/10.3390/app16073254)

[thumbnail of Watthage-Fernando-AS-2026-VVC-MV-CM-a-complexity-managed-multiview-extension]
Preview
 Text. Filename: Watthage-Fernando-AS-2026-VVC-MV-CM-a-complexity-managed-multiview-extension.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (531kB)| Preview

Abstract

Multiview video coding grows exponentially with the number of views, and VVC-based systems face particularly severe computational burdens from exhaustive inter-view prediction searches. We propose VVC-MV-CM, a complexity-managed multiview extension of VVC that combines rule-based pre-screening with CNN-based adaptive inter-view prediction bypassing within a two-stage decision engine. Performance trends are observed across 19 test sequences covering planar, arc, and spherical camera configurations under all-view and selected-view encoding modes. For planar all-view configurations, VVC-MV-CM-A achieves −52.7% BD-rate relative to MIV-A with 68% encoding time reduction. Arc arrangements yield competitive performance at −1.26% (all-view) and approximately −1% (selected-view) BD-rate. Spherical configurations demonstrate −19.8% (all-view) and −15.0% (selected-view) BD-rate gains, driven by multi-reference redundancy and temporal prediction prioritization. View density analysis reveals a 4.8 percentage-point compression difference between all-view and selected-view configurations, corresponding to approximately 2.4% efficiency gain per doubling of camera count. The proposed codec achieves 1.17–1.46× encoding time relative to MIV anchors with 18–36% decoding speedup, establishing configuration-adaptive prediction as an effective and deployable approach to multiview video coding across a wide range of geometric complexities and view-sampling densities.

ORCID iDs

Gallena Watthage, Reka Sandaruwan and Fernando, Anil ORCID logoORCID: https://orcid.org/0000-0002-2158-2367;
CORE (COnnecting REpositories)