Multi-Frame Optimized Virtual Reference Frame for Low-Delay Predictive Video Compression

View at publisher Summarise with VIRAT
Authors: Nimesh Pollwaththage, Yasith Ganearachchi, Prabhath Samarathunga, Joseph El Gemayel, Anil Fernando
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Conventional hybrid video codecs in low-delay configurations employ the first frame of a Group of Pictures (GOP) as the reference anchor in their prediction hierarchy, resulting in deterministic but non-adaptive reference structures. Although multiple reference frames are supported, the selection process remains purely temporal and lacks content-driven optimization. To address this limitation, we propose a Virtual Reference Frame (VRF) designed to act as an intelligent and statistically optimized substitute for conventional reference frames. The proposed VRF is synthesized through a constrained multi-frame optimization process that minimizes the cumulative mean squared error between the VRF and all other frames within a GOP. As a result, the VRF becomes a single representative frame that captures the dominant spatio-temporal characteristics of the GOP, reducing global residual energy and consequently the entropy of the transform coefficients to be encoded. Experimental evaluations conducted on grayscale video sequences using a low-delay encoder configuration demonstrate consistent bitrate savings and quality improvements compared to the baseline, confirming the efficiency, compatibility, and codec-agnostic applicability of the proposed approach within existing hybrid video coding frameworks.

Fields of Study & Subjects

Video Coding and Compression Technologies Advanced Data Compression Techniques Image and Video Quality Assessment

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–6
Publication date
3 February 2026
Language
EN