Adaptive Picture Mirroring for Video Coding Efficiency Enhancement

Abstract

The exponential growth of global video traffic imposes severe constraints on available transmission bandwidth, necessitating continuous improvements in compression efficiency. While modern hybrid video codecs optimize bit allocation locally, they often fail to exploit the global characteristics, such as dominant edge directionality. To address this limitation, this paper proposes a bandwidth-efficient, lightweight pre-processing framework that adaptively mirrors encoded video frames to align their structural features with the encoder prediction tools. Unlike previous approaches that rely on computationally expensive brute-force search or complex neural networks, the proposed method uses a fast, heuristic decision algorithm based on edge-direction analysis. Extensive evaluations were conducted using the state of-the-art Enhanced Compression Model (ECM) and the industry-standard HEVC, covering both All-Intra and Random Access (RA) configurations to simulate realistic broadcasting scenarios. The results confirm that the framework yields noticeable coding gains, particularly for content with dominant diagonal geometric structures, achieving average BD-rate savings of $\mathbf {-0.23\%}$ (ECM All-Intra) and $\mathbf {-0.11\%}$ (ECM Random Access), with specific high-definition sequences (Class B) reaching gains up to $\mathbf {-0.58\%}$ . Cross-codec validation using HEVC confirms the universality of the proposal, yielding an average savings of $\mathbf {-0.30\%}$ . Crucially, complexity analysis reveals that the method introduces negligible computational overhead (approximately 30 ms of pre-analysis per 4K frame, making it a highly viable, lightweight extension for video distribution pipelines where reference encoder runtimes dominate.