Graph-Based Loop Filtering Reduces Video Codec Complexity by 97%
Researchers proposed a graph-based fixed-filtering framework for Adaptive Loop Filtering (ALF) in the Enhanced Compression Model (ECM) that replaces 512 fixed filters with 16 graph filters. The approach uses polynomial graph filters to capture local structural information, achieving improved luma signal reconstruction performance while significantly reducing complexity.
Key Takeaways
- Replaced 512 traditional fixed filters with 16 graph filters using coefficient-space clustering.
- Achieved a 6.5% relative MAE reduction in luma signal reconstruction, outperforming the 5.0% reduction of the ECM baseline.
- Utilized 8-connected graph connectivity and a 5th-order polynomial to optimize the balance between performance and computational load.
- Targeted the Enhanced Compression Model (ECM), the primary software platform for developing video coding standards beyond Versatile Video Coding (VVC).
Why It Matters
This development addresses a critical bottleneck in next-generation video coding: the ballooning hardware complexity of in-loop filters. By using input-dependent graphs to represent multiple structurally similar pixel neighborhoods, the framework eliminates the redundancy found in gradient-based classifiers. For the industry, this signals a path toward achieving the ~27% bitrate gains targeted by the ECM project without prohibitive increases in decoder implementation costs. Strategists should monitor if this graph-based approach is adopted into the formal Joint Video Experts Team (JVET) Call for Proposals for the H.267 standard, currently projected for a late 2020s finalization.
Additional Context
The Enhanced Compression Model (ECM) serves as the experimental testbed for the Joint Video Experts Team (JVET), a collaboration between the ITU-T and ISO/IEC focused on the successor to Versatile Video Coding (VVC). Per industry reports from OFINNO in October 2025, the latest ECM updates (Version 19) have demonstrated roughly 27% bitrate savings over VVC. The project aims for an eventual 40% reduction, positioning it as the foundation for a future H.267 codec. However, the adoption of these tools is heavily contingent on balancing coding efficiency with the computational budgets of mobile and edge devices. While VVC adoption continues to gain traction in broadcast ecosystems like Brazil’s TV 3.0 (launched August 2025) and European DVB tuners, market analysts at Rethink Research noted in December 2025 that the codec faces competition from royalty-free alternatives like AV1. The next generation of video coding is increasingly focused on specialized applications, including Video Coding for Machines (VCM) and immersive media. According to Fraunhofer HHI in November 2025, European research leaders are now prioritizing collaborative standardization to define these 6G-era codecs, specifically exploring how to integrate emerging 3D rendering and neural network-based tools into the compression pipeline. Recent JVET meetings in 2024 and 2025 have intensified the focus on Neural Network-based Video Coding (NNVC), which reports BD-rate reductions in the 6–14% range. Despite these gains, the computational overhead remains significant—often an order of magnitude higher than VVC. Consequently, complexity-reduction techniques, such as the graph-based filtering framework, are vital for ensuring that the high-efficiency gains observed in ECM testing are commercially viable for volume hardware production.
Read full article at arxiv.org
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source