JVET-AP0186 AHG17/AHG4: A Generic Full-Reference Objective Quality Assessment Method for Compressed Video [J. Wang, X. Zhuang, J. Zhang, L. Yu (Zhejiang Univ.)]

Recent advances in block-based hybrid video codecs and neural network-based codecs for next-generation compression, alongside the widespread adoption of high dynamic range (HDR) and wide colour gamut (WCG) video content, pose significant challenges for existing objective quality metrics. These metrics often fail to maintain consistent correlation with human perception across different codec types, distortion patterns, and HDR/WCG content characteristics.

In this contribution, we propose FDIM, a generic full-reference objective quality assessment method with high correlation to human perception, strong generalization across traditional and learning-based codecs, and high robustness across SDR and HDR/WCG content, to support the objective assessment of next-generation video coding technologies.

It is recommended to consider FDIM as a candidate quality metric beyond PSNR, e.g., for studying perceptual coding algorithms, evaluating CfP proposals, etc.

The source code for FDIM had been released at https://github.com/MCL-ZJU/FDIM.

FDIM is a hybrid approach, averaging a learning based metric with traditional metric, as shown below:

Training data are obtained by crowd sourcing

FDIM has already being considered in AG 5, where investigation of metrics for better matching visual quality, this is an ongoing work item. In an activity such as the CfP, visual tests are necessary, and it would not be appropriate to over-burden proponents by computing various metrics. Further discussion was recommended in AG 5, which might also use the subjective results from the CfP to further investigate various metrics.

This was cross-checked in an AG 5 document

AHG4: Subjective quality testing and verification testing (2)

Contributions in this area were discussed during 1815–1910 on Monday 27 April 2026 (Joint meeting chaired by JRO, MW, GJS).