|
Work item:
|
J.FR-NP-REF
|
|
Subject/title:
|
Full-reference video quality assessment in the presence of non-pristine reference sources
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-12 (Medium priority)
|
|
Liaison:
|
-
|
|
Supporting members:
|
Korea (Rep. of)
|
|
Summary:
|
The accuracy of Full-Reference (FR) video quality measurement is traditionally predicated on the assumption that the reference source (SRC) is "pristine"—meaning it is free from any perceptible impairments caused by capturing and/or video processing. However, in practical deployment scenarios, a pristine reference signal often is not be available. Reference SRCs often contain inherent degradations, such as camera sensor noise, capture-related artifacts, or residual coding noise from previous processing stages.
When the reference signal itself is non-pristine, conventional FR models that rely on a pixel-to-pixel or feature-to-feature comparison may produce unreliable quality scores, as they assume that the source signals have no impairments. This work item aims to develop FR models in the presence of non-pristine reference sources.
The output of this work item may be used for the following use cases:
– QoE of streaming or social network services of user-generated contents where the reference SRCs often contain perceptible impairments.
– QoE measurements in scenarios where network bandwidth cannot be guaranteed.
– Video recorded in hostile or unstable environments (e.g., poor lighting, drone footage in windy conditions, or rapid movement causing focus issues and camera jiggle)
– Choosing coding parameters for video with camera impairments
It is noted that there are ITU FR/RR Recommendations with different scopes under different assumptions. Thus, care must be taken:
– The output of this work item must not be used to replace or suppress existing ITU FR and RR Recommendations, as they have different scopes.
– Furthermore, the validation datasets produced by this work item must not be used to undermine the performance of these existing Recommendations, as conventional FR/RR models assume that reference signals are of high quality.
This work item is expected to include discussions on the use case, subjective test method, and expected model outputs.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-06-28 17:36:42
|
|
Last update:
2026-06-28 17:39:24
|
|
