Objective Evaluation System for 2K Video Quality Transcoded from 4K Source Using Dynamic Thresholds and Statistical Method
Originally Aired - Sunday, April 14 | 11:50 AM - 12:10 PM PT
Pass Required: Core Education Collection Pass
Don't have this pass? Register Now!
Create or Log in to myNAB Show to see Videos and Resources.
Videos
Resources
{{video.title}}
Resources
This Session Has Not Started Yet
Be sure to come back after the session starts to have access to session resources.
Many video files are generally handled with compression encoding as they contain enormous amounts of data. However, compression can degrade image quality, thereby underscoring the critical need for quality assurance in video production. Several methods are available for quantifying video quality objectively, and there are already products that systematically measure and detect quality anomalies. For instance, to assess the extent of degradation in a compressed video via comparison with its original source, established metrics such as peak signal-to-noise ratio (PSNR) or the ITU-T J.144 international standard methods exist. However, these evaluation methodologies can be applied only when comparing images with the same resolution, which does not solve the problems when assessing the quality of videos with different resolutions. Therefore, even a 2K video transcoded from 4K cannot systematically be evaluated for quality relative to the original 4K source. Consequently, human operators must expend real-time effort on reviewing both 4K and 2K videos during packaging before broadcasting or distribution, lowering the productivity of integrated 4K/2K simal productions. To address this, we have devised and prototyped an innovative method for the automatic evaluation of 2K video quality transcoded from 4K. Specifically, to enable pixel-by-pixel comparisons between 4K and 2K videos, a spatial two-dimensional filter is employed for predicting reference pixel values for every pixel in the 2K video, derived from multiple pixels in the original 4K video. These predicted reference pixel values were compared with the actual pixel values of the 2K video, thereby enabling the calculation of international standard image quality metrics such as PSNR and the identification of quality anomalies. Nevertheless, a video with spatially and temporally complicated features may degrade the image quality metrics, resulting in false alarms even when the resolution conversion is performed correctly. As a solution, we adapted our detection thresholds for quality anomalies, incorporating indicators that capture spatial or temporal complexity. Furthermore, for the detection of noise limited to specific regions within the 2K video, we employed small-area PSNR as an image quality metric. However, in scenes with high complexity, this metric tends to degrade, potentially increasing the number of false detections although many of such noises cannot be noticed by human beings. Therefore, to ensure accurate detection even in complex scenes, we devised a methodology that detects quality anomalies by applying a statistical method to this metric. Accordingly, the automated quality check of 2K content can now match or surpass the accuracy of human previews. This advancement reduces the significant cost and effort associated with traditional human review, resulting in both increased productivity and quality assurance in integrated 4K/2K simal productions.
Presented as part of:
Speakers
