Wednesday 16 April 2025
As we stream more and more content online, from our favorite TV shows to live sports events, ensuring that the video quality is up to par has become a top priority for streaming services. One of the key challenges in achieving this is determining when to switch between different video resolutions, a process called resolution cross-over. A new study sheds light on how we can improve our understanding of this process and develop more accurate methods for predicting when to make these changes.
Currently, many streaming services use objective video quality metrics (VQMs) like PSNR or VMAF to determine the best resolution for a given bandwidth. These metrics are calculated by comparing the original video signal with the compressed version sent over the internet. However, researchers have found that these metrics often fail to accurately predict when to switch between resolutions, leading to suboptimal video quality.
To address this issue, the study proposes a new metric called Resolution Cross-Over Quality Loss (RCQL), which measures the quality loss caused by incorrect resolution cross-over decisions. This approach is more subjective in nature, relying on human judgment to evaluate the quality of different video streams. The researchers collected a dataset of 10,000 videos and had participants rate their quality, allowing them to develop a more accurate understanding of what constitutes high-quality video.
The study also highlights the limitations of using absolute category rating (ACR) methods, which are commonly used in subjective video quality assessment studies. ACR involves asking participants to rate the overall quality of a video on a scale from 1-5, but this approach has been shown to introduce significant uncertainty and errors in resolution cross-over predictions.
In contrast, pairwise comparison (PC) methods involve comparing different videos side-by-side and asking participants which one appears higher quality. This approach is found to be more accurate in predicting resolution cross-over decisions. The study demonstrates the effectiveness of PC by using it to evaluate the performance of various VQMs on a new dataset called LSCO.
The results show that while PSNR performs well for low-resolution video streams, its accuracy drops significantly at higher resolutions. On the other hand, EQM, another popular VQM, outperforms PSNR across different resolutions but struggles with lower-quality videos. The study highlights the importance of considering the specific characteristics of each video stream when making resolution cross-over decisions.
The implications of this research are significant for streaming services looking to improve their video quality.
Cite this article: “Unlocking the Secrets of Video Quality Assessment: A New Mathematical Framework Reveals Hidden Truths in Cross-Over Accuracy”, The Science Archive, 2025.
Video Quality Metrics, Resolution Cross-Over, Streaming Services, Video Resolution, Bandwidth, Psnr, Vmaf, Subjective Evaluation, Pairwise Comparison, Objective Assessment
