Quality Metric with Video Characteristics on Scalable Video Coding

Hana Yoo, Cheon-Seog Kim, Hojun Lee, Sungho Jin, Yong Man RoJournal of Broadcast Engineering 2008 · DOI

Abstract

As video consumption grows across diverse devices and network conditions, ensuring a consistent Quality of Service (QoS) has become a critical challenge. Scalable Video Coding (SVC) offers a solution by providing a single bitstream that can be adapted spatially, temporally, and in quality. However, traditional objective metrics like PSNR often fail to reflect human perception. This paper proposes a universal quality metric for SVC that incorporates spatial and temporal video characteristics. By modeling the relationship between physical features and subjective viewer scores, the proposed metric achieves a 0.96 correlation with human perception, significantly outperforming standard methods.

The Challenge of Measuring Video Quality

In modern media delivery, video must be optimized for a vast range of displays, from mobile phones to high-resolution monitors. SVC is highly effective in these heterogeneous environments because it allows for flexible bitstream extraction.

Despite its technical advantages, a major hurdle remains: accurately measuring the "Quality of Experience" (QoE). Traditional metrics like Peak Signal-to-Noise Ratio (PSNR) are computationally simple but do not account for how humans actually perceive video. For instance, compression artifacts are often more noticeable in static scenes than in high-motion sequences. Existing metrics struggle to generalize across different content types, making it difficult for service providers to optimize delivery based on actual viewer satisfaction.

Proposed Methodology

The researchers developed a new Quality Metric (QM) designed to bridge the gap between objective data and subjective perception. The approach follows a structured pipeline:

  1. Feature Extraction: Using MPEG-7 descriptors, the system quantifies Spatial Value (SV) (texture complexity via Edge Histogram Descriptors) and Temporal Value (TV) (motion intensity via Motion Activity).
  2. Subjective Assessment: Experiments were conducted with 16 viewers to obtain Differential Mean Opinion Scores (DMOS) across various resolutions and bitrates, providing a human-centric ground truth.
  3. Modeling and Regression: Mathematical models were created to link SV, TV, frame rate, and resolution to subjective scores. The Least Square Method (LSM) was used to determine optimal coefficients for the final metric equation.

Content-Awareness and Subjective Perception

A core insight of this study is that subjective quality is not a fixed function of bitrate or PSNR; it is heavily influenced by the video's spatial characteristics.

Subjective Quality (SQ) vs. PSNR for different Spatial Values (SV), showing how human perception varies with content complexity.

As shown in Figure 1, for the same PSNR value, viewers perceive quality differently depending on the SV level. In high-complexity scenes (High SV), viewers are often more forgiving of certain artifacts, leading to higher subjective scores compared to low-complexity scenes at the same PSNR. This demonstrates why content-aware metrics are essential for accurate quality assessment.

Performance and Results

The proposed QM demonstrated superior performance and consistency compared to existing standards. While PSNR achieved a correlation of 0.79 and the Video Quality Metric (VQM) reached 0.90, the proposed QM achieved an average correlation coefficient of 0.96 with subjective human scores.

Correlation with Subjective Quality (SQ) across varying Spatial Values (SV) for the proposed QM vs. the default metric.

Figure 4 highlights the robustness of the proposed metric. While the default metric's accuracy fluctuates significantly depending on the video's spatial complexity, the proposed QM maintains a high and stable correlation (0.95–0.97) across all tested categories. This makes it a reliable tool for general-purpose video quality assessment.

Conclusion

By integrating spatial and temporal characteristics into the quality calculation, this research provides a robust framework for measuring perceived video quality in SVC environments. The high correlation with human judgment suggests that this metric can serve as an effective decision tool for SVC bitstream extraction, ensuring that users receive the best possible visual experience regardless of their device or network constraints. Future work may extend this model to account for network-induced issues like packet loss and jitter.

Summary updated

Abstract

본 논문에서는 영상 특성에 제약 없이 적용 가능한 SVC 기반 범용 품질 메트릭을 제안한다. 이종 환경에서의 비디오 콘텐츠 소비가 증가하면서 서비스 품질(Quality of Service, QoS)은 중요한 이슈 중 하나가 되었다. 또한 스케일러블 비디오 코딩(Scalable Video Coding, SVC)은 하나의 부호화된 비트스트림이 공간적, 시간적 그리고 품질적으로 다양한 확장이 가능하다는 장점을 이유로 이종 환경에서 적응성이 높은 영상 부호화 기술로 주목받고 있다. 제한된 환경에서의 서비스 품질 보장을 위해서는 서비스의 사용 주체인 사람이 느끼는 주관적 비디오 품질이 고려되어야 한다. 주관적 품질은 영상의 시간적, 공간적 특성 모두에 영향을 받기 때문에 영상의 특성에 대한 고려 역시 중요하다. 제안하는 방법의 효율성은 실험을 통해 측정된 주관적 품질 값과 제안된 품질 메트릭을 이용해 측정된 품질 지수 사이의 상관계수를 이용하여 검증하였으며 제안하는 품질 메트릭은 SVC의 품질 확장 결정 기준으로 사용될 수 있다. In this paper, we proposed the qualify metric based on SVC and the subjective quality. The proposed quality metric is for a general purpose. It means we can use it for any video sequences regardless of its temporal and spatial characteristics. The Quality of Service(QoS) is one of the important issues in heterogeneous environment which has diverse restrictions such as limited network bandwidth and limited display resolution. Scalable Video Coding(SVC) is the efficient video coding skill in heterogeneous environment. Because SVC can be adapted to various quality bitstreams using three scalabilities(spatial, temporal, and SNR) from one bitstream which has full scalability. To maximize the QoS in this environment, we should consider the subjective quality which is the viewer response. And also we should consider temporal and spatial characteristics of video sequence because the subjective quality is affected by temporal and spatial characteristics of video sequence. To verify the efficiency of the proposed method, we perform subjective assessments. The experimental results show that the proposed method has high correlation with subjective quality. The proposed method can be a decision tool of SVC birstream extraction.
BibTeX
@article{yoo2008quality,
  author = {Hana Yoo and Cheon-Seog Kim and Hojun Lee and Sungho Jin and Yong Man Ro},
  title = {{Quality Metric with Video Characteristics on Scalable Video Coding}},
  journal = {Journal of Broadcast Engineering},
  year = {2008},
  doi = {10.5909/jbe.2008.13.2.179},
}
Figure