Q-STAR:A Perceptual Video Quality Model Considering Impact of Spatial, Temporal, and Amplitude Resolutions

Yen-Fu Ou, Yuanyi Xue, Yao Wang

Abstract

This paper presents Q-STAR, a unified model for predicting the perceptual quality of compressed video on mobile devices. By investigating the joint impact of spatial, temporal, and amplitude resolution (STAR), we developed a multiplicative model that captures quality degradation through three content-dependent parameters. Subjective tests involving 189 processed video sequences revealed that while temporal resolution effects are largely independent, spatial and amplitude resolutions exhibit significant interaction. The resulting model achieves a Pearson Correlation Coefficient (PCC) of 0.991, demonstrating exceptional accuracy across multiple datasets and encoding standards.

Introduction

Mobile video streaming requires a delicate balance between spatial resolution (SR), temporal resolution (TR), and quantization stepsize (QS). Engineers must often decide whether to sacrifice frame rate, resolution, or compression quality to meet bandwidth constraints. While previous research has examined these factors in isolation, the Q-STAR model provides a comprehensive framework to understand their combined effect on human perception, specifically within the context of high-pixel-density mobile screens.

Methodology

The researchers conducted extensive subjective quality tests using a TI Zoom2 mobile development platform. A total of 60 viewers evaluated 189 processed video sequences (PVSs) derived from seven distinct source contents. The test configurations varied across three spatial resolutions (QCIF, CIF, 4CIF), three temporal resolutions (7.5, 15, 30 Hz), and three quantization levels.

The subjective testing workflow: (a) Welcome Screen, (b) Playback Screen, and (c) Rating Screen.

Participants rated the videos on a 10-level scale ranging from "Bad" to "Excellent." This structured approach allowed for the isolation of individual STAR components and the observation of their interactions.

The Q-STAR Model

The core of the research is the discovery that quality degradation follows a generalized inverse exponential function for each dimension. The total quality is modeled as the product of three normalized functions:

$$QSTAR(s, t, q) = f_q(q; s_{max}) \cdot f_s(s; q) \cdot f_t(t)$$

A key insight from the study is the separability of temporal resolution. The rate of quality decay due to reduced frame rates was found to be independent of spatial resolution and compression levels. Conversely, spatial resolution and quantization (QS) interact significantly; high compression levels introduce blurring that exacerbates the loss of detail in lower-resolution videos. The model accounts for this by making the spatial decay parameter a function of the quantization level.

Results and Validation

The Q-STAR model demonstrated remarkable performance, achieving a Pearson Correlation Coefficient (PCC) of 0.991 and a Root Mean Square Error (RMSE) of 0.035 on the primary dataset. To ensure the model's robustness, it was validated against five external datasets featuring different encoders (H.263, MPEG-2) and resolutions up to 1080i HDTV.

A sample frame from one of the source sequences used to evaluate the Q-STAR model.

Across these diverse conditions, the model maintained high predictive power, typically exceeding a PCC of 0.85. This suggests that the multiplicative STAR framework is a fundamental characteristic of human video perception rather than an artifact of specific test conditions.

Conclusion

Q-STAR offers a simple yet powerful tool for optimizing mobile video delivery. By using only three content-dependent parameters, it accurately predicts how viewers will perceive changes in resolution and frame rate. Future work aims to automate the estimation of these parameters directly from video features, enabling real-time, content-aware streaming optimizations.

Summary updated

Abstract

In this paper, we investigate the impact of spatial, temporal and amplitude resolution (STAR) on the perceptual quality of a compressed video. Subjective quality tests were carried out on a mobile device. Seven source sequences are included in the tests and for each source sequence we have 27 test configurations generated by JSVM encoder (3 QP levels, 3 spatial resolutions, and 3 temporal resolutions), resulting a total of 189 processed video sequences (PVSs). Videos coded at different spatial resolutions are displayed at the full screen size of the mobile platform. Subjective data reveal that the impact of spatial resolution (SR), temporal resolution (TR) and quantization stepsize (QS) can each be captured by a function with a single content-dependent parameter. The joint impact of SR, TR and QS can be accurately modeled by the product of these three functions with only three parameters. We further find that the quality decay rates with SR and QS, respectively are independent of TR, and likewise, the decay rate with TR is independent of SR and QS, respectively. However, there is a significant interaction between the effects of SR and QS. The overall quality model is further validated on five other datasets with very high accuracy. The complete model correlates well with the subjective ratings with a Pearson Correlation Coefficient (PCC) of 0.991.
BibTeX
@article{ou2014q,
  author = {Yen-Fu Ou and Yuanyi Xue and Yao Wang},
  title = {{Q-STAR:A Perceptual Video Quality Model Considering Impact of Spatial, Temporal, and Amplitude Resolutions}},
  journal = {IEEE Transactions on Image Processing},
  year = {2014},
  doi = {10.1109/tip.2014.2303636},
}
Figure