Perceptual quality assessment of 3D dynamic meshes: Subjective and objective studies

Nowadays, 3D mesh animations have been increasingly used in various applications, e.g., in digital entertainment and physically based simulation. In many applications, it is possible that a surface animation undergoes some lossy operations which can impair its perceptual quality. Since the end users of mesh animations are often human beings, the perceptual quality assessment of 3D dynamic meshes has recently emerged as an interesting and important research problem. In this paper, we will first of all present the results of a large-scale subjective experimental study carried out to collect human opinion scores of the perceptual quality of distorted dynamic meshes. We integrate in our subjective study a large set of 276 impaired mesh sequences with different distortion intensities and categories, such as spatial and temporal visual masking simulation, lossy compression and network transmission error. The constructed database is publicly available and can serve as ground-truth data for the performance evaluation of objective metrics of dynamic mesh perceptual quality. A comparative study is then conducted to benchmark existing objective metrics. The main finding of this objective study is that in general model-based quality metrics outperform image- and video-based quality metrics, but none of the existing metrics can achieve high correlation with human opinion scores on the whole subjective database and for all kinds of distortions. Finally, a simple but effective full-reference objective metric is proposed for the perceptual quality assessment of 3D dynamic meshes, which incorporates both spatial and temporal features and which provides a satisfactory correlation value when tested on the subjective database.