Research on perceptual quality metrics of 3D meshes

Publication Type:
Thesis
Issue Date:
2018
Full metadata record
In mesh processing operations, the 3D mesh is always subject to geometric distortions which may degrade the visual quality of the 3D mesh. As the ultimate receptor of 3D meshes is typically human eyes, it is of great significance to develop perceptual metrics for mesh quality assessment. While most existing research works in mesh quality assessment focused on the evaluation of local distortions on the mesh, we mainly study the perceptual importance of local regions to the mesh quality in this thesis. We propose five perceptual mesh quality metrics and validate the performance superiority of these metrics over state-of-the-art metrics in the experiments. Firstly, we investigate the relationship between the statistical descriptors of local distortions and the mesh quality. We propose the Tensor-based Perceptual Distance Measure-Spatial Pooling metric by exploiting the Support Vector Regression method. The experimental results demonstrate that the severely distorted regions on the mesh have a great impact on the mesh quality. Secondly, we propose the percentile weighting method to emphasize the impact of the severely distorted regions on the mesh quality. A local distortion-based visual importance model is built by assigning a great weight to a small portion of vertices that suffer from severe distortions. We propose the Tensor-based Perceptual Distance Measure-Percentile Weighting metric by applying the percentile weighting method at the stage of spatial pooling. Thirdly, we investigate the incorporation of mesh saliency into mesh quality assessment and build a visual importance model to emphasize the impact of salient regions on the mesh quality. We propose the Tensor-based Perceptual Distance Measure-Visual Saliency metric by taking the saliency value as the weighting coefficient of the local distortion. A saliency map synthesis method is proposed to assemble the salient regions from different individual saliency maps. The experimental results reveal that the performance gain of the synthetic saliency map over individual saliency maps depends on the similarity between individual saliency maps. Lastly, we propose the saliency weighting strategy to emphasize the impact of both severely distorted regions and salient regions on the mesh quality. We propose the Tensor-based Perceptual Distance Measure-Saliency Weighting metric and the Tensor-based Perceptual Distance Measure-Percentile Weighting-Saliency Weighting metric by applying the Minkowski method and the percentile weighting method respectively to the saliency-aware distortions. The experimental results reveal that the emphasis on the impact of salient regions improves the performance of the metric, but an overemphasis may cause a performance degradation.
Please use this identifier to cite or link to this item: