基于多特征融合的三维网格质量评价方法

3D mesh quality assessment method based on multiple feature fusion

三维网格客观质量方法研究对三维网格处理方法设计有重要指导意义。现有的三维网格质量评价方法大多采用单一特征来表征三维网格的失真程度,难以全面描述人眼对三维网格失真的视觉感知。在提出基于分区的曲率奇异值分解的结构相似性(CSSIM)特征的基础上,进一步提出多特征融合的三维网格质量评价方法。提取的特征通过随机森林的聚合,训练得到三维网格的客观质量评价模型,预测评价分数。经过在两个公开的三维网格数据库(LIRIS/EPFL general-purpose和LIRIS Masking)上的验证,其结果表明相比现有代表性方法,所提出方法的评价结果与人眼主观感知评价结果的一致性更好,表面其符合人眼的视觉感知。

The research on objective quality assessment method of 3D mesh has important guiding significance for the design of 3D mesh processing method. Most existing 3D mesh quality assessment methods use a single feature to characterize the degree of the 3D mesh distortion,which is difficult to describe the human eye’s visual perception of 3D mesh distortion. Based on the structural similarity( CSSIM) feature of partition-based curvature singular value decomposition,a multiple feature fusion based 3D mesh quality assessment method is proposed in this paper. The extracted features are trained by random forest to obtain an objective quality assessment model of 3D mesh which is used to predict the quality score of 3D mesh. After verifying in two public 3D mesh databases( LIRIS/EPFL general-purpose and LIRIS Masking),the experimental results show that compared with the existing representative method,the evaluation results of the proposed method are more consistent with the subjective evaluation results,which means it’s more in line with the visual perception of the human eye.