Blind 3D mesh visual quality assessment using support vector regression

Various visual distortions can inevitably affect the 3D meshes during their transmission and geometrical processing. In most practical cases, blind quality assessment becomes a challenging issue due to the unavailability of reference meshes and distortion related information. In this paper, we present a novel method to blindly assess the quality of 3D meshes. This method relies on a feature learning based approach to predict the objective quality scores. For this, we propose the mesh dihedral angles statistics as a feature and the support vector regression (SVR) as a learning tool based quality predictor. The proposed method takes into account the main functions of the human visual system (HVS) by introducing the visual masking and the saturation effects. Experiments have been successfully conducted on LIRIS/EPFL general-purpose, LIRIS Masking and UWB compression databases. The obtained results show that the proposed method provides good correlation and competitive scores comparing to some influential and effective full and reduced reference existing methods.