Visual Mesh Quality Assessment Using Weighted Network Representation

This paper addresses the critical task of evaluating the visual quality of triangular mesh models. We introduce an innovative approach that leverages weighted graphs for this purpose. Motivated by the growing need for accurate quality assessment in various fields, including computer graphics and 3D modeling, our methodology begins by generating saliency maps for each distorted mesh model. These models are subsequently transformed into a network representation, where mesh vertices are nodes and mesh edges are edges in the graph. The determination of vertex weights relies on the salience values. We then extract a wide range of topological properties and compute statistical measures to create a signature vector. To predict the quality score, we rigorously evaluate the performance of three regression algorithms. Experiments span four publicly available databases designed for mesh model quality assessment. Results demonstrate that the proposed approach excels in this task, showcasing remarkable correlations with subjective evaluations. This preliminary analysis paves the way for further research to address potential limitations and explore additional applications of mesh network representation.