MS-GeodesicPSIM: Predicting the Quality of Static Mesh with Texture Map via multi-scale Geodesic Patch Similarity

To address the mesh quality assessment (MQA) problem, GeodesicP-SIM was proposed by jointly considering geometry and color features, demonstrating compelling performance in multiple benchmarks.However, GeodesicPSIM does not consider the multi-scale characteristics of human perception.To better mimic human subjective perception, we proposed a multi-scale MQA model called multi-scale Geodesic Patch Similarity (MS-GeodesicPSIM).Firstly, inspired by the multi-scale processing methods used in image and point cloud analysis, we propose a novel multi-scale representation of textured meshes based on mesh simplification techniques.Secondly, we extend GeodesicPSIM into a multi-scale version leveraging the proposed multi-scale representation.Specifically, we construct a multi-scale representation for the reference and distorted meshes, followed by fusing the results of GeodesicPSIM at different scales to obtain an overall quality score.Experimental results demonstrate the superior performance of the proposed MS-GeodesicPSIM compared to the single-scale GeodesicPSIM and other MQA metrics on three large and independent databases.Ablation studies further confirm that MS-GeodesicPSIM is robust to different model hyperparameter settings.The code for MS-GeodesicPSIM is available at https://github.com/ccccby/MS-GeodesicPSIM