Surface-Aware Weighted Partial Matching for Segmented 3D Shape Retrieval

Abstract

Following the growing usage of three-dimensional models, searching and browsing models in large 3D databases has become a challenging task that has attracted the focus of numerous research efforts in the area. Indexing 3D models for mesh matching can be a complicated challenge, requiring advanced computational algorithms and dedicated tools for information extraction. Most current methods treat the similarity problem between objects through global matching strategies. However, the majority of these approaches do not allow for part-based querying, in addition to their limited effectiveness for classes with articulated and incomplete objects. Recently, some researchers have focused their research on 3D shape retrieval with partial matching, which appears to be a relevant solution to these issues. Partial shape retrieval continues to pose significant challenges in 3D object analysis, especially when comparing models that contain different numbers of segments. To address this issue, we introduce a new similarity metric that incorporates both the similarity between successfully mapped segments and applies a penalty for segments that remain unmatched, with each component weighted according to surface area. This surface-aware weighting approach strengthens the system’s ability to handle partial matches and segmentation inconsistencies more effectively. Through comprehensive evaluation on a segmented 3D shape dataset using established retrieval metrics, our method demonstrates substantial improvements compared to traditional similarity measures.