Generalized Local-to-Global Shape Feature Detection Based on Graph Wavelets
Nannan Li,Shengfa Wang,Ming Zhong,Zhixun Su,Hong Qin
DOI: https://doi.org/10.1109/TVCG.2015.2498557
Abstract:Informative and discriminative feature descriptors are vital in qualitative and quantitative shape analysis for a large variety of graphics applications. Conventional feature descriptors primarily concentrate on discontinuity of certain differential attributes at different orders that naturally give rise to their discriminative power in depicting point, line, small patch features, etc. This paper seeks novel strategies to define generalized, user-specified features anywhere on shapes. Our new region-based feature descriptors are constructed primarily with the powerful spectral graph wavelets (SGWs) that are both multi-scale and multi-level in nature, incorporating both local (differential) and global (integral) information. To our best knowledge, this is the first attempt to organize SGWs in a hierarchical way and unite them with the bi-harmonic diffusion field towards quantitative region-based shape analysis. Furthermore, we develop a local-to-global shape feature detection framework to facilitate a host of graphics applications, including partial matching without point-wise correspondence, coarse-to-fine recognition, model recognition, etc. Through the extensive experiments and comprehensive comparisons with the state-of-the-art, our framework has exhibited many attractive advantages such as being geometry-aware, robust, discriminative, isometry-invariant, etc.