Stretching-Robust Laplace Spectral Descriptor for Non-rigid 3D Shape Retrieval

This paper proposes a framework based on harmonic mean normalized Laplace-Beltrami spectral descriptor for non-rigid 3D shape retrieval. A series of experiments show harmonic mean normalization is suited to classification of stretched shapes, and is robust to isometric transformation, holes, local scaling, noise, shot noise and sampling. To better distinguish among shapes with fine or rough details, weighting method and fusion method are employed. We use the two methods to reduce the adverse impact of high frequency when the shapes with fine and rough details are distinguished. In the experiments, three 3D shape retrieval benchmarks are used, and our approach has better performance than other state-of-the-art methods on both retrieval accuracy and time performance for stretched non-rigid 3D shapes.