Content-oriented3D Model Retrieval

With the recent increase in the number of3D models,3D model retrieval appli-cation is widely used in people’s life, and development of the technology for effective retrieval of3D models has become an important issue.Firstly, a fast and robust3D retrieval method is proposed based on a novel weight-ed structural histogram representation, which has the following steps:adaptively seg-ment any3D shape into a group of meaningful parts to generate local distribution matrixes, integrate all the local distribution matrixes into a global distribution matrix, simultaneously considering their weight factors, and retrieve3D shapes by calculating the distance between their global distribution matrixes.Considering histogram descriptor is a discrete description about shape feature, we propose a new shape matching method based on kernel density estimation, by first extracting angular and distance feature pairs from pre-processed3D models, then esti-mating their kernel densities after quantifying the feature pairs into a fixed number of bins. During3D matching, we adapt the KL-divergence as a distance of3D compari-son.On these bases, we further explore spherical harmonics based3D model retrieval methods. Considering geometrical details may be lost during the harmonics decom-position process and make these methods not robust to solve shape matching or3D retrieval tasks, we use a radial basis function to model local geometrical constraints of a3D shape, and present a novel Gaussian spherical convolution function to simulta-neously integrate global shape constraints and local geometrical details. By spherical harmonics decomposition of the convolution, an efficient3D shape retrieval algorithm is introduced. Furthermore, considering retrieval results based on spherical harmonics are not sufficiently better than other kinds of algorithms, we further analyze spherical harmon-ics and find that useful information maybe lost during extracting spherical features. As an improvement, we finally propose a new algorithm by mixing spherical harmonic and geometric histogram methods in3D retrieval. Our experimental results prove that the proposed algorithm has a surprising retrieval performance.