Harmonic Shape Histograms for 3D Shape Classification and Retrieval

In this paper, we present a novel approach towards 3D shape recognition and retrieval using histograms of rotation invariant local features. Features are extracted for every point of voxelized 3D shape objects by use of functions on spheres which are invariant towards rotation of the object. The fast computation of the local features is performed via convolution methods in frequency space. Histograms of these features describe an object in terms of distributions of local geometric properties such as orientation and angle of edges, distances and convexity. Object classification is performed by Support Vector Machines with histogramintersection kernels. In experiments on the Princeton Shape Benchmark [1], our approach outperformed many existing methods in several classification and retrieval tasks.