Pose Insensitive 3D Retrieval by Poisson Shape Histogram

摘要

With the rapid increase of available 3D models, content-based 3D retrieval is attracting more and more research interests. Histogram is the most widely in constructing 3d shape descriptor. Most existing histogram based descriptors, however, will not remain invariant under rigid transform. In this paper, we proposed a new kind of descriptor called poisson shape histogram. The main advantage of the proposed descriptor is not sensitive for rigid transform. It can remain invariant under rotation as well. To extract poisson shape histogram, we first convert the given 3d model into voxel representation. Then, the poisson solver with dirichlet boundary condition is used to get shape signature for each voxel. Finally, the poisson shape histogram is constructed by shape signatures. Retrieving experiments for the shape benchmark database have proven that poisson shape histogram can achieve better performance than other similar histogram-based shape representations.