A Binary Volumetric Data Retrieval Method Based on Neighboring Voxel Pattern Descriptors

Abstract

The use of 3D polygonal data and volumetric data has attracted attention in recent years, because of advancement in 3D printing technology. The amount of online 3D data is increasing, and systematic searches and classifications are required for 3D model databases. It is important to extract shape descriptors from such 3D data. This paper describes a shape descriptor extraction method for volumetric data, and proposes descriptors called Neighboring Voxel Patters (NVP). These shape descriptors are computed based on voxel patterns in target volumetric data. The technique uses small local 3 x 3 x 3 voxel regions to compute all the possible voxel pattern combinations, and classifies these voxel patterns into a small number of groups. NVP descriptors are computed easily from volumetric data. Although the descriptors are small in size, these descriptors efficiently classify 3D models that are represented by volumetric data. Preliminary experiments are conducted with voxelized 3D model benchmark data. Various sizes of volumetric data are analyzed for similarity retrieval experiments, and the web-based similarity retrieval system is implemented.