3D model retrieval based on volume distribution

The key problem of 3D model retrieval is the technique of feature extraction. Some approaches have appeared in recent years. In this paper, we propose a new technique of feature extraction based on volume distribution for measuring similarity between 3D models. It is fast and stable, and can process various 3D mesh models with arbitrary geometry and topology. The main idea is 3D models are represented by volume distribution along three principal axes. First principal component analysis (PCA) is used for rotation normalization; then we transform the model's center of gravity to the origin for translation normalization. After the normalization, we use a set of parallel planes to slice the model into several high equal segments along three principal axes, respectively. From cross section in each plane, a polygon set composed of one or more polygons can be obtained and the area of each cross section can be consequently computed. By normalizing the areas, volume distributions of the model along the principal axes can be plotted by three characteristic curves and the curves give the model a description of the shape. Experiment results are given in this paper to show the efficiency and discriminative power of our approach.