3D shape retrieval using the filtering method

Curve-skeletons are a 1D subset of the medial surface of a 3D object. Curve-skeletons with thickness are used in the 3D shape matching field. In the present paper, we introduce a filtering method to retrieve similar or partially similar shapes from a 3D shape database. The concept of curve-skeletons with thickness is that define the distance transform (DT) value of the voxel on the curve-skeleton as the thickness of the skeleton. This concept can be used to efficiently retrieve shapes for which there are no branches on their curve-skeletons. Therefore, the retrieval of shapes having curve-skeletons that are structural complex is challenging. We herein propose a general and robust method by which to retrieve mostly similar shapes by their parts. The proposed method filters out the shapes that have the greatest number of similar parts. The main idea is that if two 3D models are similar, then most of the parts of one of the models have a similar correlative part to the shape of the other model. The similar parts are similar with respect to their thickness distributions along the curve-skeletons. The proposed method is particularly suited to searching objects with a distinct part structure and is invariant to part architecture. The proposed method can also be used in web shape search engines to find matching 3D shapes from a 2D curve.