Three dimensional object retrieval using a skeleton-based approach

This paper proposes a new technique for 3D object retrieval using skeletons of objects´ orthographic projections. The proposed method exploits the 2D multi-views wherein silhouettes are obtained for different viewpoints for each model. A feature vector is extracted for the skeleton of each viewpoint of an image obtained by contour partitioning with Discrete Curve Evolution (DCE). The vectors are further trained in a matrix for pairwise comparison. The experimental results have been given by conducting experiments on the Princeton Shape-Benchmark (PSB), a publicly available database of 3D models. The obtained results are quite encouraging in terms of accuracy.