Unsupervised 3D Shape Classification Algorithm Using Density Peaks

In this paper, we propose an unsupervised classification algorithm by using density peaks for automatic content-based 3D model classification. Firstly, the algorithm extracts multiple kinds of feature vectors for each model in the given shape collection. Secondly, it uses robust principal component analysis to denoise the feature vectors and reduce their dimensions simultaneously. Finally, the algorithm determines the number of categories of the 3D models and realizes an unsupervised classification in an intuitive and visual way by computing the density peaks of the feature vectors' distribution and a corresponding decision graph. Extensive experimental results show that the number of categories of clustering is much easier to determine and the results are more accurate and robust in our algorithm when compared with the traditional algorithms.