Research On 3D Model Segmentation And Classification Algorithms

With the development of 3D scanning technology and modeling technology, 3D models have been widely used in the field of games, virtual reality, city planning, film, television animation and biology and so on. The rapid increase of the number of 3D models has brought new challenges to digital geometry processing. The urgent problem to be solved is how to sort and classify them based on their geometric shapes. And, in the field of geometry modeling and model processing, the automatic segmentation of 3D model is a basic operation. It can not only help the understanding of graphics, but also is very important for many computer graphics problems, including the mesh parameterization, skeleton extraction, resolution modeling and graphics retrieval and so on. In this paper, the classification and segmentation of 3D models are studied, and two related algorithms are proposed.For 3D shape segmentation, we propose a novel unsupervised algorithm for automatically segmenting a single 3D shape or co-segmenting a family of 3D shapes using deep learning. The algorithm consists of three stages. In the first stage, we pre-decompose each 3D shape of interest into primitive patches to generate over-segmentation and compute various signatures as low-level shape features. In the second stage, high-level features are learned, in an unsupervised style, from the low-level ones based on deep learning. Finally, either segmentation or co-segmentation results can be quickly reported by patch clustering in the high-level feature space. The experimental results on the Princeton Segmentation Benchmark and the Shape COSEG Dataset exhibit superior segmentation performance of the proposed method over the previous state-of-the-art approaches.For 3D shape classification, we propose a novel unsupervised 3D shape classification algorithm. The method significantly improves the classification accuracy by using the density peaks clustering. The method first extracts multiple kinds of features vectors for each model in the given shape collection. Then, it uses robust principal component analysis to denoising and dimension reduction of the feature vectors simultaneously. Finally, the algorithm classifies these 3D shapes by employing density peak clustering. The advantage of our method is that the number of categories of the 3D model can be determined in an intuitive and visual way through decision graphs, which distinguishes our algorithm from traditional classification algorithms. Extensive experimental results show that the proposed method is more accurate and robust compared with the traditional algorithms.