三维模型分割与分类算法研究

发布时间：2018-10-15 16:29

【摘要】：随着三维扫描技术和建模技术的不断发展,三维模型被大量运用到游戏、虚拟现实、影视动画、生物学等领域。三维模型数量的快速增加为数字几何处理提出了新的挑战,其中亟待解决的问题就是如何基于它们的几何形状进行整理和分类。另外,在几何建模和模型处理等领域,如何对三维模型进行自动地分割也变得越来越重要。它不但能帮助图形理解,而且对很多计算机图形学问题也是至关重要的,包括网格参数化、模型简化、异常检测、骨骼提取、分辨率建模和图形检索等等。本文对三维模型分类与分割问题进行了研究,并分别提出了相关的算法。在三维模型分割算法中,该算法可以自动地对单个三维模型进行分割或对同一类三维模型进行协同分割(co-segmentation)。首先对每个三维模型进行预分割(over-segmentation)来得到一个个小块,然后利用特征描述符计算每个块的特征并以此作为此三维模型的低级特征。然后再基于深度学习算法并以无监督的方式从各个三维模型的低级特征中学到高级特征。最后,通过在高级特征空间中对各个三维模型的一个个块进行聚类,从而可以很快的得到分割或协同分割的结果。实验结果表明,与其它三维模型分割算法相比,本文的算法展现出更优异的分割效果。在三维模型分类算法中,该算法基于密度峰值聚类(Density Peaks Clustering)来有效提高分类的准确率。算法首先利用多种特征描述符分别对每个三维模型提取相应的特征向量,并将由此得到的特征向量运用鲁棒主成分分析(Robust Principle Component Analysis)去除噪声并降维,最后通过密度峰值聚类实现三维模型的分类。本文算法具有可以通过决策图以直观的方式确定三维模型分类类别数的优点,克服了传统算法在这方面的缺陷。实验结果表明,与传统算法相比,本文算法具有准确率高,鲁棒性强等优点。
[Abstract]:With the development of 3D scanning and modeling techniques, 3D models are widely used in the fields of game, virtual reality, video animation, biology and so on. The rapid increase in the number of 3D models poses a new challenge to digital geometry processing, among which the problem to be solved is how to sort and classify them based on their geometric shapes. In addition, in the field of geometric modeling and model processing, how to segment 3D models automatically becomes more and more important. It not only helps to understand graphics, but also plays an important role in many computer graphics problems, including mesh parameterization, model simplification, anomaly detection, bone extraction, resolution modeling and graphic retrieval. In this paper, the classification and segmentation of three-dimensional models are studied, and relevant algorithms are proposed. In the 3D model segmentation algorithm, the algorithm can automatically segment a single 3D model or co-partition the same kind of 3D model (co-segmentation). At first, each 3D model is prepartitioned (over-segmentation) to obtain a small block, and then the feature descriptor is used to calculate the features of each block as the low-level feature of the 3D model. Then, based on the depth learning algorithm and in an unsupervised manner, we can learn advanced features from the low-level features of each 3D model. Finally, by clustering the blocks of each 3D model in the high-level feature space, the results of segmentation or co-segmentation can be obtained quickly. Experimental results show that the proposed algorithm is more effective than other 3D model segmentation algorithms. In the 3D model classification algorithm, the algorithm based on the peak density clustering (Density Peaks Clustering) to effectively improve the accuracy of classification. The algorithm firstly extracts the corresponding feature vectors for each 3D model using a variety of feature descriptors, and uses the robust principal component analysis (Robust Principle Component Analysis) to remove noise and reduce the dimension. Finally, the classification of 3D models is realized by peak density clustering. This algorithm has the advantage that the classification number of 3D models can be determined intuitively by the decision graph, which overcomes the defects of the traditional algorithm in this respect. The experimental results show that the proposed algorithm has the advantages of high accuracy and robustness compared with the traditional algorithm.
【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】