基于机器学习的复杂曲面特征提取与分割方法及其应用研究

发布时间：2018-06-18 17:26

本文选题：模型分割 + 特征提取　；参考：《浙江大学》2017年硕士论文

【摘要】：复杂曲面三维模型的分割是三维模型处理中的关键步骤,三维模型的分割在三维模型检索、几何压缩传输、三维模型简化等领域有着至关重要的作用。基于曲率信息的三维模型分割方法是三维模型分割方法中一种重要的分割方法。本文在对三维模型特征提取和分割相关技术的研究现状进行总结和分析的基础上,针对基于曲率信息的三维模型分割方法存在效率低、准确性无法保证的问题等进行了深入研究,根据研究成果开发了复杂曲面三维模型分割系统,在牙齿三维模型的分割中进行了应用验证。论文的主要研究内容包括:第一章介绍了基于机器学习的三维模型特征提取和分割技术的国内外相关研究,包括三维模型特征提取技术、基于机器学习估计阈值技术和三维模型分割技术。分析了上述技术的研究进展以及目前存在的问题,提出了本文的研究内容并分析其研究意义,最后简单介绍本文的行文结构。第二章提出了复杂曲面三维模型顶点高斯曲率密度直方图的特征描述方式。此特征描述对于三维模型高斯曲率阈值的获取有较大的辅助作用。基于复杂曲面顶点高斯曲率密度直方图的特征描述,提出将三维模型众多特征融合的方案设计。同时利用相关工具设计了实验,提取了三维模型顶点的高斯曲率密度直方图,验证了复杂曲面三维模型顶点高斯曲率密度特征提取方法的正确性。第三章提出了基于机器学习的复杂曲面高斯曲率阈值估计方法。依据复杂曲面模型和模型特征对应的高斯曲率阈值的可靠样本,训练出复杂曲面特征和对应高斯曲率阈值的回归方程,通过复杂曲面模型测试样本对回归方程的参数改进、修正,随着样本量的增加,回归方程的可靠性也得到提升,达到了指定复杂曲面模型对应的高斯曲率阈值估计的准确度,从而可以根据复杂模型特征求解出复杂曲面模型分割所需的高斯曲率阈值,解决了基于曲率的复杂曲面分割过程中,高斯曲率阈值难以计算的问题。第四章提出了基于高斯曲率阈值估计的三维模型分割方法。利用第三章中提出的高斯曲率阈值估计方法,建立了复杂曲面模型特征描述和对应的高斯曲率阈值之间的回归模型,进行待分割模型的高斯曲率阈值估计,实现了三维模型的准确分割,减少了人工交互设置高斯曲率阈值的繁琐步骤,提高了复杂曲面模型分割的效率,通过对实际三维模型进行分割,对基于高斯曲率阈值估计的三维模型分割算法进行了实验设计并验证了提出方法的有效性。第五章开发了复杂曲面三维模型分割系统,实现了基于机器学习的复杂曲面三维模型高斯曲率阈值估计模块、复杂曲面三维模型边界线划分模块和复杂曲面三维模型分割模块。并将此系统应用于牙齿三维模型分割。第六章对本文研究存在的不足进行了总结,并对今后的研究工作进行了展望。
[Abstract]:The segmentation of 3D model of complex surface is a key step in 3D model processing. The segmentation of 3D model plays an important role in the fields of 3D model retrieval, geometric compression and transmission, and simplification of 3D model. 3D model segmentation based on curvature information is an important method in 3D model segmentation. On the basis of summarizing and analyzing the research status of 3D model feature extraction and segmentation, this paper aims at the low efficiency of 3D model segmentation based on curvature information. Based on the research results, the 3D model segmentation system of complex curved surface is developed, which is applied to the tooth 3D model segmentation. The main contents of this paper are as follows: in chapter 1, we introduce the research of 3D model feature extraction and segmentation based on machine learning, including 3D model feature extraction. Based on machine learning estimation threshold technique and 3D model segmentation technology. This paper analyzes the research progress and existing problems of the above technologies, puts forward the research contents of this paper and analyzes its significance, and finally briefly introduces the structure of this paper. In chapter 2, the feature description method of the vertex Gao Si curvature density histogram of 3D model of complex surface is presented. This feature description is helpful to obtain the Gao Si curvature threshold of 3D model. Based on the feature description of the Gao Si curvature density histogram at the vertices of complex surfaces, the scheme of fusion of many features of 3D models is proposed. At the same time, experiments are designed by using related tools to extract the Gao Si curvature density histogram of the vertices of the 3D model, which verifies the correctness of the method for extracting the Gao Si curvature density features of the 3D model vertices of complex surfaces. In chapter 3, a method for estimating Gao Si curvature threshold of complex surfaces based on machine learning is proposed. According to the reliable samples of the Gao Si curvature threshold corresponding to the complex surface model and the model characteristics, the regression equations of the complex surface features and the corresponding Gao Si curvature thresholds are trained, and the parameters of the regression equations are improved by the complex surface model test samples. With the increase of sample size, the reliability of the regression equation is improved, and the accuracy of Gao Si curvature threshold estimation corresponding to the specified complex surface model is achieved. Thus, the Gao Si curvature threshold needed for the segmentation of complex surface model can be solved according to the characteristics of complex model, and the problem of Gao Si curvature threshold is difficult to calculate in the process of complex surface segmentation based on curvature. In chapter 4, the segmentation method of 3D model based on Gao Si curvature threshold estimation is proposed. Using the Gao Si curvature threshold estimation method proposed in Chapter 3, the regression model between the characteristic description of complex surface model and the corresponding Gao Si curvature threshold is established, and the Gao Si curvature threshold is estimated for the model to be segmented. It realizes the accurate segmentation of 3D model, reduces the tedious steps of manual interaction setting Gao Si curvature threshold, and improves the efficiency of segmentation of complex surface model. The experimental design of 3D model segmentation algorithm based on Gao Si curvature threshold estimation is carried out and the validity of the proposed method is verified. In the fifth chapter, a 3D model segmentation system for complex surfaces is developed, which realizes the Gao Si curvature threshold estimation module based on machine learning, the boundary line partition module of complex surface 3D model and the 3D model segmentation module of complex surface. The system is applied to the tooth 3D model segmentation. The sixth chapter summarizes the shortcomings of this study and prospects the future research work.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP181;O186.11

【相似文献】