基于模板的陶瓷文物碎片分类与拼接方法研究

Aiming at the NP hard problem about many fragments of archaeological excavations and complex relationship of fragments assembly, a new method of intelligent classification is proposed. We will reduce the size of the fragments through the target guidance of template annotation to move the hard problem forward and break through the bottleneck of the application of the virtual restoration project of the damaged cultural relics. The research based on the template target guidance and involved in feature extraction, template construction, intelligent classification and assembly and restoration: ① Study on multi scale feature representation and multi feature understanding of ceramic cultural relics is aim at filling the gap between the physical characteristics of ceramic cultural relics and cultural semantic features. ② According to the existing and restoration of complete cultural relics, a fusion approach of model set consistent segmentation and image set cooperative segmentation is proposed. Building the template library of ceramic artifacts based on ontology description to provide a template specification for the restoration of broken artifacts. ③A new method of migration and classification is proposed. Based on the multistep connection refinement, semi-supervised manifold alignment method, comprehensive examples of constraints and characteristics of contact and other new ways, we construct the intelligent classifier model of cultural relic fragments, and use the template to achieve the accurate classification of large fragment sets. ④A assembly method of surface adjacency constraint based on template guidance and incomplete model repair method based on template matching analysis are proposed, aiming at improving the efficiency of virtual repair. It could form the theoretical method and system of the virtual restoration of the ceramic cultural relics, and enhance the social and cultural value of the damaged cultural relics.

针对考古发掘碎片多、拼接关系复杂的NP难问题，研究提出碎片智能分类新方法，通过模板的目标引导减小待拼接碎片集合规模，将拼接难度前移，突破破损文物虚拟复原工程应用的瓶颈。研究基于模板目标引导，涉及文物特征提取、模板构建、智能分类、拼接修复四方面：①研究陶瓷文物的多尺度特征表示、多特征理解，填补陶瓷文物物理特征与文化语义特征之间的鸿沟；②依据现存及复原的完整文物，提出模型集一致性分割和图像集协同分割的融合方法，构建本体描述的陶瓷文物模板库，提供破碎文物复原的模板规范。③提出迁移分类新方法，通过基于多步连接精化和半监督流形对齐方法、综合实例约束和特征联系等多途径，构建文物碎片智能分类器模型，应用模板指导实现对大碎片集子类准确划分。④提出模板导引的表面邻接约束碎片拼接方法和基于模板匹配分析的残缺模型修补方法，提高虚拟修复的效率。形成支撑陶瓷文物虚拟复原的理论方法与体系，提升破损文物的社会文化价值。

针对考古发掘碎片多、拼接关系复杂的NP难问题，本研究基于陶瓷文物模型，以模板构建与模板应用为核心，针对陶瓷文物模型拼接中的四大关键问题：特征提取、模板库构建、智能分类方法与拼接修复技术进行深入研究。研究成果主要包括：（1）研究了陶瓷文物三维模型预处理和特征提取技术，提出了基于离散Morse理论的散乱点云特征提取；改进和提出了适合文物三维模型检索的有效特征提取算法；（2）通过提取融合文物的物理与文化特征，构建兵马俑模板库和部位特征的标准模板库；（3）构建多步连接精化思想指导下的少样本量的高效分类器，提出了基于C均值聚类和图转导的半监督分类算法、基于多特征和SVM的兵马俑碎片分类以及基于深度学习和模板引导的三维兵马俑碎片分类；（4）深入研究了文物模型修复和拼接技术，提出了奇异值控制的图像特征点匹配方法、基于关键点特征描述子的三维文物碎片重组、结合表面纹理与断裂轮廓的碎片拼接方法、基于断裂面特征点匹配的文物碎片重组方法、基于表面纹理特征定义的碎片拼接方法、基于轮廓线特征点的交互式文物拼接等十余种模型匹配拼接修复算法。.项目达到并超额完成预定目标，举办参加9次国内外学术会议；发表论文共41篇（其中SCI 11篇，EI 12篇）；申请发明专利3项，授权1项；软著2项；获省奖1项；并培养3名博士生和12名硕士生。成果进一步形成支撑破碎陶瓷文物复原的理论基础与方法体系，成果方法已服务于兵马俑实体考古复原工作，提供有效指导，提升破损文物的社会文化价值，理论和应用意义显著。