Templates Based 3D Terracotta Warrior Fragments Part Annotation

Archeological excavation of the Terra Cotta Warriors acquired actual measured 3D data, which has high dimension and complicated adjacency relationship. Automatic mosaics of a number of high-dimensional complex fragments belong to the NP-hard problem. This article focus on objective model's part templates obtain, applying the part templates to label the position of a number of 3D fragments, make the part templates as the goal orientation of intelligent stitching, classify and label a number of complex adjacent fragments by position, thereby reduce the scope of test mosaic fragments, improve the mosaic efficiency and reduce the mosaic difficulty. The main research works and innovations are as follows:(1) Single model part templates acquisition:To solve the traditional model segmentation algorithm problems that curvature calculation is carried out in the whole model has high time complexity, and not suitable for high dimensional data, a 3D model segmentation algorithm based on regional curvature constraint is proposed. The algorithm reduces the discrete curvature calculation scope, avoid over-segmentation, and improve the single model segmentation algorithm efficiency in large-scale data. Applied to single Terra Cotta Warriors model with obvious main-branch structure, "meaningful" segmentation parts can be obtained, form the accurate part templates.(2) The model set part template acquisition: For complex model, it is difficult to produce a consistent result by single model segmentation algorithm. A consistent model set segmentation algorithm based on multi-feature spectral clustering is proposed. The algorithm utilize multi feature descriptors to generate multiple similarity matrices, calculate the weight of different similarity matrix automatically, make use of multi-feature advantages to realize unsupervised segmentation, get the part template of the entire model set, the problem of accuracy of the consistent segmentation algorithm is overcomed.(3) Fragment part labeling based on surface feature similarity retrieval:Fast labeling algorithm for similar fragments based on Exact Euclidean Locality-Sensitive Hashing (E~2LSH) is proposed. Firstly extract the surface features of fragments, and hash index the surface features based on E~2LSH. Secondly for the part template the random sample at the same scale of the fragment is conducted, the surface features of the samples are extracted as the labeled samples. Finally, retrieve the fragments that are similar to the labeled sample based on E~2LSH, and label the fragments with part tag. The retrieval algorithm based on E~2LSH reduces the time complexity to sub-linear, comparison with k-means retrieval method proves the efficiency and accuracy, the algorithm has a good practicability for large-scale high-dimensional data.(4) The semantic web construction of Terra Cotta Warriors 3D digital models:Proposed method of the semantic web construction of Terra Cotta Warriors digital 3D models based on ontology. Building the domain ontology of Terra Cotta Warriors, and extend the ontology based on WordNet, utilize semantic web to retrieve similar fragments or part template. Realize the effective management of domain knowledge and mosaics relationship.