Research On Interactive Digital Geometric Modeling And Key Techniques

With the rapid advance and development of existing 3D scanning and related techniques,3D digital geometry models have nowadays become an emerging type of digital media after sound, images, and video, which have been intensively used in many fields such as industrial manufacturing, digital entertainment, biology, medicine and digital protection. And how to create the corresponding models required by the user is always one of the central problems in computer graphics and digital geometry processing. As the main digital geometric modeling method, interactive digital geometric modeling is a challenging task. In this thesis, according to the application requirement of digital geometric modeling, we focus on several problems, such as the convenience of interaction, the diversity of modeling and the flexibility of annotation, and research on several techniques, such as sketch-based 3D modeling, example-based 3D model synthesis, and discovery-based 3D model classification. Regarding these issues, this thesis engages in the following work:(1) We propose a rectangular polyhedron modeling method based on single perspective sketchy drawing and decision tree. This method use single perspective sketchy drawing to support the user’s interactive design of 3D rectangular polyhedron in order to reduce the user’s effort of the sketch drawing. Meanwhile, the incremental decision tree algorithm is used to train and update the rule of sketch analysis and 3D mapping automatically, which improve the definition efficiency and expansibility of the rule set. Accordingly, by the trained rules, the single sketchy drawing can be turned into the corresponding rectangular polyhedron, which improve the convenience of interaction of the constructive sketch-based 3D modeling.(2) We propose an example-based 3D model synthesis method based on boundary consistency analysis and Petri net. This method use the boundary consistency analysis method to discovery the potential adjacent relation in the example, and connect the corresponding construction unit to create the novel local structure, which improves the novelty of local topology. Meanwhile, Petri net is used to describe the construction constraints and processing of the 3D model in order to realize the mutual delivery of the parameters of scale extending and geometric deformation, which enriches the variation mode of the synthesis process. Accordingly, through the creation of the novel local structure and the combination of multiple variation mode, the modeling diversity of the example-based 3D model synthesis is improved.(3) We propose an iterative 3D model classification method based on online metric learning. This method use an iterative labeling method based on unsupervised clustering to realize the recurrent classification process of the model set, which removes the dependence of pre-labeled samples and pre-trained classifiers. Besides, each iteration includes two operations:member selection and group labeling, which leads to a personalized result in order to improve the adaptation of the classification for different applications. Moreover, the incremental updating strategy based on online metric learning is used to realize the incremental classification of the scalable model set, which improve the efficiency, and reduce the interactive effort of 3D model classification process.