Research On Feature Mapping Mechanisms For Sketch-Based Modeling

Sketch-based modeling is one of the most challenging tasks in computer graphics. The principal mission is to construct 3D models corresponding with users’ sketch according to human perceptual system and object knowledgement. As a new research subject in computer graphics, sketch-based modeling is built on the achievements of many existing research subjects, and it has wide range of applications including filmmaking, avatar design in games and cartoons, virtual reality, distance education and so forth.In this thesis, we studied the problem of feature mapping mechanisms in sketch-based modeling from single user’s drawing. We research on several techniques of high computational complexity in field-based modeling, data gathering problem of procedurally described models, scalability of procedurally modeling with constructive template, corresponding problem of strokes and model feature lines in deformation methods. Regarding these issues, this paper engages in the following work:(1) We proposed a sketch-based garment modeling method in which a simplified distance field is used. In this method, a distance field based on distances along z-axis is defined according to garment characteristic, so that the 3D space sampling problem in distance calculation is changed into 2D space sampling procedure. Then a distance propagation mechanism based on iterative diffusion is adapted to convert 2D silhouette features into 3D distance features well-distributed. In field interpolation, a stretch strategy of garment grids and a distance degeneration mechanism based on relative position are used to cover the simplified distance field area with 2D garment grids, and finally the smooth and well-constructed garment mesh is obtained. The proposed method significantly reduces the computational complexity of field calculation in sketch-based modeling with field structure, and it improves the interactive performance of modeling method while the models constructed are reliable.(2) We proposed a sketch-based body modeling method with multilayer cross-section templates. In this method, a body template structure with two layers of skeleton and skin is defined based on human articulation structures, then an interactive labeling and segmentation procedure is adapted to extracting body template from 3D body triangular mesh. Furthermore, we use the distance descriptor based on anatomy bodies and a sketch-based shape retrieval method to find the body template consistent with user’s drawing. The proposed method improves the scalability of modeling method by expanding template data sources to general triangular meshes. Since the intent-driven templates construction method is implemented by using the sketch-based retrieval, the modeling applicability with different sketches is also strengthened.(3) We proposed a matching method between strokes and model feature lines based on conditional random fields. In this method, a multi-feature conditional random field model is established between strokes and feature lines in matching area, and the relations between strokes and feature lines are calculated by combining shape similarity features and gradient features of depth map together. Furthermore, a feature line rational metric with topology and shape features together is defined, and then an iterative procedure is adapted to migrate matching area and approach best feature lines. Finally the global optimized model feature lines are extracted. In proposed method, the users’drawing mind and model geometric features can be reflected in the matching results between strokes and feature lines. The problem that complex sketch can not be introduced into sketch-based deformation method is preliminary solved, and the field of application of sketch-based deformation method is expanded too.