A Research On Narrative-Based Automatic Scene Generation For Computer Animation

Scene design is one of the most important yet labor-intensive tasks in animation generation. Along with the development of computer graphics technology in the movie and animation generation, people could implement the scene design by computer. But, the routine work of scene design costs too much time. In this thesis, we present a case-based and rule-based method for automating scene design.The main contributions of the thesis are as follows:1. Definition of a formal language for describing a scene. Based on this formal language, we could describe succinctly a scene and this description could be identified easy by computer.2. Design of a multi-layered method for scene description and storage. In order to reuse and share the information of a scene case, the scene is divided into many parts: scene attribute, scene element attribute, relations of the scene elements, scene color schemes, scene environment lighting and so on.3. Design of a scene markup method for scene annotation, and a rule representation for representing scene design knowledge. Using these annotated resources and formal knowledge, the scene design system can automatically generate different coherent scenes.4. Introduction of a scene design algorithm and implementation of a scene design system. In addition to the basic requirements of a scene task, our system also considers some problems relevant to the characteristics of characters (also called the scene hosts), actions which could happen in the scene, matching of scene element sizes and characters. These are more relevant to narrative requirements.Experiments show that our method accords with the human scene design process and the results of experiments accord ultimately with the user's requirements. Furthermore, the plenty of semantic information which is included in scene cases and scene element cases could be reused in other virtual reality environments.