Automatic Scene Generating Based On Fuzzy Semantic

Automatically generating computer animation is a research applying technology with many different areas of knowledge,including natural language processing,artificial intelligence,multimedia technology,digital image processing,virtual reality,even involved in the kinematics,visual physiology,visual psychology and art field.Changing the scene described by natural language into a 3D animation scene is an important part of automatically generating computer animation.This paper presents an automatic scene generator of static scene based on semantics.The generator proposed in the paper aims at automatically laying static objects(including models of human or animals without skeletons and motion information)in scene in accordance with the information of the object's name and the noun of locality in fuzzy natural language.But except that can't deal with the characters,such as human and animals with skeletons and motion information.And also,the generator can enrich and perfect the content of the scene automatically,making the static scene more natural and reasonable.In the research on natural language processing,there has some methods to extract the time,events(including subject,object and the contents of the event)and the spatial relationship of objects,therefore,this part is not the work of this paper.The work of this paper can be concluded into two parts as follows:Firstly,this paper presents a method to recover the spatial relations by adding other implicit information in the description of the scene in the text.First,we use a method based on analysising huge amounts of the corpus to add defected spatial relations.Then we need to calculate the spatial relationship between all the objects which should be put into the scene.Then according to these spatial relationships,we can get a more abundant and lively scene.Secondly,this paper presents a method of quantizing the distance between objects and changing fuzzy spatial relations into specific coordinate.This method is based on supervaluation to describe or define the context-dependent 'near' and 'far'relation.Then it calculates the crisp number of distance from a threshold gained by observations.After acquiring all the possible areas of one object by all the spatial relationships that related to the object,we can ensure the overlap part is the final putting area.Based on the methods mentioned above,this paper designs a scene generating system.After extracting the spatial relations from natural language input and recovering them,the system calculates the positions of all objects and generates a 3D scene.Some experiments shows that methods are effective.