动漫领域语义检索及场景生成关键技术研究

发布时间：2018-04-29 12:35

本文选题：动画生成 + 语义检索　；参考：《湖南大学》2012年硕士论文

【摘要】：动画生成技术是结合了人工智能理论与现代多媒体技术的全新动画制作过程。系统接收以自然语言描述的剧本,使用人工智能、语义检索、空间推理与布局等相关技术对故事剧本进行分析、理解,提取场景、事件、人物等信息,并在动漫领域知识库的协助下,采用语义标注技术将多媒体信息(动漫资源)语义化,并采用语义检索引擎对相关剧本进行语义检索查询和推理,最终实现动画的生成。本文主要对动漫领域语义检索及场景生成关键技术进行了研究,在论文的前半部分我们主要对动漫领域语义检索的关键技术进行了研究,通过分析语义检索中的一些关键技术,提出了基于本体的查询扩展机制和加权排序优化算法,并结合本体的具体特点,提出了面向查询的动漫语义检索框架,借助LarKC开发平台,采用LarKC插件集中的SPARQL识别器初步实现了一种面向查询的动漫语义检索系统。本系统采用OWL作为本体的描述语言,基于动漫制作领域建立相关本体,实现了动漫领域知识的描述,建立了动漫本体知识库和规则库；并通过语义标注来进行动漫素材的语义化,并借助Jena开发工具,实现了针对动漫领域知识的本体推理和查询,并对系统的查询效率、查全率与查准率做了一系列的实测实验。实验结果表明：实验得到的动漫素材能够满足检索需求,同时也验证了动漫素材领域本体,确实保证了系统的查准率与查全率,验证了检索框架的实用性与有效性。论文的后半部分我们主要对场景生成的相关方法进行了研究,论文从定性和定量两方面将场景生成进行了简要的讨论,分析了场景生成的一些关键技术和方法,提出了场景生成算法和针对物体间关系用来计算物体位置和方向的场景物体空间关系布局推理方法等。最后,我们提出了场景生成系统的框架和设计思想,对于某一段故事剧本,动漫本体描述语言能将其以RDF/N3形式进行描述,也就是说,我们能以RDF三元组的形式来描述某一段剧本,或者说可以把某一段剧本以RDF文件的形式来表达。然后通过SPARQL语言查询匹配动漫本体描述语言的事件(Action),再根据Action各故事划分成场景,然后根据每个场景的背景描述查找场景库,生成该背景下的典型场景,之后对动漫本体描述语言中空间信息进行分析,然后加上物体规则库动态修改场景信息,最后添加该场景中出现的人物信息,得到最终的场景。
[Abstract]:Animation generation technology is a new animation production process which combines artificial intelligence theory and modern multimedia technology. The system receives scripts described in natural language, uses artificial intelligence, semantic retrieval, spatial reasoning and layout techniques to analyze, understand, extract scenes, events, characters, etc. With the help of animation domain knowledge base, semantic annotation technology is used to make multimedia information (animation resource) semantic, and semantic retrieval engine is used to query and infer the relevant scripts. Finally, animation generation is realized. In the first half of this paper, we mainly study the key technologies of semantic retrieval in animation field, and analyze some key technologies in semantic retrieval. A query extension mechanism based on ontology and a weighted sorting optimization algorithm are proposed. Combined with the specific characteristics of ontology, a query oriented animation semantic retrieval framework is proposed, which is based on LarKC development platform. A query oriented animation semantic retrieval system is implemented using SPARQL recognizer in the LarKC plug-in set. This system uses OWL as the description language of ontology, establishes the related ontology based on the animation production domain, realizes the description of the knowledge in the animation field, and establishes the knowledge base and rule base of the animation ontology. With the help of Jena, the ontology reasoning and query of animation domain knowledge are realized, and a series of experiments are made on the query efficiency, recall and precision of the system. The experimental results show that the animation materials obtained from the experiment can meet the retrieval requirements, and also verify the ontology of animation material domain, which ensure the precision and recall of the system, and verify the practicability and effectiveness of the retrieval framework. In the second half of the thesis, we mainly study the related methods of scene generation, discuss the scene generation from qualitative and quantitative aspects, and analyze some key techniques and methods of scene generation. A scene generation algorithm and a reasoning method for spatial relation layout of scene objects are proposed, which are used to calculate the position and direction of objects. Finally, we put forward the framework and design idea of scene generation system. For a story script, animation ontology description language can describe it in the form of RDF/N3, that is, We can describe a script in the form of RDF triples, or we can express a script as a RDF file. Then the events matching animation ontology description language are queried by SPARQL language, and then divided into scenes according to each Action story. Then, the scene library is searched according to the background description of each scene, and the typical scene under this background is generated. Then it analyzes the spatial information in the animation ontology description language, then dynamically modifies the scene information by adding the object rule base, and finally adds the character information in the scene to get the final scene.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.3

【参考文献】