Study on intelligent video retrieval system based on semantic representation of video data

多媒体信息系统中的一个重要问题是缺乏多媒体对象的语义表示模型，无法从大型档案中搜索和检索适当的多媒体对象。在本研究中，我们对视频数据进行了多方面的研究。在我们的系统中，典型场景由连接的图结构注释，其中图节点对应于特定的影响单元。情感单位是情感因果关系的抽象。我们的视频描述模型是基于面向对象编程领域中提出的“带原型的面向对象模型”思想来实现的。原型是一个预制对象，具有预定义的结构、内容和行为，可以通过复制原型来创建新对象。通过使用原型注释，我们的系统使用户能够创建视频内容的语义表示，对相关的场景描述符进行分层分组，使用这些描述符对视频内容进行注释，并重用描述工作。通过使用我们对视频内容的语义表示，数据库可以为用户提供三种类型的检索工具：基于关键字的、基于故事的和两者混合的。

An important issue in multimedia information systems is the lack of models for semantic representation of multimedia objects to enable the search and retrieval of an appropriate multimedia object from large archives. In this research, we have studied on various aspects especially concerning video data. In our system, the typical scene is annotated by connected graph structures where graph nodes correspond to specific Affect Units. An Affect unit is an abstraction of affective causality. Our video description model is implemented based on the ideas of "object- oriented model with prototypes" which is proposed in the field of object-oriented programming. Prototype is a prefabricated object with predefined structure, contents, and behavior, from which new objects can be created by copying the prototype. By using the prototype annotation, our system enables users to create semantic representations of video content, group related sets of scene descriptors hierarchically, use these descriptors to annotate video content, and reuse descriptive effort. By using our semantic representation of video content, the database can provide the user three types of retrieval tools : keyword-based, story-based, and both-mixtured.