Research on 'Database Vision' based on media cooperative recognition and purposive utilization

在本研究项目中，成功地提出了一种实现高效多媒体数据库系统的新方法--数据库视觉框架，其中基于媒体融合的自动图像识别和对已有数据模型的有目的和自适应利用两个新的思想发挥了重要作用。取得了以下具体成果：(1)提出了数据库视觉方法的框架和有目的的应用：在这个子项目中，开发并实现了一种基于状态转换机制的通用自动图像理解方法，提供了适用于各种目的的灵活的图像数据模型。(2)开发了一个基于媒体融合的戏剧电影自动理解系统：实现了一个具体的运动图像理解系统，该系统使用图像、文档和语音的协同识别，成功地实现了对真实电视剧电影的困难的场景分解和人类识别。(3)目的图像数据库的实现：两个全自动图像数据库，一种是针对体育场景静止图像的图像序列分析方法，另一种是针对电视场景运动图像的序列图像分析方法。(4)提出了一种新型的基于颜色复用的图像序列分析方法，并通过实验验证了其有效性。

In this research project, a new approach to realize effective multi-media database systems, named 'Database Vision' framework, has been successfully proposed, where two new ideas of automatic image recognition based on media fusion and purposive and adaptive utilization of achieved data model, playa important role. The following concrete results have been obtained.(1) Proposal of the framework for database vision approach and purposive utilization: In this sub-project, a general purpose automatic image understanding method using 'state-transition' mechanism has been developed and implemented, which provides flexible image date models adaptively applicable to various purposes.(2) Development of an automatic drama movie understanding system based on media fusion: A concrete moving image understanding system using cooperative recognition of images, documents and voice, has been implemented to successfully realize difficult 'scene decomposition' and human identification for real TV drama movies.(3) Realization of purposive image databases: Two full-automatic image databases, one for sports scene still images and the other for TV scene moving images, have been realized and successfully evaluated by experiments.(4) A new-type of Image Sequence Analysis Method using color multiplextion has been invented and effectiveness has been revealed by experiments.