英文摘要：
Video is the most influential media, but it’s difficult to nonlinearly search video content. The integrated creative work of this paper includes: Based on image processing to recognize video subtitle and convolutional neural networks to recognize faces of characters, the problem of film and TV video nonlinear retrieval is solved. Further, we extract important entities from subtitle text and enhance their relevant knowledge with large scale knowledge base and e-books, which constructs a cross-media application system of video, text, and e-book. Word cloud of subtitles and character entities are designed to facilitate video overview understanding and navigating retrieval. Crowdsourcing technology is used to update the amendments of subtitles, e-books, face recognition and entities information. A typical cross-media convergence system are completely implemented including movies in modern history, conference of the Chinese poetry, and information technology documentary video.
|