Reducing the cost of metadata generation by using video/audio indexing and natural language processing techniques

Reducing the cost of generating metadata will allow more broadcast contents to be transmitted with advanced viewing options. In this article, we describe SceneCabinet, a system that automatically extracts scene-based semantic metadata from video content. It extracts meaningful video slices and their associated textual information such as the title, synopsis, and keywords by using natural language processing based on the results of speech and on-screen text recognition. Moreover, it can import video program scripts and use them for automatic keyword extraction. SceneCabinet provides an intuitive user operation interface including a video browser with key images detected automatically based on scene changes, on-screen text, camerawork, speech, and music information. Experiments showed that SceneCabinet can significantly reduce metadata generation costs.