NAIST Digital Video Library: Tools for Restructuring Video Data for the Content Based Information Retrieval - A Representative Image of Shot Concept for the Internet

Abstract

Nara Institute of Science and Technology started services of a digital library system for campus use since April 1996. This system includes the function to access the digital video data. However, this function just provide showing video data on the terminals. In order to make the digital video library more useful to browse and retrieve, we proposed a new structure of the digital video data.

There are a lot of elements relating to the video media, such as representative frames of each topics, the sound information which is accompanied with such the representative frame, the text information which explains about the contents, and so on. A new structure extracts some elements from the digital video data before storing the video data in the database(Fig.1(left)). This structure provides a useful function in order to retrieve, browse and edit the target data in the digital library systems for the Internet(Fig.1(right)).

We use following two terms: “shot” and “RS-frame” (Representative frame of Shot). Shot is a short sequence of the stream. A long video stream is composed of a large number of topics. For example, a news program about 30 minutes length includes a number of different topics of few minutes. Therefore, a motion video stream can be divided into a number of shots of about only a few minutes stream depending on the contents. Users can get the target scene directly by searching with shots without playing the video long stream consecutively. RS-frame is a representative frame of shot, which is selected from the shot properly.