Automatic Digest Generation by Extracting Important Scenes from the Content of Presentations

Recently, combining a video recording of a presentation along with the digital slides used in it has become popular in e-learning and presentation of archives. For users of the archives, it is useful to preview a digest of such content to grasp the atmosphere and/or an outline of the presentation. This paper proposes a method of automatic digest generation by extracting important scenes from the presentation content. The extracted scenes are chosen based on several factors such as frequency and specificity of words, scene duration and order. Finally, the effectiveness of the proposed methods are evaluated by comparing with testers' answer sets for actual lectures.