Use of Shot / Scene Parsing in Generating and Browsing Video Databases

There is a large volume of video which exists in the world today, stored in a multitude of different forms. In order to make this information more easily accessible, we need methods of organizing this data in an intelligible manner. This thesis suggests a method of cataloging and browsing videos by extracting key frames from the individual shots and scenes in the film. The algorithms presented herein were tested on video containing a wide variety of images (interviews, sports, computer animation, etc.) They provide a way to automatically segment an audio/video stream into shots and scenes, which could then be used to generate a database for many applications, from logging a broadcasting station's archives to browsing video in a digital library. Thesis Supervisor: Michael Hawley Title: Assistant Professor of Computer Science and Media Technology