Browse Prior Art Database

Automatic Selection of Representative Keyframes from Video Sequences Using Scene Content

IP.com Disclosure Number: IPCOM000123529D
Original Publication Date: 1998-Dec-01
Included in the Prior Art Database: 2005-Apr-05
Document File: 2 page(s) / 88K

Publishing Venue

IBM

Related People

Diklic, D: AUTHOR [+3]

Abstract

Background: A proliferation of digital video as a means of communicating ideas, products, improving education, enabling content production etc. The latest example is a video of President Clinton being widely distributed over the World Wide Web. Video used electronically to promote and advertise products is emerging as a very important future application of media on the Internet. Effective manipulation, authoring, search and browse of video in a networked environment like the Internet poses some problems: video files are quite large and users do not want to wait for long periods for files to download or to watch the whole video in order to get to the scene of interest.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Automatic Selection of Representative Keyframes from Video Sequences
Using Scene Content

   Background:

   A proliferation of digital video as a means of
communicating ideas, products, improving education, enabling content
production etc.  The latest example is a video of President Clinton
being widely distributed over the World Wide Web.  Video used
electronically to promote and advertise products is emerging as a
very important future application of media on the Internet.
Effective manipulation, authoring, search and browse of video in a
networked environment like the Internet poses some problems: video
files are quite large and users do not want to wait for long periods
for files to download or to watch the whole video in order to get to
the scene of interest.

   Traditional means for solving these problems are based on
breaking video into scenes, describing each scene with one or more
keyframes, and allowing users to click on the desired keyframe to
access (i.e. play back) that particular portion of the video.  The
key problem here is the cost of manual labor involved in breaking the
video into scenes and manually extracting the most representative
keyframes.  For many applications this cost is prohibitive.

   Problem this disclosure solves:

   While automatic detection of scene changes, i.e. the
process of breaking video into scenes, has been solved (Ref. 1); the
problem of quality automatic selection of one or more representative
keyframes from video or video scene has not been addressed.  This is
the main focus of this disclosure.  Currently, keyframes are selected
from scenes either manually or at fixed positions (i.e. every 5th,
first, middle, last etc.)  This approach suffers from the fact that
in selecting keyframes scene content is not taken into account, which
can cause some valuable keyframes to be missed.  This disclosure
provides for automatic selection of keyframes using scene content
i.e., content of video frames.

   Solution:

   Assume there is a video sequence, preferably a video scene
representing relatively similar visual content.  Such video scenes
can be automatically extracted using known scene cut algorithms
(Ref. 1).  Let each frame of a video scene at position i...