Automated Shot Boundary Detection - Computer Science

Automated Shot
Boundary Detection
in VIRS
DJ Park
Computer Science Department
The University of Iowa
Introduction





VIRS?
Visual Information Retrieval System
Works on multimedia information domain such
as video, sound, etc
Aims for content-based analysis, retrieval and
presentation
Find all videos that Iowa football team plays
Building VIRS






Shot Boundary Detection
Scene Boundary Detection
Key Frame Generation
Pattern Recognition
Indexing with Meta Data
Interface and Querying Method
Shot Boundary Detection






A sequence of frames captured by a single
camera in a single continuous action
Most basic unit of video data
A scene is a logical group of shots into a
semantic unit
Many different styles and techniques ( cut, fade,
wipe, dissolve )
How to detect among these styles?
What about camera/object movements?
General Approach






Frame to Frame Comparison
Take 2 subsequent frames and measure their
differences
How to define these differences discriminates
each method
Pixel Difference
Color Histogram Difference
Edge Difference
Pixel Difference




One of the earliest methods
How many pixels are different between
subsequent frames?
Problems?
Improvement – using blocks instead of pixels
(X,Y)
Frame t
(X,Y)
Frame t+1
Color Histogram Difference



Much more accurate than simple pixel-wise
approach
Compare the number of occurrences for each
color between subsequent frames
Problems?
Frame t
Histogram t
Edge Difference




No color information is
needed
Compare entering edge
pixels and exiting edge
pixels
Better than other
methods in detecting
fade, wipe, dissolve
Problems?
Future Works




Evaluation and Combination of existing
methods
Relating shot boundary information with scene
boundary – how to group shots into semantic
unit?
Using sound information for shot ( or scene )
boundary detection?
Using object-oriented approach or 3-D image
modeling technique?
Questions?