ICME Fabchannel Demo Poster

1
Spatiotemporal Block Based Analysis of Video for
Fast and Effective Shot Transition Detection and
Identification
Umut Naci and Alan Hanjalic
Detecting Dissolves
The features
Dissolves are characterized by smooth and
monotonic changes in the block luminance values
in time.
The method takes the spatiotemporal blocks in
video as the basic units in stead of 2D frames.
Detecting Cuts and Wipes
F1 is the ratio of the total luminance change to
the total absolute luminance change in the cube
and is an indication of the monotonic change. F2
value is high if maximum luminance change in time
is small compared to the total luminance change.
Dissolve probability in time interval k is
calculated as the average of F1F2 values over
all blocks with time index k.
Cut and wipe detections are based on the maximum
time derivative indices for every
spatial location (m,n) of each spatiotemporal
block (i,j,k)
If there is an abrupt change, maximum time
derivative points are expected to form a plane
perpendicular to the time direction, yielding
high planarity values.
Test Results
Future Work
In the above graph, each column is the vector of
planarity values indicating the cut probabilities
in the blocks with the same time index. Cuts are
characterized by the columns with high planarity
values. In case of a wipe, blocks experiencing
abrupt change are spread in a time interval and
create a pattern specific to the type of wipe.

• More compatibility with TRECVID rules
• Analysis in motion direction (not in time
  direction)
• Better thresholding scheme
• Developing a unified low level analysis
  framework which involves shot change detector,
  camera motion detector and a general purpose
  motion estimation system

MultimediaN