|
Efficient
procedures for browsing, filtering, sorting or retrieving pictorial
content require accurate content characterization. Methods for
automatically classifying a video streams into categories are
presented.
In
the first method motion information is extracted from a video
sequence and used to train Hidden Markov Models to classify sports
and news video sequences and different types of movie trailers.
The
second method uses local activity estimation and shot length values
to classify movie trailers according to their activity/violence
characteristics with a simple 2-D classifier.
Results
and directions for future research are discussed.
N.
Vasconcelos and A. Lippman. Towards Semantically Meaningful Feature
Spaces for the Characterization of Video Content, ICIP'97, Santa
Barbara, California, 1997, IEEE. ftp://ftp.media.mit.edu/pub/nuno/papers/chips.ps.gz
G.
Iyengar and A. Lippman. Models for automatic classification of
video sequences. SPIE vol. 3312, pp. 216-227, Photonics West '98,
San Jose, California 1998.
The
other papers you were interested in are:
Joshua
S. Wachman (1996) TR#383: A Video Browser That Learns by Example
Appears in: Masters Thesis, MIT Media Lab ftp://whitechapel.media.mit.edu/pub/tech-reports/TR-383.ps.Z
N.
Vasconcelos and A. Lippman. A Bayesian Video Modeling Framework
for Shot Segmentation and Content Characterization, Wkshp on CAIVL,
CVPR'97, San Juan, Puerto Rico, 1997, © IEEE. ftp://ftp.media.mit.edu/pub/nuno/papers/CAIVL97.ps.gz
|
